Grounding Meaning in Everyday Experience in the World. An Embodied Construction Grammar Analysis of Italian Caused-Motion Constructions more

Università degli Studi di Pavia Facoltà di Lettere e Filosofia Corso di Laurea Specialistica in Linguistica Teorica ed Applicata Tesi di Laurea Grounding Meaning in Everyday Experience in the World. An Embodied Construction Grammar Analysis of Italian CausedMotion Constructions Relatore: Chiar.ma Prof.ssa Sonia Cristofaro Correlatore: Chiar.mo Prof. Cristiano Broccias Candidato: Enrico Torre Anno Accademico 2009-2010 To my father Nardi and my mother Bruna. ...linguistic symbols are social conventions for inducing others to construe, or take a perspective on, some experiential situation. (Michael Tomasello) ...our structured experience is an organism-environment interaction in which both poles are altered and transformed through an ongoing historical process. (Mark Johnson) Contents Acknowledgments List of Figures and Tables List of Abbreviations 1 Introduction 2 From cognition to grammar 2.1 Embodiment, cognition, and language 2.1.1 Embodied Cognition 2.1.2 Cognitive Linguistics 2.2 Insights from the NTL paradigm 2.2.1 The NTL paradigm in a nutshell 2.2.2 Simulation semantics 2.2.3 The computational dimension 2.3 Embodied Construction Grammar: setting the scene 2.3.1 Cognitive approaches to grammar 2.3.2 The ECG model 3 An ECG analysis of Italian caused-motion constructions 3.1 Introduction: data and methodology 3.2 Analysis 3.2.1 Active caused-motion constructions 3.2.2 Passive caused-motion constructions 3.2.3 Two examples of possible extensions 3.3 Discussion 3.3.1 Compositionality 3.3.2 The intersection between compositionality and polysemy 3.3.3 Some more issues related to our analysis 3.4 Summary and remarks 4 Conclusion: ECG and beyond References i iii vii 1 11 11 12 31 49 49 52 62 73 73 98 125 125 138 139 170 202 210 211 219 230 237 245 261 Acknowledgments Acknowledgments Since thesis work does not come out of nothing, there are several people I must (and am pleased to) acknowledge. First of all, I want to sincerely thank Prof. Sonia Cristofaro (Dept. of Linguistics, Università degli Studi di Pavia), for accepting to be my supervisor and for letting me develop my own work, without nonetheless failing to provide me with her constructive criticism, valuable advice, and also with several stimuli for reflection. I am also truly grateful to Prof. Cristiano Broccias (DISCLIC, Università degli Studi di Genova) for willingly accepting to be my co-supervisor. Furthermore, I am and shall always be deeply indebted to him, for making me fall for linguistics and, especially, for introducing me to the amazing world of Cognitive Linguistics. I must also thank Prof. Jerome A. Feldman (Dept. of Electrical Engineering and Computer Science, University of California at Berkeley), Dr Ellen K. Dodge (Dept. of Linguistics, University of California at Berkeley), and Mr Luca Gilardi (International Computer Science Institute) for kindly answering my e-mails and sending me some useful papers. Furthermore, I also want to thank Prof. Francisco Gonzálvez García (Dept. of English and German Philology, Universidad de Almería) who provided me with an insightful assessment of my thesis project when it was still at its very, very beginning. Moreover, I am grateful to Dr Ad Foolen (Dept. of Linguistics, Radbaud Nijmegen Universiteit) and Dr Emanuele Bardone (Dept. of Philosophy, Università degli Studi di Pavia) for reading and commenting on an earlier essay paper of mine on which a part of this thesis is based. Finally, I need to heartily thank the people who are closest to me: my best friend Giorgio Caviglia (INDACO, Politecnico di Milano), with whom I often discuss topics related to epistemology, cognition, and communication when we meet in our hometown Arenzano, for his fifteen years of true friendship, his help with computer issues, our Skype chats, and a billion more reasons I just cannot list here; my father Bernardo Torre, my mother Bruna Magliano, and my sister Marina Torre, for always providing me with all of their love, help and support; my girlfriend Sara Piperissa, for loving me, standing by me, and being my sunshine since November 2003. Probably, the present M.A. thesis might have been much better, but without you it would have surely been substantially worse. Thank you so much! i Figures and Tables Figures and Tables Figures 2.1. 2.2. 2.3. 2.4. 2.5. 2.6. 2.7. 2.8. 2.9. 2.10. 2.11. 2.12. 2.13. 2.14. 2.15. 2.16. 2.17. 2.18. 2.19. 2.20. 2.21. 2.22. 2.23. 2.24. 2.25. 2.26. 2.27. 2.28. 2.29. 2.30. 2.31. 3.1. 3.2. 3.3. 3.4. 3.5. 3.6. 3.7. 3.8. 3.9. 3.10. 3.11. 3.12. 3.13. 3.14. 3.15. 3.16. 3.17. 3.18. 3.19. 3.20. The Process schema The ComplexProcess schema The TagliareSimplePast3sg construction The SimplePastTense construction The Finite construction The HasVerbFeatures construction The VerbFeatureSet schema The Singular construction The 3rdPerson construction The WordForm schema The CauseEffectAction schema The ForceApplication schema The MotorControl schema The ForceTransfer schema The EventDescriptor schema The VerbPlusArguments construction The TransitiveCEA construction The Declarative construction The RD schema The Sara construction The NominalFeatureSet schema The FemaleSingularNoun construction The Pagnotta construction The SingularNoun construction The Female construction The Noun construction The Nominal construction The HasNominalFeatures construction The FemSgDefiniteArticle construction The La construction The DeterminerPlusNoun construction The Lanciare1SimpleFuture3Pl construction The Lanciare1 construction The Lanciare construction The CauseMotionAction schema The MotionAlongAPath schema The Motion schema The TL schema The SPG schema The ActiveCauseMotion1 construction The ActiveCauseMotion construction The Bambini1 construction The I construction The Sassi construction The MalePluralNoun construction The Bersaglio1 construction The Preposition construction The SpatialPreposition construction The PathPreposition construction The A construction The A2 construction 108 109 110 110 110 110 111 111 111 111 112 112 113 113 114 114 115 116 116 118 118 119 119 119 119 120 120 120 120 121 121 139 140 140 140 140 141 141 141 142 142 144 144 145 145 146 146 146 147 147 147 iii Grounding Meaning in Everyday Experience in the World 3.21. 3.22. 3.23. 3.24. 3.25. 3.26. 3.27. 3.28. 3.29. 3.30. 3.31. 3.32. 3.33. 3.34. 3.35. 3.36. 3.37. 3.38. 3.39. 3.40. 3.41. 3.42. 3.43. 3.44. 3.45. 3.46. 3.47. 3.48. 3.49. 3.50. 3.51. 3.52. 3.53. 3.54. 3.55. 3.56. 3.57. 3.58. 3.59. 3.60. 3.61. 3.62. 3.63. 3.64. 3.65. 3.66. 3.67. 3.68. 3.69. 3.70. 3.71. 3.72. 3.73. 3.74. 3.75. 3.76. 3.77. 3.78. The Proximity schema The Il construction The ComplexPathPreposition construction The Al2 construction The PathPP construction The Posare1SimplePast2sg construction The DropTu construction The Tu construction The SingSubjPersPron construction The PortacenereSg construction The BoundedObject schema The BoundedRegion schema The Su2 construction The Comodino construction The AvereAuxSimplePresent3Sg construction The AvereAux construction The AuxFeatureSet schema The Aux construction The Sollevare1UnmarkedPastParticiple construction The FiniteAuxPlusVP construction The Netturbino construction The Giovane1m construction The SingularAdjective construction The ScalarModifier construction The ModifiedNP1 construction The Bidone construction The Un construction The Marciapiede1 construction The Da2a construction The ActiveCauseMotion2 construction The Spingere1Imperfect1Pl construction The DropNoi construction The Noi construction The Ceste1 construction The Le construction The Ripostiglio construction The In2 construction The PassiveCauseMotion1 construction The EssereAuxSimpleFuture3Sg construction The EssereAux construction The Scagliare1FemSgPastParticiple construction The Una construction The Pietra1 construction The Folla1 construction The ControPrep2 construction The Infiltrato1 construction The Da3 construction The CausativePreposition construction The Causation schema The CausativePP construction The VenireAuxSimpleFuture3Pl construction The VenireAux construction The Scaraventare1MalePlPastParticiple construction The Gli construction The Arbitri1 construction The Ring1Sg construction The Ring1 construction The Da2b construction 147 147 148 148 149 151 152 152 153 153 154 154 154 154 156 157 157 157 158 158 159 159 160 160 161 162 162 162 163 165 166 166 167 168 168 168 169 171 172 173 174 174 174 175 175 176 176 176 176 177 179 179 180 180 180 181 181 182 iv Figures and Tables 3.79. 3.80. 3.81. 3.82. 3.83. 3.84. 3.85. 3.86. 3.87. 3.88. 3.89. 3.90. 3.91. 3.92. 3.93. 3.94. 3.95. 3.96. 3.97. 3.98. 3.99. 3.100. 3.101. 3.102. 3.103. 3.104. 3.105. 3.106. 3.107. 3.108. 3.109. 3.110. 3.111. 3.112. 3.113. 3.114. 3.115. 3.116. 3.117. 3.118. 3.119. 3.120. 3.121. 3.122. 3.123. 3.124. 4.1. 4.2. 4.3. The Tavolo construction The Undertaker construction The ComplexPathPP construction The ComplexSPG schema The NonFiniteAuxPlusVP construction The EssereAuxSimplePresent1Sg construction The EssereAuxMaleSgPastParticiple construction The Gettare1MaleSgPastParticiple construction The DropIo construction The Io construction The Cella1 construction The Agente1m construction The Paonazzo1 construction The ModifiedNP2 construction The L'DetM construction The ComplexCausativePreposition construction The PassiveCauseMotion2 construction The EssereAuxImperfect2Pl construction The EssereFemPlPastParticiple construction The Tirare1FemPlPastParticiple construction The DropVoi construction The Voi construction The Stanza1 construction The Amiche1 construction The Ubriache1 construction The Possessor construction The Possession schema The Vostr- construction The -e construction The Possessed construction The Possessive construction The PossessivePlusNP1 construction The WhSpecifier construction The PluralWhSpecifier construction The Quali construction The WhSpecifierPlusNP construction The SubjWhSpec-Question construction The ActiveCauseMotionX construction The FrontedObjWh-Question construction The LiClitic construction The ActiveCauseMotionY construction The DeclarativePlusLeftDislocation construction The Declarative construction The DeclarativePlusRightDislocation construction The FocalPair schema The ActiveTransitive construction The Lanciare construction The Posare construction The LexVerb construction 182 184 184 184 186 187 187 188 189 189 190 190 190 191 192 192 194 194 195 195 196 196 197 198 198 198 198 199 199 199 200 201 203 203 203 204 204 205 206 207 208 209 209 210 229 229 250 250 251 Tables 2.1. 2.2. 3.1. Semantic properties of prototypical parts of speech Typological prototypes Italian possessives 94 95 232 v Abbreviations Abbreviations 1 2 3 A-S ACC AOR ASL Aux CAS CEA CG CxG CxGx CL ECG F IND INF Intr L1 L2 loc M NEG NP NTL Obj Obl PART PDP PL PP PRES PRT PST RCG rec REFL RD SemSpec SG SPG Subj SVO TL Tr V VOC VP x-net x-schema first person second person third person argument structure accusative aorist American Sign Language auxiliary Complex Adaptive System CauseEffectAction Cognitive Grammar constructionist approaches to grammar Goldberg's Construction Grammar clitic pronoun Embodied Construction Grammar female indicative infinite intransitive first language second language location male negation noun phrase Neural Theory of Language object oblique participle parallel distributed processing plural prepositional phrase present preterite past Radical Construction Grammar recipient reflexive referent descriptor semantic specification singular SourcePathGoal subject Subject-Verb-Object TrajectorLandmark Transitive verb vocative verb phrase execution-net execution-schema vii 1. Introduction 1. Introduction Introducing the present thesis is a relatively complicated affair, for two main reasons: on the one hand, this study can be seen as having a sort of double-layered nature, which characterizes several of its aspects; on the other hand, in regard to most issues, it also seems to show a nested structure. What is more, this rather complex organization is further characterized by the presence of links which establish a connection between these two properties. In the following lines, we shall try to unravel the problem and present our work in the clearest way we can. The present study can be seen as having two foci: on the one hand, it is concerned with theoretical issues on the nature of language and cognition, being aligned with two particular research programs, namely Cognitive Linguistics and Embodied Cognition; on the other hand, it also has the narrower and more concrete scope to carry out an empirical investigation of a very circumscribed phenomenon of a particular language, namely caused-motion constructions with force-exertion verbs in Italian. With regard to the latter issue, we are going to present an analysis of a small sample of homogeneous utterances, in order to provide a characterization of the various constructions (formmeaning pairings) they instantiate and the way these contribute to the overall meaning of the expressions, allowing us to characterize the online process of comprehension by which the language understander grasps the semantic content of the message correctly. The instruments we shall be exploiting to carry out this duty are those provided by the Neural Theory of Language (NTL henceforth) paradigm, an interdisciplinary research program based in Berkeley, California which involves a pool of computer scientists, linguists, and other cognitive scientists in a research effort which aims to unravel the neural correlates of human manifest behavior, with particular emphasis on language. The projects developed by NTL scholars are aligned with the theoretical assumptions of Embodied Cognition with regard to the nature of human cognition in general (with a particular focus on the working of our neural circuitry in carrying out cognitive processes), and of Cognitive Linguistics, as to the linguistic ability in particular, with a special concern on the relation between language and embodiment (see e.g. Narayanan 1996; Feldman and Narayanan 2004; Feldman 2006; Bergen and Feldman 2009). Adopting a vast amount of concepts developed by theorists of Embodied Cognition and Cognitive Linguistics, NTL researchers make use of different empirical research methods (psycholinguistic experiments, corpus-based analyses, and computational modeling), in order to test theoretical issues, looking for converging evidence to refine and correct predictions, to the purpose of aiding the formulation of empirically plausible theories and models. One of these models, called Embodied Construction Grammar (ECG from now on), will be used in the present study to carry out an analysis of a sample of Italian sentence-level constructions denoting caused-motion. In a sense, ECG can be seen as a practical tool for the use of language analysts: it is a computational formalism which stands on notions inherited from theories linking language to cognition and, in turn, cognition to the bodily and experiential nature of the organism continuously interacting with the environment. This model, designed to be precise and flexible at once, possesses formal structures which allow the analyst to investigate linguistic phenomena at the lexical, phrasal, and clausal levels, examining both their formal and semantic properties, in order to illustrate the mechanisms whereby the 1 Grounding Meaning in Everyday Experience in the World interaction between the words and the syntactic patterns instantiated in an utterance drives the understander to grasp the correct meaning of the message being conveyed. We have to emphasize the fact that, apart from the ECG formalism proper, there are also studies carried out within the NTL program whose results can turn out to be useful in phase of discussion of the results obtained carrying out the ECG analysis. Moreover, we can also highlight that, in a sense, the ECG model and the NTL paradigm in general can be considered as the point of intersection between the two levels of our study: our theoretical commitment to Embodied Cognition and Cognitive Linguistics on the one hand, and our concrete scope to provide a detailed and precise representation of linguistic data on the other. In regard to more theoretical issues, the starting point of the present study can be found in the conception of language as a cognitive function. Accordingly, we conceive language as inextricably linked to human cognition. Intuitively, from this perspective there are (at least) three key aspects to be investigated. First of all, it is inevitable to ask ourselves what kind of relation holds between language and the other facets of cognition. Second, it is necessary to inquire into the relation between language, mind, and brain. Finally, it follows that an investigation into the relation between cognition as a whole, mind, and brain will also be all-important. Obviously, providing even partial answers to these questions is definitely “a big deal”, since it entails a painstaking study of the facts which have to do with approximately all the activities which are part of our lives as human beings, thus encompassing an enormous amount of possible topics. As a consequence, in order to undertake such an ambitious enterprise, an interdisciplinary approach unifying the work carried out by a large number of scholars is required. Indeed, during the last six decades, a vast community of academics have contributed a remarkable amount of studies on these subjects. In the light of what said above, when we began to figure out the topic and the structure of the present thesis, we were conscious of the fact that a substantial amount of our background readings would have dealt with cognition, in order for us to develop our own work on language with clearer ideas. As a result, we can say that our study can be seen as falling within both the domains of linguistics and cognitive science. Since there are different points of view on the relation between these two academic fields, it is worth to clarify the position we are going to adopt about this point. In the present study, we shall make use of the term cognitive science as a label to refer to the interdisciplinary effort devoted to the study of cognition which relies on insights from basically six disciplines: philosophy, psychology, linguistics, neuroscience, computer science and anthropology. It follows that we shall employ the label linguistics to refer to the branch of cognitive science which deals with language. Consequently, to apply the simple logic of modus ponens, if a subject pertains to the field of linguistics, it also belongs more generally to the field of cognitive science. For several decades, the field of cognitive science has been dominated by a research paradigm that we label Mainstream Cognitive Science (see §2.1.1 below), adopting a theoretical approach (whose philosophical roots can be ultimately led back to Cartesian rationalism) which considers cognitive processes as computational activities which operate on amodal mental representations. In other words, this approach sees cognitive functions as processes of manipulation of abstract symbols carried out by the mind, a computational device which happens to be physically implemented in the brain. This rationalist orientation involves the adoption of the so-called “mind as a computer” metaphor, which can be represented by the following proportion: 2 1. Introduction (1) mind : brain = software : hardware From this point of view, cognitive processes can thus be analyzed independently of their cerebral realization, since their nature is considered to be independent of their physical correlates (just like the same piece of software can be run in different pieces of hardware). This kind of approach to the study of cognition was proposed and adopted by several among the most influential academics in the field of cognitive science. Among these leading scholars, we can mention the American philosopher and psychologist Jerry Fodor, who argued for the existence of a mind-internal code in which thought processes are carried out and whereby the output of perception is coded. Such a “language” is posited to be innate and endowed with an expressive potential comparable to that of a natural language (Fodor 1975). Furthermore, he also introduced the concept of the modularity of mind, according to which mental processes have a modular nature: the mind comprises several airtight “modules”, which are specialized, encapsulated systems which elaborate the output of perceptual systems and transmit an output to higher intellectual faculties (Fodor 1983). Accordingly, a rationalist approach to the study of language, Generative Grammar (see e.g. Chomsky 1965), found its way through the wide range of theoretical linguistics, becoming the dominant paradigm in the field of linguistic studies. Consistent with the view of the mind as a modular device, proponents of this approach consider language as an autonomous faculty, governed by domain-specific cognitive principles. Language, from this perspective, consists of an innate set of formal rules which operate on a list of abstract symbols. The various facets of language (phonology, morphology, syntax, etc.) represent independent submodules of the language module, with syntax representing the core component of language, while the other submodules are seen as peripheral (and thus rather uninteresting). Advocates of a generative approach to the study of language draw a sharp distinction between competence (language knowledge) and performance (language use): the scope of this formal theory of language is to model the linguistic knowledge of the native speaker of a given language, which is considered to be independent of actual language use. Moreover, human beings are supposed to be endowed with an innate set of principles of language organization, collectively referred to as Universal Grammar, which are thought to be shared across all language communities. Though probably these rationalist approaches can still be considered as the standard paradigms in the domains of cognitive science and linguistics, in the present study we are not going to commit to them. On the contrary, as mentioned above, we are going to adopt an alternative perspective, which embraces the tenets of two frameworks known as Embodied Cognition and Cognitive Linguistics, respectively. The former is a research program developed in the last decades by scholars working in the different disciplines which constitute the cognitive science field. These scholars reject the view of the human mind as a piece of computer software, claiming instead that cognition emerges from the interplay of our brains, our bodies, and the world (see e.g. Varela et al. 1991; Lakoff and Johnson 1980, 1999; Clark 2008)1. In other words, it is shaped by our daily experience as embodied agents in constant dynamic interaction with the physical as well as the sociocultural setting we are situated in (e.g. Gibbs 2005). As a 1 The reader will have noticed that the relation between cognition, mind, and brain previously mentioned is extended to include also the organism as a whole and the surrounding environment. 3 Grounding Meaning in Everyday Experience in the World result, scholars involved in this research program emphasize the fact that cognition is moulded by the bulk of situations experienced in everyday life simply by actively taking part in the functioning of a biocultural environment: kinesthetic action, empathy, perception, manipulation of objects, development, emotions, feelings, and so on. In recent years, this basic assumption has come to be supported by a substantial body of converging evidence from the various branches of cognitive science, and theories of Embodied Cognition have grown stronger and stronger (see Gibbs 2005; Bergen 2007; Shapiro 2010). At the same time, in the linguistic field, the Cognitive Linguistics 2 framework has come to be taken seriously as an alternative to Generative Grammar. Advocates of Cognitive Linguistics reject the basic tenets of Generative Grammar, asserting that the linguistic system emerges from the interaction of domain-general cognitive principles and social interactions3. Moreover, they do not conceive language as a set of formal rules operating on a list of symbols in order to generate well-formed strings; on the contrary, the central element of language is identified in its semantic component rather than its form: the basic function of language is to convey meaning, which overlaps with a portion of the conceptual knowledge of the speaker and is acquired through repeated exposure to the use of linguistic items and embodied experience in the world (see e.g. Langacker 1987). Furthermore, cognitive linguists get rid with the idea of the modularity of mind and language: linguistic units include information on several aspects of language and, at the same time, language and other aspects of cognition exert influence on each other (see e.g. Goldberg 2003; Tomasello 2003). Since a great deal of our background readings focused on the origin and nature of cognition, it is no chance that Embodied Cognition is awarded an important role in the present study: in a sense, it represents half of the theoretical background of our work and the basic foundation of our viewpoint on language and the human mind. The conception of cognition as shaped by the interaction of brain, body, and experience in the environment was crucial for the development of the present study, since a different perspective on cognition would have led to a significantly different approach to language, and the reader would now be busy in the reading of an entirely different piece of work. Furthermore, the theoretical perspective adopted on cognition might also be useful for the reader to understand the reason why (at least most) cognitive linguists share certain assumptions on linguistic issues, for the relation between Embodied Cognition and Cognitive Linguistics is rather tight: as will be highlighted more than once in the following chapters, these two frameworks can be seen as, to a great extent, mutually informative, and some of the leading scholars in one field are also key figures in the other4. So far, the reader has already been provided with an overview of several factors which display the “double-layered” nature and the “nested” structure which above have 2 There is a widespread convention among cognitive linguists to make a terminological distinction between Cognitive Linguistics (capitalized) and cognitive linguistics (uncapitalized): the former term refers to the framework being described here, whereas the latter encompasses all theoretical approaches which focus on the relation between language and mind, thus including Chomsky's Generative Grammar (see e.g. Mazzone 2004; Zlatev 2007a; Gonzálvez García 2009b). In the present study, the latter notion will be avoided. 3 Therefore, language is considered as an integrated branch of human cognition, rather than an autonomous faculty. 4 Of course, this is just a general characterization and does not mean that all cognitive linguists share the same level of enthusiasm on Embodied Cognition (compare, for instance, Rohrer 2007 with Zlatev 2007a). In turn, not all theorists of Embodied Cognition are also involved in Cognitive Linguistics. 4 1. Introduction been claimed to characterize the present study. Nevertheless, we believe that providing the reader with a brief outline of what they are going to find in the remainder of the present study, will make some more dual and/or nested facets turn up. As a consequence, now we shall proceed to supply an overview of the content of the following chapters. Ch. 2 will be dedicated to a detailed characterization of the background we have been introducing so far. As a matter of fact, the first section will focus on the two broad research paradigms briefly sketched above, Embodied Cognition (§2.1.1) and Cognitive Linguistics (§2.1.2), which represent the theoretical spine of our work, providing a more detailed description and explanation. Then, in the second section (§2.2) we shall briefly outline the NTL paradigm. First of all, we shall supply a very brief sketch of the program in general (§2.2.1), then we shall concentrate on two aspects of the NTL enterprise which are particularly relevant to the present study: first, we shall focus on the notion of simulation semantics, to which various scholars working within the NTL paradigm have dedicated a large number of studies5, providing both theoretical motivation and a substantial amount of empirical evidence for its existence (§2.2.2). Simulation semantics is important for the ECG model, since the model rests on the idea that the understanding of a linguistic utterance depends on the active enactment of its semantic content. Finally, some time will be devoted to depict the computational aspect of the projects deveoped within the NTL program: indeed, NTL scholars make extensive use of computational models in order to simulate cognitive processes. This point is relevant for us for two important reasons: first, the ECG model exploits a computational formalism to supply a description of the process of online language understanding with an adequate level of formal precision (and is also devised to be implemented on computers); second, probabilistic and connectionist computational models designed by NTL researchers can sometimes be helpful to understand the mechanisms of the process of online language comprehension (§2.2.3). The third section of ch. 2 will focus on grammar from a cognitive perspective (§2.3). First of all, it will briefly list the common properties of cognitive approaches to grammar, illustrating some of the main points on which these differ from the mainstream generative model, and will also briefly outline three among the most influential of these models, which represent, strictly speaking, the “linguistic” background of ECG. Then, the ECG model will be briefly characterized, with a discussion of the main points on which it converges with other cognitive approaches and those on which it diverges from them (§2.3.1). Finally, a more detailed account of the working mechanisms of the model will be supplied, together with a detailed illustration of the ECG formalism at work, showing the analysis of a simple Italian sentence carried out making use of an ECG grammar built on the model of the one exploited by Dodge and Bryant (forthcoming) in their study on English transitives (§2.3.2). In ch. 3, we are going to provide the reader with the report of our case-study, the analysis of a handful of (quite prototypical) Italian caused-motion constructions involving force-exertion verbs, aimed at evaluating the adequacy of the ECG model to account for facts of the Italian language. In the first section, we shall provide the reader with the relevant information on the formal and semantic nature of the phenomenon under investigation, also drawing a brief comparison with the same phenomenon in 5 In the present study, we shall confine ourselves to an overview of some recent studies carried out by Benjamin K. Bergen and his co-workers. 5 Grounding Meaning in Everyday Experience in the World English (the language with which the bulk of the ECG studies are concerned), along with the rationale for the choice to investigate this specific phenomenon in this specific language. Then, we shall supply the reader with information on the sentences analyzed in the study, including the source of the data and the criteria which led us to choose the examples to be illustrated. Moreover, we shall outline the hypothesis underlying our investigation and briefly introduce the main theoretical and practical issues of our casestudy, telling in advance the solutions adopted to handle them (§3.1). In the second section, we shall first provide a detailed analysis of the four active target sentences we dealt with in our study (§3.2.1), followed by the analysis of the passive ones (§3.2.2), supplying an explicit representation of the formal and semantic structures of all the lexical, phrasal, and clausal constructions instantiated in each sentence, using the tools of the ECG model to show the way they interplay in order to convey the message expressed by the composite sentence as a whole. Furthermore, we shall sketch some of the possible inferences that the process of analysis of each sentence on the part of the speaker may trigger. Finally, we shall suggest a way in which our grammar might be extended in order to account for less prototypical syntactic patterns (§3.2.3). The third section will be dedicated to a discussion of the phenomena observed in the course of analysis, focusing on the role of every single construction instantiated by the sentences in the dynamic activity of online comprehension on the part of the language understander, speculating about the mechanisms coming into play in the process, on the basis of the evidence provided by some probabilistic studies carried out by NTL scholars and other academics. We shall of course award particular emphasis to the main theoretical issues mentioned in the first section of the chapter (§3.3.1), arguing that their interaction is crucial in order to allow the understander to grasp the meaning of an utterance (§3.3.2). Finally, we shall also spend some time to make some observations with regard to other, less central, phenomena observed in the case-study (§.3.3.3). In the final section of the chapter (§3.4) we shall proceed to summarize the results of our case-study, briefly evaluating the strong and weak points of our grammar and assessing the explanatory adequacy of the ECG model for Italian data, proposing some suggestions for future studies. Moreover, it will also concisely outline some goals to be reached to increase the analytic power of the formalism, in order to cover a broader range of linguistic phenomena. Finally, ch. 4 will provide the reader with a concise summary of the present study as a whole and draw some conclusions, distinguishing two different levels at which our work can be appraised: not only a micro level, which has the narrow scope to evaluate the grammar we built and exploited in ch. 3 to analyze our target data and, to some extent, the adequacy of the ECG model to be applied to the analysis of linguistic phenomena of the Italian language, but also a macro level, at which it is possible to evalutate the overall assets and liabilities of the ECG model, situating it in the broader context of the NTL paradigm and, more generally, discussing its role in the world(s) of Embodied Cognition and Cognitive Linguistics, since the application of a model to a previously “unexplored” language6 may sometimes underline some liabilities that are more difficult to spot in an already investigated language. At the very end of the chapter, it will be argued that research work of scholars involved in the ECG enterprise and the other projects which are part of the NTL paradigm might fruitfully be integrated in the 6 Of course, by claiming that the Italian language is “unexplored”, we mean that it is unexplored from an ECG perspective. 6 1. Introduction unified emergentist approach which propose the idea that language should be considered as a complex adaptive system, described in the collection of papers which can be found in the volume edited by Ellis and Larsen-Freeman (2010a). This approach, which overturns the traditional assumptions of Generative Grammar (e.g. language acquisition driven by innate mechanisms, comprehension governed by formal rules, etc.) and replaces them with usage-based, emergentist accounts of linguistic systematicities, is based on widely shared theoretical assumptions and supported by a huge and fast-growing body of converging evidence from studies carried out within different academic fields adopting a plurality of empirical methods, and thus seems to present itself as a candidate to be the focal point for an even broader unified framework which might establish itself as a viable alternative to the prevailing nativist paradigm. It will be argued that, since the basic assumptions of the NTL research program virtually overlap with those of this approach, an active interaction between the ECG model and the projects carried out by the advocates of the view that language is a complex adaptive system might offer theoretical advances and make a valuable contribution to the development of a unified emergentist paradigm. To recapitulate, at the very beginning of this chapter we mentioned the fact that the present study can be better characterized as having a double-layered nature; at the same time, it was claimed that it displays a nested structure, and that these two properties are deeply intertwined. As a matter of fact, they should not be seen as two alternative keys to the reading, but rather as two inherently interrelated aspects of our work. Concluding this introduction, we shall try to make this point explicit. The present study deals with language; therefore, it falls within the field of linguistics. Nevertheless, here we conceive language as an integrated facet of human cognition, and consequently, we are also (though in a somewhat looser sense) concerned with cognition in general. As a result, the present study can be seen as falling in the broader field of cognitive science as well. This is a point in which the double-layered nature of the study is apparent. At the same time, since we view linguistics as the branch of cognitive science dealing with language, a nested structure also displays: the dual scope of our work seems to be primarily explainable in terms of levels of relations between general sets and more specific subsets. Being interested in features of language as well as of cognition in general, and in the relation between them, our study displays a commitment to two theoretical frameworks: Cognitive Linguistics, which is prominently concerned with linguistic issues, and Embodied Cognition, which encompasses all aspects of human cognition. Again, the reader can easily observe the duality of our work. Since it would be somewhat rash and unaccurate to claim that Cognitive Linguistics is the branch of Embodied Cognition dealing with language7 (at least at present), here we cannot see exactly a nested structure; however, these two paradigms share a substantial number of basic assumptions and most often interact providing useful insights to each other. As a result, they are better seen as strongly intertwined. Resting on these important theoretical considerations, which deal with very broad topics, our study is also aimed to provide an account of a very specific and circumscribed class of constructions which are part of the grammatical system of the Italian language, carrying out an investigation of a sample of data in order to shed some light on the nature of the mechanisms which underpin online language processing. Therefore, once again we can observe the pervasive dual nature of the present study: on 7 A characterization which some cognitive linguists as well as some theorists of Embodied Cognition would probably not accept, indeed. 7 Grounding Meaning in Everyday Experience in the World the one hand, it is concerned with very general theoretical subjects; on the other hand, it also has a more practical dimension, aiming to supply a characterization of a specific, empirically observable phenomenon of a particular language, in order to provide some insights on the cognitive mechanisms which underlie a particular process which takes place whenever two or more human beings are involved in linguistic communication. Here, it is possible to observe a nested structure again: both the grammatical constructions of Italian used as a means to inquire into the process of language understanding and this process itself fall within the domain of the study of language and, as a consequence, of cognition. From this perspective, as underlined above, the NTL paradigm can be considered as the link which allowes us to connect the theoretical and the practical sides of the present study: on the one hand, it relies on the conceptual apparatus developed by theorists of both Embodied Cognition and Cognitive Linguistics; on the other hand, it rests on studies carried out exploiting different empirical methods and also supplies the analyst with operational tools which permit them to perform investigations of concrete phenomena through the analysis of real data. In this way, the NTL program serves as the bridge between several dual aspects of our study: theory and practice, linguistics and cognitive science, Cognitive Linguistics and Embodied Cognition. As a result, while it temporarily “neutralizes” the dual nature of our study, it is part of all the elements of our study mentioned above, highlighting the nested structure of our work once more. It was already emphasized that the NTL program proved helpful for our analysis in two different but related ways (and here the double-layered nature of our work strikes back): first of all, one of the projects developed by NTL researchers, i.e. the ECG model, was used to investigate our data and provide the reader with a detailed characterization of the online process of language comprehension undertaken by the understander to grasp the meaning of the sentence; second, a series of studies carried out by NTL scholars on the statistical and incremental nature of language processing allowed us to make some observations on the mechanisms whereby this cognitive function is performed. Since both the ECG model and the probabilistic studies just mentioned represent projects which are part of the NTL paradigm, we can see the nested structure of our study emerge again. Furthermore, during the evaluation of the analysis carried out in our case-study we distinguished two different levels to assess: one more strictly related to the efficiency of the ECG grammar we built to account for the sentence-level constructions analyzed and more generally to the adequacy of the ECG formalism to deal with Italian data; the other directed to provide an appraisal of the current assets and liabilities of the ECG model overall. Again, we can observe that our study displays both a double-layered nature and a nested structure. While the brief discussion of the organization of our work supplied above may result somewhat prolix and repetitive, it was meant to provide the reader with a point-by-point explanation of the reason why we consider the present study to be double-layered and nested. From our point of view, this was an important point to be made, in order to provide the reader with a suggestion on how the present study should be approached. As a matter of fact, the simultaneously dual and nested organization of our work can be seen, in a sense, as a reflection of our belief according to which the general and the specific, the theoretical and the practical, the whole and the part, the abstract and the concrete, are deeply intertwined. This perspective will often emerge throughout the present study: indeed, even though it will be not explicitly mentioned, (keeping in mind what said above) the reader will be able to notice that the constant tension and mutual 8 1. Introduction influence between the two elements included in the pairs listed above is a recurrent theme in the following chapters8. 8 In a sense, this view might also be considered (at least partially) to lie at the roots of the hypothesis proposed at the very end of ch. 4, according to which the results empirically obtained by the studies carried out by NTL scholars, and those achieved by supporters of the approach which conceives language as a complex adaptive system (which share virtually the same assumptions and methodology) seem to support the intuition that an interaction between these two enterprises might represent a step toward the emergence of a broader unified framework able to stand as an alternative to the prevailing nativist paradigm in cognitive science, challenging it on theoretical as well as empirical grounds. 9 2. From Cognition to Grammar 2. From cognition to grammar In this chapter, we shall outline the theoretical background which can be seen as the “backbone” of ECG, the approach adopted in the case study illustrated in ch. 3 below. The present chapter is divided into three sections, differing from one another in terms of abstractness/concreteness and generality/specificity. The first section will briefly introduce Embodied Cognition, a philosophical and scientific reference paradigm, which represents the very foundations of the present study. Moreover, §2.1 will briefly outline Cognitive Linguistics, a linguistic framework which includes a collection of different but related and largely compatible approaches to the study of language, highlighting the connection between the Cognitive Linguistics enterprise and the more general Embodied Cognition research program. In the second section, we shall introduce the NTL paradigm, the interdisciplinary research program aimed to uncover the complex and controversial relationship between language, mind, and brain within which ECG is a project. While researchers involved in the NTL program deal with a wide range of topics, we shall focus on a couple of subjects which are particularly relevant to the purpose of our study: simulation semantics and computational modeling. Finally, in §2.3, which represents the main part of this chapter, we shall briefly introduce three cognitive/constructionist approaches to grammar which exerted a remarkable influence on the development of the ECG formalism and, at last, we shall provide the reader with an overview of the ECG model itself. 2.1. Embodiment, cognition, and language In this section, we are going to introduce the very starting point for the present study: the belief that language and other human cognitive abilities are tightly interconnected, and that cognition is grounded in everyday embodied interactions with the physical and sociocultural environment. The first subsection will introduce the Embodied Cognition research program, which challenges the tenets of Mainstream Cognitive Science, denying the fact that cognition is a purely intracranial affair which involves the manipulation of abstract symbols by means of formal rules, and claiming that instead the mind extends beyond the brain to the whole organism, and is shaped via bodily experience in the world. The second subsection will briefly outline the Cognitive Linguistics enterprise, a growing paradigm in the field of linguistic theory which covers a set of projects arisen as a reaction against the formal linguistics tradition, with the intention to show that language is not an autonomous rule-based computational system, but rather an integral aspect of human cognition. Throughout the section, it will be often highlighted the fact that Embodied Cognition and Cognitive Linguistics are tightly connected, sharing many theoretical assumptions and being backed by converging empirical evidence. 11 Grounding Meaning in Everyday Experience in the World 2.1.1. Embodied Cognition Cognitive science, an interdisciplinary research enterprise aimed at the study of human cognition, came to light in the 1950s riding on the wave of the advances in computing technology, and was strongly influenced by the work of academics such as Alan Turing, Jerry Fodor, and Noam Chomsky. Basically, scholars working in the framework of Mainstream Cognitive Science1 hold the view that the human mind works as a sort of Turing machine or, alternatively, as a piece of computer software devoted to the formal manipulation of abstract symbols, which happens to run in the hardware, i.e. the brain and the central nervous system and, only by extension, the human body (the so-called “mind as a computer” metaphor). As a consequence, the capacity to perceive, think, use language, and all complex cognitive processes are considered as functional states resulting from the manipulation of symbols following precise algorithms independent of the physical system in which they are implemented2. It is assumed that the mind is endowed with a sort of a priori forms (similar to those posited by Kant in the 18th century), which guide all our potential knowledge of cognition and the world. These forms are considered to be real and independent from experience. In order to provide a clearer account of cognition, theorists of Mainstream Cognitive Science often divide the cognitive architecture into three intertwined levels: i) the computational theory (which clarifies the scope of the computation), (ii) representation and algorithm (which gives a representation of the input, then transformed into an output by the algorithm); (iii) the hardware implementation (the physical realization of algorithm and representation). In this perspective, knowledge can only be transmitted from an interlocutor to the other through a shared symbol system. The first level is only semantic, while the second is only symbolic: a nonsense but well-formed sentence (consider e.g. Chomsky's famous example “colorless green ideas sleep furiously”) is problematic at the first level, whereas an ill-formed but meaningful English sentence may cause some problems at the second level, but both may successfully transmit a message. A solution to a problem may be found making use of several different algorithms. Similarly, in cognitive psychology each level is independent from the lower ones. Therefore, knowledge does not rely on symbols, and neither knowledge nor symbols rely on the hardware level. As consequence, it is possible to implement the same higher level in different types of hardware (e.g. a human being vs. a robot). In other words, cognition is essentially seen as autonomous from the body, and although nowadays cognitive scientists are generally no longer committed to Cartesian dualism and the sharp distinction between body and mind inherited from several centuries of mainstream Western philosophy, the phenomenological body is rarely recognized a constitutive role in human cognition: as pointed out by Gibbs (2005: 5), often mental activity is reduced to a series of brain processes and, as a result, “In some cases, the reduction of mind to brain carries with it the reduction of body to brain.” 1 This framework is also often referred to as “first-generation cognitive science” (e.g. Lakoff and Johnson 1999), “classical cognitive science” (e.g. Sinha 2007), or “standard cognitive science” (e.g. Shapiro 2010). Here, we prefer, to adopt the label “Mainstream Cognitive Science”, but this is just a matter of terminological choice. 2 From the Internet Encyclopedia of Philosophy: “The cognitivist/classicist research program can be defined as a rule-based, information-processing model of cognition that 1) characterizes problemsolving in terms of inputs and outputs, 2) assumes the existence of symbolic, encoded representations which enable the system to devise a solution by means of computation, and 3) maintains that cognition can be understood by focusing primarily on an organism's internal cognitive processes (that is, those specifically involving computation and representation).” 12 2. From Cognition to Grammar Some of the leading scholars in Mainstream Cognitive Science (notably Fodor) are also committed to the modularity of mind, i.e. they propose that the mind is composed of genetically specified, encapsulated modules which developed for different functional purposes (e.g. vision, motion, language, etc.) and can operate without reference to other psychological systems. Information from the environment passes first through a system of sensory transducers that convert the data into formats that each specific module can process. Then, each module outputs data in a common format for central, domaingeneral processing, analogous to the CPU of personal computers. During the last decades, an alternative paradigm has arisen within the cognitive science enterprise, which rejects the “mind as a computer” metaphor and calls into questions the basic tenets of Mainstream Cognitive Science. This framework is generally known as Embodied Cognition (or “second-generation cognitive science”)3 and its theorists claim that the very fact that we have the kind of body we have, with its structure and its abilities, and our embodied interactions in and with a specific environment, shape the way we think, reason, and act: bodily experience is considered the primary source of human cognitive processes. Cognition, therefore, can be seen as the result of the interaction between the brain, the body as a whole, and the (physical as well as socio-cultural) world. This claim is supported by a growing body of empirical evidence from the different disciplinary fields which are assumed to be part of the larger cognitive science enterprise, which include philosophy, psychology, neuroscience, artificial intelligence, linguistics, and anthropology. However, this plurality of sources is a sort of double-edged sword: while on the one hand such a wide range of converging evidence can be considered an asset which strengthens the claims for a model of cognition which takes into account the role of the body and its interactions with the world, on the other hand, this multiplicity of sources can also represent a problem, especially on the way to the formulation of a single unified theory. As a matter of fact, at present there are several “embodied” approaches to the study of cognition, which often focus on different aspects of cognition and, as a consequence, hold different perspectives on embodiment, sometimes not fully consistent with one another. Consequently, we find it suitable to follow Shapiro (2010: introduction) in considering Embodied Cognition as a (relatively fluid) growing research program rather than a precisely defined theory. On the basis of this observation, in the present section we find it more convenient to provide the reader with only a brief outline of the approach adopted in this study, rather than supplying fragmentory information on several strands of Embodied Cognition. In the present study, we are going to adopt an approach which, in a sense, can be regarded as quite close (though not perfectly correspondent) to a point of view which has been quite influential especially among the Cognitive Linguistics community, generally labeled experientialism or embodied realism, extensively advocated for in Lakoff and Johnson (1999), whose core concept can be, in our opinion, summarized in the following excerpt (see also Rohrer 2007): What we understand the world to be like is determined by many things: our sensory organs, our ability to move and to manipulate objects, the detailed structure of our brain, our culture, and our 3 It should be stressed that Embodied Cognition is neither the only nor the first alternative proposed to Mainstream Cognitive Science, which regularly faces other challengers such as, for instance, ecological psychology and connectionist modeling. Actually, it is no chance that theorists of Embodied Cognition often “cast a glance at” and interact with these two paradigms, especially the latter. 13 Grounding Meaning in Everyday Experience in the World interactions in our environment, at the very least. What we take to be true in a situation depends on our embodied understanding of the situation, which is in turn shaped by all these factors. Truth for us, any truth we can have access to, depends on such embodied understanding. (Lakoff and Johnson 1999: 102, italics original) Owing to its strong influence, Lakoff and Johnson's 1999 book is generally considered a manifesto of the Embodied Cognition research program and it also represents one of the major theoretical pillars of the NTL paradigm we shall be dealing with in §2.2. Nevertheless, while the importance of Lakoff and Johnson (1999) for the development of the present study is beyond doubt, we prefer to refer to Gibbs (2005) as our main reference text. Actually, Gibbs's approach is mostly compatible with Lakoff and Johnson's, with which it shares virtually all the basic assumptions, but Gibbs' book avoids some of the potential drawbacks which can be found in Lakoff and Johnson's work (see also Johnson and Lakoff 2002), for a variety of reasons. The first remarkable difference between the two texts rests on the different range of their scope. Indeed, Lakoff and Johnson (1999) goes far beyond the presentation of their particular embodied approach to the study of cognition and the mind, which is essentially an updated and more radical version of the perspective adopted since Lakoff and Johnson (1980): in their extensive 1999 book, the two scholars assert that Embodied Cognition in general and their approach in particular call into question the basic tenets of centuries of Western philosophy, culture, and thought, which in their opinion need to be deconstructed and rethought in the light of the assumptions of the philosophy of embodied realism. Indeed, after outlining the starting points of their approach, Lakoff and Johnson (1999) undertake an analysis of some basic philosophical ideas in the light of their view of the embodied mind, and then review the theories of several major Western schools of philosophy, arguing that all of them were basically wrong. On the other hand, the scope of Gibbs' book is limited to providing a detailed exposition of the reasons why the properties of the human body and our everyday sensorimotor experience in the world we live in should not be disregarded but instead awarded a prominent role in any realistic account of human cognition. Since a detailed discussion of embodied realism and its (potential) implications for Western philosophy is not our concern here, the narrower scope of Gibbs' work seems more in line with the present study. Moreover, the advantages provided by more humble objectives and a more moderate positions are remarkable, especially in terms of potential criticism. In fact, while Lakoff and Johnson's comprehensive and ambitious book provoked a heated debate among scholars and exerted a strong influence on subsequent work, it was also subject to harsh criticism, sometimes based on strong arguments and supported by a fine-grained analysis (see e.g. Zlatev 2007a). Second, Gibbs agrees with Lakoff and Johnson in the recognition of three different levels of embodiment, also providing us with a short but clear definition of each of them. Following Lakoff and Johnson (1999), Gibbs (2005: 39-40) posits the existence of a neural level, a cognitive unconscious, and a level of phenomenological conscious experience. The first level regards the neurophysiological structures which characterize our concepts and cognitive operations, since concepts and experience are embodied in the brain. Nevertheless, Gibbs specifies that neural assemblies do not simply receives input from the enviroment and provide output to the body, but operate in relation to the entire body as it functions within concrete situations. The second level, the “cognitive unconscious”, consists of all the unconscious mental operations underlying conscious experience. This level both uses and guides the perceptuo-motor aspect of our bodies 14 2. From Cognition to Grammar and includes all our unconscious knowledge and thought processes. At this level, the importance of the body is critical, for all our cognitive structures “are grounded in patterns of bodily experience and activity”. Finally, the phenomenological level consists of everything we can be aware of (our mental states, our bodies, the environment, etc.) and it is the level at which we feel experience. However, while Lakoff and Johnson (1999) award the cognitive unconscious a prominent role in the execution of cognitive tasks and try to venture (at least potentially controversial) guesses regarding the functioning of this level4, Gibbs avoids to quantify the role of the three levels, preferring to simply assert that “explanations at all three levels are necessary for an adequate account of the human mind” (Gibbs 2005: 40), thus conveniently getting around the risk to underestimate the role of consciousness in cognitive activities. A further difference between Lakoff and Johnson (1999) and Gibbs (2005) regards the notion of mental representation. The fact that this notion can be problematic is hardly surprising, as testified by the fact that it has always been a matter of debate among philosophers, psychologists, and linguists and a solution to the problem seems (at least, at present) improbable. However, for simplicity's sake, we shall follow Shapiro (2010: 217) and simply reduce it to the crude definition of “something that is used to stand-in for something else”, (cfr. the Saussurean notion of sign). Lakoff and Johnson take a strong anti-representational stand, vigorously denying the existence of mental entities referring to objects in the external world, replacing representations with the notion of “interaction”, as later reasserted in Johnson and Lakoff (2002: 249-250): “the only workable theory of representation is a flexible pattern of organism-environment interactions, and not some inner mental entity that somehow gets hooked up with parts of the external world by a strange relation called 'reference'”. Nevertheless, while the notion of abstract, arbitrary symbolic representations used in Mainstream Cognitive Science is (from an embodied outlook) untenable and need to be radically revised, Lakoff and Johnson's denial of the notions of “representation” and “reference” seems a bit too strong, as underlined by Zlatev (2005: 318): it might be rash to define, say, an apple simply in terms of neural structures and the interaction between an organism and the environment, excluding any level of abstract representation arising from such an interaction5. Gibbs, unlike Lakoff and Johnson, does not reject the notion of representation: in his work, he preserves this notion, often specifying that “embodied action underlies part of conceptual representations”. He provides evidence in support to the necessity for an embodied approach to cognition drawing from the work of both scholars who admit the existence of representations (e.g. various representatives of cognitive psychology) and others who do not (notably, supporters of the Dynamic Systems Theory). Cautiously, Gibbs confines himself to highlighting the contributions to the Embodied Cognition research program made by scholars working in different frameworks and discussing their assets and liabilities on the basis of empirical evidence, including the results of his own research. Though he does not explicitly argue for representations, he does not rule them out; on the contrary, most often his beliefs seem to lean toward their existence, though with a low level of abstractness grounded in patterns of embodied experience6. A similar, less radical attitude on representations is 4 As argued by Zlatev (2005, 2007a), some points of Lakoff and Johnson's exposition seem to be somewhat confusing and contradictory, sometimes making their arguments rather cryptic. 5 Gallese and Lakoff (2005) offer a more moderate point of view (closer to Gibbs') without giving up the assumption that representations have an interactional origin based on the brain's sensorimotor activity. 6 For a discussion of the nature of representations and a review of some contemporary theories of 15 Grounding Meaning in Everyday Experience in the World more in line with the present study and, perhaps, more consistent with the ECG model which, as we shall see in §2.3 and in ch. 3, makes an extensive use of the notion of “reference”. Moreover, Gibbs' work is somewhat more attentive than Lakoff and Johnson's to the sociocultural component of human cognition. As a matter of fact, while Lakoff and Johnson occasionally recognize the importance of the role of the social and cultural background in shaping cognition, they generally neglect this aspect in favor of a more intrasubjective dimension of cognitive processes, somehow underestimating the notion of situatedness, which can be seen, in a sense, complementary to that of embodiment (see e.g. Sinha 2007). Indeed, since human beings are social creatures, most of human activities are strongly intersubjective (language, for instance, is related to the need to communicate with conspecifics). Therefore, in order to realistically account for human cognition it seems necessary to integrate an account of its intrasubjective dimension with an account of its sociocultural dimension (as suggested in several studies such as, for instance, Tomasello 1999, 2003; Sinha 2002, 2007). Accordingly, Gibbs (2005) explicitly asserts that cultural background is an essential aspect of human cognition: Rather than being a biological given, embodiment is a category of sociocultural analysis, often revealing complex dimensions of the interactions between bodies and personhood. Talk of embodiment is situated in reference to topics such as health and illness, kinship, modes of production and exchange, gender and age hierarchies, language practics, religious and political disciplines, jural rules, pervasive metaphors, spirit possession, historical experiences, and myths. People's bodies are more than surfaces for social inscriptions (i.e, “the body as text” metaphor), but incorporate cultural meanings and memories. Culture does not just inform embodied experience; embodied experience is itself culturally constituted (...) Certain aspects of sensory perception are emergent and dependent on culture, which influences the embodiment of dispositions through everyday practice (...) (Gibbs 2005: 37) Diametrically opposed to what claimed by advocates of Mainstream Cognitive Science, this view thus depicts cultural variation as an intrinsic property of human cognition. As a result, although the intrasubjective dimension of cognition is perhaps still awarded a prominent role, Gibbs' work also encompasses its sociocultural aspect, sidestepping one of the most common criticism directed toward embodied approaches to verbal communication7. Furthermore, while Lakoff and Johnson (1999) are sometimes blamed of providing too little empirical evidence in favor of Embodied Cognition, this kind of criticism can hardly be extended to Gibbs (2005). Indeed, we believe that one of the major assets of Gibbs' book is the richness of empirical evidence the author mentions and discusses to sustain his theses in favor of an embodied outlook on the human mind. Actually, Gibbs provides the reader with an impressive amount of converging evidence several of the disciplines which populate the realm of cognitive science: while the majority of the work cited in his book is represented by experimental evidence from psychological studies, Gibbs also makes reference to a substantial amount of work carried out in the fields of neuroscience, artificial intelligence, linguistics, and, though maybe to a minor extent, anthropology and philosophy. What is more, Gibbs backs his theses mentioning representation, the reader is referred to Vigliocco and Vinson's (2007) contribution. 7 The recognition of the constitutive role of the sociocultural background in human cognition also seems to prevent Gibbs's work from the accusation of “physico-biological reductionism” (see e.g. Itkonen 2003; Zlatev 2007a) Lakoff and Johnson have been charged with, together with other linguists and cognitive scientists . 16 2. From Cognition to Grammar a small but significant amount of clinical evidence from neurological studies carried out on impaired subjects, whose lack was one of the causes of Zlatev's (2005) severe criticism of a substantial amount of work carried out from the perspective of Embodied Cognition (with particular reference to the work of several cognitive linguists, especially Lakoff and Johnson and their collaborators). Finally, it seems relevant to underline that while Lakoff and Johnson are usually prone to make strong claims on the basis of their studies, another arguably positive side of Gibbs' (2005) stands on the carefulness of the author in drawing his conclusions: though his theses are generally based on a solid background and supported by a considerable body of empirical evidence, normally Gibbs does not take for granted that his observations are completely and undisputedly correct; on the contrary, he often leaves room for possible complementary or even alternative explanations8. Now that we have outlined the rationale for taking Gibbs (2005) as our main reference point, we can procede to provide the reader with a brief explanation of the particular perspective adopted in the present study9. We may state that the basic starting point of this perspective on human cognition regards the fact that the body is not just a sort of physical accessory which is necessary to allow the mind to perform its cognitive activities; instead, it is a constitutive part of our “self”, which is not something trascendent and disembodied, but rather is grounded in our embodied perception and action10. Indeed, since a child is born, they grow into their bodies. Infants gradually discover that the various parts of their bodies are each capable of particular movements. Kinesthetic action plays a role of paramount importance in infants' physical experiences of the world, which allows them to progressively build more complex mental understandings of basic concepts (e.g. containment or balance) and the consequences of their actions upon the external world. Human beings are then seen as having both a body schema and a body image. The former labels the way in which the body integrates posture and position in the world. It is made up of unconscious and constantly updated information about the body which provides feedback for the execution and the coordination of our motor program, and it is essential to both the way we move and the recognition of our unique identity in the world. The latter, on the other hand, refers to conscious representation of the body, including how it works as an object of feelings and emotions. Body schema and body image are not independent of each other: on the contrary, there are important interactions between them. For instance, the unconscious operation of the body schema influence significant aspects of our conscious experience of the body. As an example, changing how the body performs affects people's perceptions of their bodies, together with their perceptions of the environment11. An important aspect of embodiment regards our relation with others, which is fundamentally based on empathy. In particular, it is relevant here to mention ideomotor action, i.e. body movements that tend to arise in observers when they watch other 8 An acquaintance of ours with years of work experience in the field of physical therapy pointed us out that this kind of caution is particularly important when dealing with clinical cases, for sometimes different subjects affected by the same neurological impairment can show contrasting behaviors. 9 It seems relevant to remind the reader that other schools of thought in Embodied Cognition may share several of the basic points of view of the approach adopted and outlined here (see Shapiro 2010). 10 Following Gibbs (2005: 14), we may say that “a body is not just something that we own, it is something that we are.” 11 Gibbs (2005: 33) also underlines the fact that “neurological damage can have terrible effects on people's perceptions of their own bodies and their bodily experiences. When people's perceptions of their bodies are disturbed, they suffer mental/emotional disarrey.” 17 Grounding Meaning in Everyday Experience in the World people performing specific action. As Gibbs points out: Contemporary research suggests that simulation mechanisms provide the common code between perception and action, such that perceiving an action induces the production of a similar act, or the urge to act, in the observer (...) When we see someone perform an action, the same motor circuits that are activated when we perform such actions are concurrently activated (...) Activating shared representation between perception and action is clearly an important element in the experience of empathy. But empathy is deeply grounded in the experience of our lived bodies, and experience enables us to directly recognize others, not as bodies endowed with minds, but persons like us (...) One possibility is that the mechanism for this sharing of experience is simulation (...) by modeling some behavior, our actions provide a simulated representation of the same process that can be used to produce it, on the one hand, and to decode it when performed by someone else, on the other. (Gibbs 2005: 35-36) The concept of simulation of an action performed by others carried out in the observer by the same mechanisms which are activated when they perform the same action themselves, which is very important for Embodied Cognition and for the NTL program, will be addressed in more detail in §2.2.2. For the time being, it is important to keep in mind that there is a substantial amount of research highlighting the fact that being aware of our bodily responses is essential in order to understand and feel emotion for other people's experiences (see e.g. Gallese 2006). One of the foremost claims made by theorist of embodiment regards the crucial role of direct embodied action in the environment in shaping an individual's perception, which is not, consequently, considered to be the result of computational operations which simply take place in the brain of the perceiver, but rather the result of the situated, dynamic interaction between the whole organism in action and the world12, which in a sense co-determine each other: on the one hand, the structure of the environment constrains the activity of the organism, and at the same time the activity of the organism affects the shape of the environment. Action is considered to be an essential condition for perception. For instance, experiencing the movement of one's own body in action is considered to be responsible for the perception of causality. In this perspective, children are thought to acquire knowledge through their repeated direct perception of causal relations as they interact with objects and other people. Gibbs asserts that this knowledge of causal relation, over time, can be used to interpret other events in which causal relation is not directly perceived through our embodied experience (e.g. an action performed by other people)13. Gibbs (2005: 57-63) reports that several studies carried out in the field of neuropsychology seem to show that the perception of an object involves the activation of its affordances14, thus supporting the idea that action and perception are deeply interrelated, and possibly share common neurological mechanisms: “A growing body of 12 This point is clearly exemplified in an influential study on color perception carried out by Varela et al. (1991: ch. 8). 13 This posited process reminds us of some notions adopted in Cognitive Linguistics, such as the usagebased model of grammar and constructional polysemy, (see §2.3 and the next chapter), thus (indirectly) arguing against the modularity of mind. 14 “Affordance” is “J.J. Gibson's term for a property of an object that invites opportunities for particular sorts of action” (Shapiro 2010: 211). Though neither Shapiro nor Gibbs address this topic, it seems plausible that, for certain objects, two persons with different sociocultural backgrounds will activate different kinds of affordances. 18 2. From Cognition to Grammar research has demonstrated that people readily perceive objects in terms of the possible bodily action they perform (...) The results of these studies are consistent with the idea that anticipated bodily interactions are a significant part of perceptual experience” (Gibbs 2005: 61)15. We have also to clarify that research also seems to suggest that the perception of an object does not involve the activation of all its affordances together, but instead distinct contexts are likely to activate different affordances for the same object. The relation between perception and action in this embodied perspective can be summarized with the following excerpt from Gibbs' book: ...perception involves bodily movements of various sorts and the anticipation of action when adapting to environmental situations. In this way perception establishes a reciprocal relation in the physical coordination of the organism with the environment. This kind of body-world structural coupling is fundamentally grounded in self-movement. An embodied approach to perception and action sees these as dynamically intertwined, in that the physical properties of real world are not entities to be statically perceived, but are opportunities for action. (Gibbs 2005: 77) Action is also awarded an important role in conceptualization. Indeed, in this perspective concepts are considered to arise from and be structured in terms of patterns of embodied activity. Supporters of this embodied perspective take issue with the traditional view of concepts. According to the traditional view, concepts are stored mental representations which directly reflect the nature of objects and events in the external world, categorized on the basis of their correspondence to necessary and sufficient conditions and hierarchically organized in terms of their objective properties16. A property is assumed to be represented across different categories by a single amodal, disembodied symbol, independent of language and context. The example of a category judged “best” is labeled “prototype”, and categories are considered to be mentally represented in terms of prototypes, with the degree of category membership being determined by the degree of similarity to the prototype. From the perspective of Embodied Cognition, this view is inconsistent with a number of empirically demonstrated facts. First of all, conceptualization is highly sensible to situational and cultural context, and so there is no 1:1 correspondence between mental concepts and external objects; instead, people are likely to represent the same property “locally” in different contexts rather than “globally” as a single symbol (e.g. the red of red hair is not the same red of blood). Moreover, as argued by scholars such as Lakoff (1987), categorization is centered on basic-level categories, cognitive categories in the middle of taxonomic hierarchies, a level which has been empirically shown to be the most salient, since it is the level at which people interact optimally with the world (given the structure of their bodies and brains and the environment they inhabit). In other words the basic-level is (Gibbs 2005: 82, from Lakoff 1987: 46): • the highest level at which category members have similarly perceived overall shapes; • the highest level at which a single mental image can represent the entire category; • the highest level at which a person uses similar motor actions in interacting with 15 A representative example of this kind of studies can be found for instance in Gentilucci (2003). 16 For instance, the concept “cars” will fall into the middle level of a taxonomic hierarchy with “vehicles” as a superordinate and specific kinds of cars (e.g. “station wagon”, “sedan”, etc.) as subordinates. 19 Grounding Meaning in Everyday Experience in the World category members; • the level at which most of our knowledge is organized. With regard to prototypes, supporters of the embodied approach here in focus see them as sensory, imagistic mental events shaped by real-world bodily experience and closely tied to individual contexts, rather than rigidly static representations, and conceptual categorization is considered to be highly flexible, constantly evolving on the basis of the continuous update of our background knowledge on the basis of new interactions with the world. In other words, “What people learn from their interactions with items is incorporated into their concepts and consequently affects categorization.” (Gibbs 2005: 85). These observations suggest that concepts are, at least partially, temporary constructions in working memory, based on embodied simulations, rather than fully stable structures stored in long-term memory. Gibbs suggests that concepts may be seen as statistical patterns in sensorimotor systems that take different form in different context. In order to account for the perceptual origin of concepts, supporters of this embodied approach exploit the notions of perceptual symbol, image schema, and conceptual metaphor. With regard to the former, Gibbs (2005: 86) observes that perceptual symbols “are derived from the representations generated from perceptual input systems, but are acquired by performing operations on perceptual representations and are similar to these operations”. As a result, they are schematic but similar to the object they represent, unlike the amodal symbols posited by traditional view of concepts: salient aspects of a perceptual state (including both sensory states and mental events) are extracted by selective attention and stored in long-term memory. When the sensory input is absent, the stimulus is internally re-enacted by the activation of conjunctive neurons: these re-enactments allows context-specific representations of a category. Then, we can state that perceptual symbol theory (which is generally traced back to some studies carried out by Lawrence Barsalou) suggests that the conceptualization of a category includes background information partly based on people's embodied simulations of acting in determined contextual situations. This re-enactment is not necessarily complete, but each conceptualization represents a category in a way that is relevant to the specific, relevant situation. Perceptual symbols can thus be seen as representations arising online, from the interaction of background knowledge and a specific context. The notion of image schema, introduced by scholars working in the field of Cognitive Linguistics (in particular, see Johnson 1987; Lakoff 1987), is often exploited in studies dealing with spatial semantics and force-dynamics. They can be considered dynamic representations of spatial relation and movement derived from perceptuomotor processes. Gibbs (2005: 91) defines image schemas as “imaginative, nonpropositional structures that organize experience at the level of bodily perception and movement”17. Gibbs provides the illustration of the SourcePathGoal schema (SGP), an instance of image schema (which will be extensively used in the case study illustrated in ch. 3) which is thought to emerge as a result of visual experience in the world. Moving their eyes through their visual field, an infant is able to recognize a pattern of trajectory from a point in space to another. Later, experiencing our bodies moving in the environment (reaching objects, moving from one place to another, etc.), a 17 This characterization of image-schemas is essentially faithful to Johnson's (1987: 29) original definition: “A schema is a recurrent pattern, shape, and regularity in, or of, these ongoing ordering activities [i.e. actions, perceptions, and conceptions].” 20 2. From Cognition to Grammar variety of SPG experiences become salient. All these different experiences rest on the same image schemas, which supports both literal and metaphorical meanings. Metaphor, traditionally considered a linguistic artefact, was introduced as a constitutive component of human thought and reason by Lakoff and Johnson (1980). These scholars argued that we conceptualize abstract objects and situations (source domain) in terms of more concrete objects and situations often based on bodily experience (target domain) using the cognitive mechanism of cross-domain mapping. Metaphorical conceptualization gives rise to a number of conventional expressions in possibly all languages18. For instance, if we observe even a small amount of English data, it will be quite easy to spot several conventional expressions cluster together under a few common metaphorical systems of understanding, involving tight mapping from entities in the source domain to entities in the target domain. An often cited instance of a conceptual metaphor is “love is a journey” which underlies several conventional expressions (e.g. “our marriage is on the rocks”, “we are spinning our wheels”, etc.) and is based on the SourcePathGoal schema. As Gibbs (2005: 93) points out, it seems that “there are direct connections between recurring bodily experiences, metaphorically understood abstract concepts, and both conventional and creative language that refer to these abstract concepts.” In summary, from this perspective (both concrete and abstract) concepts are, to a great extent, temporary, dynamic, embodied, and situated representations, which arise from acts of perceptual simulation and are not merely accessed as static representations in long-term memory19. This perspective allows to explain the flexibility, multimodality, and productivity of concepts, which also give rise to inferences based on the situational context. This is not to neglect the role of representations, as Gibbs clarifies: My advocacy of embodied simulation in the creation of concepts in context does not necessarily imply that the sensorimotor nature of conceptual processing is inherently non representational. After all, simulation processes operating to create specific concepts in context use various kinds of knowledge, including that about the body, that is represented. These simulations are not identical to the neural states that underlie perception, action, and cognition. But conceptual simulations surely involve brain processes in cooperation with the entire nervous system and body to create imaginative understandings of events, both when environmental information is present, and when it is not. (Gibbs 2005: 121-122) This embodied perspective also has, of course, some important implications with regard to mental imagery, which is claimed to be linked to kinesthetic activity. Indeed, a substantial body of research shows that imagery and action extert strong influence on each other: for instance, a young person thinking of an elder one, can be unconsciously induced to slow down their pace. At the same time, aspects of imagination are shaped by bodily action: e.g., the mental rotation of a hand is more quickly accomplished when the hand is physically rotated than when it is not. Moreover, there is evidence that mental representations of a human body in motion is constrained by the the same biomechanical factors that limit actual movements: planning movements underlies both 18 It seems relevant to clarify that metaphorical systems are not exactly universal. Though there is often a certain level of overlap between two different languages, it is possible to observe a wide range of cultural variation, as illustrated by a body of linguistic studies, such as Maalej's (2008) contrastive work on “heart”-based metaphors in English and Tunisian Arabic. 19 This is not equal to claiming that concepts have no stable properties deposited in long-term memory, but that these are just one of the factors involved in the dynamic process of conceptualization. 21 Grounding Meaning in Everyday Experience in the World action and mental imagery performance (cfr. the anticipation of an action mentioned above), and predicting the result of future action seems to require the ability to run an internal representation of the situation and then inferentially draw a conclusion on the basis of this representation20. There is also a body of neuropsychological studies supporting the claim that basic processes underlying action are activated in absence of physical movement. For instance, brain-imaging studies show that the activation of the sensorimotor cortex occurs even when people perform mental tasks without actually moving. What is more, the frontal motor cortex is activated when the subject observes movements performed by somebody else. Moreover, lesions in the parietal cortex strongly interfere with people's ability to guess the outcome of an action right. Gibbs (2005: 134) asserts that “conscious imagery is consequently essential for the planning of human action, and embodied movement provides the roots for conscious experience, emphasizing the close connection between action and imagery.” Then, according to Gibbs (2005: 136), “Mental imagery (...) is a kind of simulator of action that is based on real-life actions and potential actions that a person may engage in.” This means that, being a simulator, mental imagery provides a kinesthetic feel that cannot be simply reduced to the output of abstract computational operations of amodal symbols, supplying the “flavor” of bodily experience in a three-dimensional world. Memory and reasoning are also affected by embodied activity. The former can be seen as made up of both internal representations and manipulations of environmental structures. As a matter of fact, when repeatedly faced with a particular memory task, we get used to transform the environment in order to simplify our duty21. An instance of this fact is offered by experiments aimed at explaining expert bartenders' ability to dispense drinks when faced with many orders in a noisy and crowdy environment: comparing their performance with that of novice bartenders, it was observed that experts select differently shaped glasses at the time of ordering, then using these cues to recall and sequence the specific orders. As a result, their performance significantly worsen when tested on uniform glassware. This difference in performance is not observed in novice barmen. Thus, it is apparent that the experts' skill derive from the interaction of internal representations and environmental components (they have learned to transform the environment in order to reduce their cognitive workload). Moreover, the task of remembering is eased by our embodied experience. As an example, people's memory for places is grounded in their actual interaction with those locations, rather than on abstract schematic representations, as shown, for instance, by the fact that, when asked to reproduce on a small scale the locations and orientations of the holds of a wall they had just climbed, expert climbers focused on functional as well as structural aspects of the wall, correctly recalling more holds than novice climbers, who just took structural aspects of the wall into account. Memory is then influenced by environmental information. Environmental hints can nevertheless be suppressed in the case of difficult retrieval tasks, which need a conscious and effortful use of memory and the elimination of information which can interfere with the task. With regard to reasoning, there is some evidence that people employ several complementary strategies to manipulate the environment in order to improve their 20 Gibbs (2005: 131) also points out that the connection between mental imagery and kinesthetic action is supported by clinical evidence suggesting that congenitally blind people are quite capable to form imagistic representations. 21 This point is also in line with claims made by proponents of a particular strand of Embodied Cognition generally termed “distributed cognition” or “extended cognition” (see Clark 2008). 22 2. From Cognition to Grammar reasoning abilities. These strategies generally include embodied actions. This fact is apparent when we use a pen and a sheet of paper to do complex mathematical calculations: the physical action of writing down numbers and operators allows us to perform substantially better. For instance, if one tries to calculate a multiplication such as 1,549 * 368 in their head, without the help of any embodied action (including counting on their fingers), they will find the task very demanding, while using pen and paper the operation is quite straightforward (the result is 570,032). Similarly, calculating the amount of a set of coins is easier if we are allowed to touch and move the coins with our hands, and mental rotation of an object is straightforward if we can actually grasp the object with our hands and rotate it. We constantly combine our bodies' abilities and perceptual information from the environment in order to reduce cognitive effort in reasoning. Gibbs mentions an interesting piece of evidence from a teacher of mathematics and physics, who reports that his students learned ideas about space, time, and speed more easily by his having them moving their bodies in various ways. Apparently, doing so enabled his students to connect static representations of motion directly to their own embodied actions. In summary, we can follow Gibbs' remarks on the importance of embodied actions in grounding several aspects of imagery, memory, and reason: On-line embodied processes emphasize over sensorimotor activity to assist with cognitive tasks that interact with the intermediate world. Off-line embodiment occurs when sensorimotor processes run covertly to assist with the representation and manipulation of information in the temporary absence of task relevant input or output. Both of these aspects of embodiment work to create an embodied model of mind that is not internal to people's heads, but is distributed as a “cognitive web” across brains, bodies, and world. This distributed, embodied view of cognition offers a vision of human thought that is far less internally computational and far more bodily extended into the real world of action than is traditionally understood in cognitive science. (Gibbs 2005: 157) Communication is another topic on which Embodied Cognition strongly disagrees with Mainstream Cognitive Science. While traditional theories of language and communication maintain that embodied experience exerts little or no role in speaking, understanding, and communicating, there is a body of evidence which suggests that this is not in fact the case. Gibbs (2005: ch. 6) shows that studies on different topics in linguistics demonstrate that embodiment plays an important role in speaking, understanding, and also in nonverbal communication. Although offering a detailed account of all this evidence would go far beyond the scope of the present study, it may be relevant to mention some of these studies. A field of linguistic studies suggesting that embodiment shapes language and communication is semantic change. In a collection of studies on meaning extension in Indo-European languages, the American cognitive linguist Eve Sweetser argues that a considerable amount of polysemous words acquired abstract meanings via metaphorical extension from earlier physical meanings (Sweetser 1990; see also Bybee 2003). For instance, words related to the field of sensory experience generally get mapped onto ideas about intellect. A widely cited example of this tendency regards the “understanding is seeing” metaphor, which is a crosslinguistically recurrent pattern of semantic change: as an example, an expression like “open your eyes” is massively used to invite somebody to “realize, understand a situation”: an intellectual activity is conceptualized in terms of a physical experience. Similarly, Sweetser considers the diachronic development of modal verbs as driven by the “mind as a body” metaphor, 23 Grounding Meaning in Everyday Experience in the World which derives the epistemic value of an expression (which refers to the speakers' attitude toward a situation) from the deontic value (which regards the necessity or the possibility of actions carried out by responsible agents) via a metaphorical process shown by the tendency to bring linguistic expressions referring to the external world into the internal world considering these worlds as having a similar structure: duties imposed by the external world become mental premises guiding the opinions of the language user. As a result the simple sentence John may go can be understood in two different ways: (i) John is not prevent to go by any external force; (ii) John is not prevented by the speaker's mental premises22. Quite obviously, an embodied perspective on cognition also has some implications for our beliefs on language. When people speak, they often move their bodies, and their movements are frequently tightly connected to the message they want to convey. Indeed, a lot of information can be inferred by the addressee simply by taking notice of these gestural movements. Studies on the coordination of gestures and speech seem to suggest that these movements are grounded in the same underlying cognitive processes. First of all, psychological experiments suggest that gesture is not simply involved in lexical access, but it also helps the speaker to “package” information into verbalizable entities, i.e., it plays a role in speech production because it plays a role in the process of conceptualization, rather than simply facilitate lexical retrieval, as exemplified by the fact that young children use more (both redundant and nonredundant) gestures when asked to explain whether two items did or did not have the same quantity than when asked to describe how two items, one higher than the other, looked different (although prompting similar utterances, these two tasks demand different packaging of the information). Moreover, the hypothesis of a close link between gesture and speech is supported by a substantial body of neuropsychological research. Some studies point out that hand and arm movement are represented in areas of the brain which are closely related to those responsible for movement of the vocal tract; what is more, common brain mechanisms seem to exist for language and sequential motor functions, in an area whose stimulation proved to disrupt both oral facial movement and speech production. As Gibbs (2005: 167) states, these results seem to raise the possibility that “the tight temporal link between speech and gesture may be achieved because of spreading activation from the brain region responsible for speech production to that associated with hand and arm movements, and vice versa.” Other studies are reported to have found that high levels of electrical activity can be registered in motor areas of the brain when people are asked to read silently. In particular, it was found that verbs and names for tools exhibit the strongest brain activities in the motor cortex, perhaps because people encode the motoric function of these words as part of their semantic representations, activated when users attempt to retrieve them from memory. Furthermore, other brain studies report a certain level of overlap between brain areas activated during language and motor tasks, showing a significant activation of Broca's area during both these different kinds of task, especially those requiring hand action and, what is more, there seems to be some evidence of the activation of Broca's area when people just think about moving their hands (see e.g. Rizzolatti and Arbib 1998; Glenberg 2007; Rizzolatti and Craighero 22 More recently, an alternative interpretation of this kind of semantic change have been proposed by scholars working within the functional-typological as well as the cognitive semantic traditions (see Traugott and Dasher 2002; Hopper and Traugott 2003: ch. 4), according to whom this change is better explained in terms of metonymic inferences in context, rather than metaphorical processes. However, this different approach is not incompatible with an embodied perspective on language change. 24 2. From Cognition to Grammar 2007). This kind of evidence appears to contradict the widespread belief that Broca's area is only associated with aspects of language production . Evidently, the relationship between gesture movements and speech has also strong implications on discourse analysis, since embodied activity is revealing about speakers' goal in conversation (e.g., eye gaze often says a lot on the interlocutors' attitude toward what is being said and toward each other). The centrality of this activity in discourse is highlighted by the fact that a remarkable amount of body movements are conventionally associated to a determined meaning and/or intention, and thus becomes part of the background information which is shared by members of a certain sociocultural community23. Psycholinguistic studies show this “common ground” between speakers and listener, which is considered to be vital for successful communication, to derive from three main sources, clearly and syntethically outlined by Gibbs in the following chunk: The first source is “linguistic co-presence,” where the listener takes as common ground all of the conversation up to and including the utterance currently being interpreted. A second source for common ground is “physical co-presence,” where the listener takes as common ground what he or she and the speaker are currently experiencing in terms of their immediate physical environment, including the actions and positioning of their own bodies. The final source of evidence is community membership. This includes information that is universally known in a community...” (Gibbs 2005: 172) Embodiment in language is also recognized a prominent role in some cognitive linguistic theories of polysemy. The lexical organization of polysemous words is considered to be structured by systematic, recurrent cognitive principles arisen from bodily experience: metaphor, metonymy, and different kinds of image schemas. This hypothesis was empirically tested via psycholinguistic experiments with regard to the different senses of the verb stand, and Gibbs reports that the results of experiments suggest that actually people do make sense of different uses of stand on their basis of their implicit understanding of various image schemas which arise partly from embodied experience of standing. These image schemas are argued to underlie people's understanding of both literal and metaphorical senses of stand, a result which Gibbs explains by positing that people perceive different senses of stand as similar in meaning partially on the basis of the underlying image schema for each use of the word in context. Nevertheless, Gibbs points out that his experimental results (which hold for later studies as well) just hint that people tacitly recognize some connection between these schematic bodily experience and other aspects of linguistic meaning, but they neither rule out the possibility that other aspects of word meaning can also come into play in people's understanding of word meaning and their judgments of similarity of meaning for distinct senses of a word nor imply that people automatically recall some certain pattern of image schemas any time they encounter a particular use of a word. However, a considerable amount of work on image schemas supports the hypothesis that some aspects of meaning are grounded in speakers' bodily experience, and this idea is backed by other studies as well. Indeed, research shows that people convert words and sentences into a “flow of events” similar to normal perceptual experiences. As an example, it is often claimed that expressions in which events are represented in language reflect their chronological order (the “principle of iconicity”). Experiments 23 It is important to underline that this kind of nonverbal communication is subject to cultural variation and diachronic change just as language is. 25 Grounding Meaning in Everyday Experience in the World also suggest that people build dynamic representations as part of their understanding of word meaning. In other words, “people's construal of events, based on their embodied understanding, plays an important role in the processing and representation in memory of words in linguistic expressions” (Gibbs 2005: 179). Experiments also suggest that the interpretation of figurative language is influenced by embodied experience, with image schemas serving as source domains in a range of linguistic phenomena. Gibbs mentions a study on people's mental images for clusters of idioms with similar figurative meanings which shows that the general schemas underlying people's images are not simply representations of the idioms' figurative meanings, but “capture more specific aspects of the kinesthetic events with the images” (Gibbs 2005: 182). For instance, it is reported that participants imagined for anger idioms (e.g. “blow your stack” or “flip your lid”) some force causing a container to violently release pressure, despite the fact that there is nothing related to this image in the form of these idioms. Similar results are reported for studies on mental imagery of proverbs. Therefore, Gibbs proposes that specific embodied knowledge helps to structure people's metaphorical understanding of several concepts, restricting the range of images created for idioms and proverbs. Again, it seems that the metaphorical conceptualization of experiences really supply part of the motivation for people to have consistent mental images for idioms and proverbs with analogous figurative meaning. Moreover, an experiment on the possible influence of bodily action on people's speeded processing of simple metaphors shows consistent results: Wilson and Gibbs (2007) tested the hypothesis that if abstract concepts are indeed understood as items that can be acted upon by the body, then performing a related action should facilitate making a sensibleness judgment for a figurative phrase that mentions this action. The results of the experiments suggest that this is indeed the case: performing an action facilitates the understanding of a figurative phrase containing that action word, as it does for literal phrases, and these results were then confirmed by a comparison with the results of a control study. Then, it seems plausible that people understand a simple metaphor such as “grasp the concept” in terms of actually grasping an object, hinting that processing metaphorical meaning involves some imaginative understanding of the body's role in developing a framework for abstract concepts. There are several more research areas in which there is evidence of the fact that the dimension of human embodiment influences aspects of language and communication, many of which involve metaphorical thought. One of these topics is for instance the understanding of time expressions, which in English and various other languages often exploit two different metaphors involving spatial motion (see Lakoff and Johnson 1980, 1999): an “ego-moving” metaphor (the observer progresses in the time-line toward the future in front of them) and a “time-moving” metaphor (the future moves toward the observer from behind), corresponding to two distinct conceptual schemes (whose psychological reality is suggested by several experiments) for the sequencing of events in time. There is also evidence that native speakers of languages which differ in the metaphors they use to denote abstract concepts have different mental representations of such concepts: Boroditsky's (2001) influential experimental study (which was followed by a series of analogous experiments obtaining similar results) shows that not only do English and Mandarin Chinese speakers talk about time differently, but their abstract thought about time is differently shaped by distinct embodied metaphorical concepts24. 24 Converging evidence on the role of language in the way we conceptualize the world is also available from experiments carried out in the field of artificial intelligence (see De Beule and De Vylder 2005). 26 2. From Cognition to Grammar Moreover, experiments in which people in different settings were asked questions about time support the idea that people's thoughts about temporal events are influenced by their thoughts on spatial motion rather than the physical experience itself (see Boroditsky and Ramscar 2002; Matlock et al. 2003)25. Further support to the view that some aspects of language are shaped by bodily action is also provided from some research on American Sign Language, which shows that while spoken and signed language exploit structurally different conceptual metaphors, they share many of the same schematic mappings betwen embodied experiences and actions and more abstract conceptual domains, as well as by research on text understanding, which suggests that “people create meaningful construals by simulating how the objects and actions depicted in language relate to embodied possibilities. Thus, people use their embodied experiences to “'soft-assemble' meaning, rather than merely activate pre-existing abstract conceptual representations” (Gibbs 2005: 200-201). This is just a very brief summary of (a fraction of) the evidence provided by studies in some areas of language suggesting that embodiment influences language and communication, but it should be sufficient to provide the reader with an idea of the fact that many aspects of language and communication arise from and are guided by bodily experience. From the perspective of Embodied Cognition we are considering, infants' early kinesthetic action is considered to have also an important function in cognitive development, and it is argued to be instrumental for the development of physical reasoning. For instance, it is argued in some studies that newborn babies' habit of bringing the object they hold in their hand up to their mouths for oral contact enable them to learn important properties of objects (e.g. boundedness). Moreover, other studies contradict the belief that that infants' hand and arm movement are random, arguing that thiese motor activities are important for adaptation to the (both physical and social) environment. Embodied experience is thus considered crucial to children's development of object exploration skills. Supporters of Embodied Cognition emphasize that infants do not build their early object concepts on the basis of visual experience only, but rather experiencing how objects change in different perspectives through their own (both self-propelled and caused) movements. Gibbs suggest that the movement of objects may resemble certain aspects of infants' own kinesthetic actions, which can be self-instigated or caused. Since very young infants are sensitive to the difference between self-motion and caused-motion, they seem to understand difference in the movement of animate and inanimate objects. Children may use perceived differences in motion patterns to ground their early comprehension of the distinction between animate and inanimate entities, though different explanations for this phenomenon are also viable. Moreover, studies in developmental psychology show that young children recognize abstract similarities between different sensory experiences. Infants seem to able to make cross-sensory connections between visual and tactile information, as well as between auditory and visual information. 2-5-month-old infants are able to perceive coherent unitary multimodal events (e.g. the relationship between a person's face and voice) and 25 Gibbs (2005: 189) suggests that “People's understanding of time is not necessarily based on online sensorimotor activity, but rather on people's representations of and thoughts about their past and present spatial experiences”. This viewpoint is representative of the idea, briefly mentioned above, that awarding our sensorimotor system and our interactions with the environment a prominent role in shaping human cognition should not lead us to deny the role of abstract representations. 27 Grounding Meaning in Everyday Experience in the World by 5-7 months they can match their own body motions with a visual display of the motion. The discovery of intermodal relations is not a case of association of two experiences which occur simultaneously: experimental results show that young children are capable of acquiring abstract relations between events in different sensory modalities. Gibbs mentions three basic principles of intermodal perception: (i) global, abstract intermodal relations (e.g. shared synchrony between sight and sound) are detected earlier than more specific, nested relations (visual and auditory details of an event); (ii) amodal relations (e.g. the synchrony between a person's voice and their mouth movements) are perceived earlier than arbitrary relations (e.g. the specific sound of the voice); (iii) the detection of amodal relations facilitates perceptual learning about arbitrary relations (e.g. the synchrony between a person's face and voice allows an infant to associate the unique sounds of the voice with the relevant person). Experimental work on cross-sensory perception demonstrates that young children can make cross-sensory connections, which enable them to grasp important aspects of objects and events in the world. Support for the role of embodiment in cognitive development also derives from studies on imitation, which show that very young children are endowed with a surprizing ability to match visual with proprioceptive actions, a finding which may be explained positing that the infant is able to represent visual and proprioceptive information on a form common to both modalities: the infant may compare sensory information from their bodily movements with a sort of “supramodal” representation of the visually perceived gesture and match them. This idea is consistent with the possibility that children's early sensorimotor actions provide part of the basis for an embryonic body schema. Gibbs also mentions an amount of research on another type of imitation, namely older infants' analogical reenactment of previously seen events, backing the assertion (which can be led back to the Swiss psychologist and pedagogist Jean Piaget) that sensorimotor behavior becomes representational via a process of interiorization. In this view, overt behavior is seen as working as the model for understanding real-world events through a process of motor analogy. Mental imagery is then considered as arising from kinesthetic imitations as a result of progressive interiorization. Finally, embodiment seems to also have a place in the acquisition of a theory of others' mind, an important event in child development which takes place when the baby is about four years old. Gibbs advocates for a theory according to which understanding others is an ability derived from young children's engagement in sensorimotor, perceptual, and emotional practices. As a matter of fact, infants analyze the bodies and the expressive movements of others in order to recognize their intention or to understand the meaning of some objects and events. Studies on 5-7 months old babies show that infants can detect relationships between visual and auditory information specifying emotional expressions on the basis of the movement of others. From this perspective, understanding others is present before the development of a proper theory of mind: the child understands the meaning of being an experiencing subject, the meaning of certain kinds of entitites (rather than others) being such subjects, and the fact that these entities are partially similar to and partially different from oneself. The construal of human acts as intentional and goal-directed may then arise from a “like me” analogy that represents the starting point for social cognition (cfr. Gallese et al 2004; Gallese 2008). Indeed, the acquisition of a theory of mind involves being part of a community, where people with different minds share common purposes 28 2. From Cognition to Grammar and understanding. Constructing a model of minds does not stand on individual cognitive processes, but is grounded “both in the conditions of experience in the special social and cultural world, and in the phenomenology of experiencing on the part of the child” (Gibbs 2005: 236). Experimental results indeed suggest that changes in the understanding in (one's own and others') mental states reflect the child's participations in communities where social interactions are directed toward specific goals. Embodiment is also critical to the understanding of emotional experience. Gibbs (2005: 244) argues that “the fundamental relation between embodied action and emotion is captured by the idea that to 'be moved' refers to feeling as if one is in a different position in regard to one's situation”. The word 'emotion' is derived from the Latin 'e' (out) and 'movere' (to move).” The felt dimensional space can be characterized in terms of an “affective space” we move through when experiencing distinct emotions26. Different emotions, therefore, imply different levels of closeness to others: for instance, love suggests proximity, whereas loneliness makes us perceive our bodies as far from others. Each emotion is accompanied by a complex set of action tendencies, and cannot be seen as completely mental phenomena, since they rely on tactile feelings from the outside which become part of our inner emotional experiences and moving through affective space has a textured, felt dimension, similarly to “the way we can feel different textures of substances we touch with our skin” (Gibbs 2005: 246). It seems relevant, from an embodied perspective, to emphasize the fact that emotional experience is related to bodily changes. Some empirical research addressing people's description of bodily sensations arising when they are experiencing distinct emotions demonstrates that people tend to associate several emotions with specific bodily changes. At the same time, some emotions seem less related to specific body symptoms, suggesting that changes may not accompany all emotional experiences to the same degree. Other experimental studies show that body movements are, to a certain extent, specific for certain emotions, with the most significant variation regarding hand and arm movement. This experimental evidence suggests that people believe different emotions to be associated with different bodily symptoms; nevertheless, it is not clear if these beliefs match actual physiological responses to emotions. A proposal suggests that emotional experience is the result of the integration of bodily sensations with other associated mental representations, including previously acquired conceptions of how one customarily feels in a certain class of situations. This view seems to be corroborated by the fact that empirical studies reveal some cultural variation in bodily and mental emotion experience. For one thing, linguistic studies show that speakers from different cultures differ in regard to how much attention is paid to the body in emotion talk: for instance, it is reported that Russian speakers tend to describe emotions in terms of external behaviors, whereas Americans describe them as internal states, and that Chinese seem to locate experience in the body more frequently than North Americans. A culture's implicit folk model of self and of emotion will influence the form of emotional experiences (see also Maalej 2008). The relation between culture and body expression of emotion is nonetheless complex: first of all, experimental data show that the degree of similarity across cultures in the association of bodily experience and emotion is still greater than the degree of cultural variation. Furthermore, some emotions (e.g. joy) show more cultural variations than others (e.g. fear), and at the same time certain bodily sensations are more influenced by cultural variation than others. In general, the essence of any emotion is not completely captured by the way it is articulated in the body, but 26 This fact is exemplified by the vast amount of body-based metaphors we use to describe our emotions. 29 Grounding Meaning in Everyday Experience in the World affective processes may be better seen as emotional gestalts emerging from the interaction between bodily, environmental, and cognitive variables. Finally, we can deal with the perspective of Embodied Cognition on consciousness. While the nature of consciousness is traditionally considered disembodied and nonmaterial, theorists of Embodied Cognition take issue with this view: conscious ideas and images are always owned in a highly physical and body-based way (...) We own our conscious thoughts in the same way that we feel ownership of our bodies. Conscious experiences are not ethereal, but quite often have a raw feel to them that is testimony to their embodied nature. (...) consciousness is directly tied to action, both when these actions are physically performed and when they are just mentally entertained. (Gibbs 2005: 263-265) The approach we are considering posits that our conscious experience is grounded in our perceptually guided activity in the environment. Accordingly, we are most conscious of things that offer us opportunities for action. The affordances that arise from our bodily interactions with objects produces in us a usually inhibited inner movement that brings them into consciousness. It is the combination of this perceptual grounding and our subjective experience of bodies that supports the experience of consciousness. From this perspective, the notion of “enactment” is crucial. Enactment is a mental activity in which people simulate concrete action in the real world without overtly performing this action. This activity is not mental computation but it is essentially constituted by action (see §2.2.2). Consciousness enables people to find natural, efficient ways of performing actions in the world. It is not a neural state that acts in the preparation of action, but it emerges from action. The basic kinesthetic abilities by which any creature distinguishes parts of itself as an animate form constitute a “corporeal consciousness”, originally focused on the movement of one's own body. Then, as they move, creatures initiate movement by responding adequately to the surrounding ennvironmental context. Gibbs argues that there is an inherent kinesthetic specificity of animate form that provides for a wide range of movement possibilities, which constitute a creature's sense of agency27. With regard to conscious thoughts, they can vary depending on whether focal prominence is awarded to the self (yielding tactile sensation) or the world (yielding perceptual experience). These two attentional patterns show remarkable differences: (i) bodily sensations move with one's own movement, while perceived object features do not; (ii) bodily sensations are known only through proprioceptive awareness; (iii) bodily sensations have hedonic qualities of a kind different from externally perceived objects. These differences stress the fact that while our bodies are located in the world, the two are not homogeneous parts of the same experiental world. Thus, while processes which are crucial for consciousness are claimed to cut across brain-body-world divisions as part of an individual's embodied capabilities, organism and environment are still recognized as distinct entities. In summary, conscious experiences are seen as kinds of actions, even when these actions are not overtly manifested. These actions cause people to experience bodily sensations when they are engaged in conscious reflection. Although consciousness occurs at different levels of experience, each level is constituted by its own sensations of felt movement, either in terms of the body's direct interactions with the world or when one imagines themself engaging in past or future actions. 27 In this view, consciousness and intentionality are not human species-specific characteristics. 30 2. From Cognition to Grammar In the present subsection we have outlined the foundations of the embodied perspective on cognition adopted in this study, which has been argued to be most in line with the point of view exposed in Gibbs' (2005) piece of work, which served as the main reference point for drafting these pages. Nonetheless, the reader will have noticed that no general definition of embodiment has been provided, yet. It is no chance, indeed, that sometimes supporters of Embodied Cognition are charged with the accusation of pleading the case of embodiment without providing a definition of what embodiment is supposed to be (see e.g. Zlatev 2007a). Gibbs, too, does not supply a definition in the traditional sense, but in the concluding remarks of his book he formulates the “embodiment premise”, which in a few lines defines the basic assumptions of (at least his version of) Embodied Cognition and, at the same time, implies what the nature of embodiment is understood to be: People's subjective, felt experiences of their bodies in action provide part of the fundamental grounding for language and thought. Cognition is what occurs when the body engages the physical, cultural world and must be studied in terms of the dynamical interactions between people and the environment. Human language and thought emerge from recurring patterns of embodied activity that constrain ongoing intelligent behavior. We must not assume cognition to be purely internal, symbolic, computational, and disembodied, but seek out the gross and detailed ways that language and thought are inextricably shaped by embodied action. (Gibbs 2005: 276) Concluding the present subsection with the few lines quoted above from Gibbs has two main assets: on the one hand, the “embodiment premise” can be considered as a sketchy but forceful summary of the very foundations of this embodied approach to the study of human cognition; on the other hand, it also provides us with an effective verbalization of the core theoretical assumptions which represent the spinal cord (and, in a sense, the spirit) of the present study. 2.1.2. Cognitive Linguistics Contrary to what is observed for Embodied Cognition, which is still an extremely fragmented world, in the last two decades Cognitive Linguistics as a research effort has reached a much more unitary status, as testified by the lively activity of the several associations of scholars spread in the world, above all the International Cognitive Linguistics Association, whose journal Cognitive Linguistics is published regularly since 1990. This does not mean that Cognitive Linguistics can be seen as a properly unified theory, but at least it is easier to identify and define it as a field of research then it is for Embodied Cognition. Usually, there is a higher level of overlapping between the background of distinct cognitive linguists and it is easier for them to recognize each other as involved in (various projects of) the same scientific enterprise, than it is for the community of scholars involved in the much broader Embodied Cognition research program. As a result, while attempts to draw a picture of Embodied Cognition virtually always end up in the description of one or few specific approaches, a generical snapshot of Cognitive Linguistics is more easily obtained. Consequently, an extensive introductory literature on Cognitive Linguistics is nowadays available: the interested reader can find several handbooks (e.g. Croft and Cruse 2004; Evans and Green 2006; Ungerer and Schmid 2006), a few position papers (e.g. Mazzone 2004; Janda 2006; 31 Grounding Meaning in Everyday Experience in the World Geeraerts and Cuyckens 2007b), a glossary (Evans 2007), a comprehensive handbook including almost fifty chapters written by various influential cognitive linguists (Geeraerts and Cuyckens 2007a), along with many other resources (readers, journals, websites, etc.). The scope of the present section will be limited to three narrow purposes: summarizing the theoretical foundations of Cognitive Linguistics, outlining the four basic assumptions of cognitive semantics, and sketching the notions of image schema (already encountered in §2.2.1), which will turn out to be crucial in §3.2, and of frame, which will be occasionally mentioned throughout ch. 3, looking to Evans and Green (2006), Geeraerts and Cuyckens (2007b), and Evans (2007) for “guidance”. With regard to cognitive approaches to grammar, the reader is referred to §2.3.1 below. Cognitive Linguistics is a theoretical paradigm which collects a cluster of different, but related approaches to the scientific study of language in its cognitive function which originated in the late 70s and early 80s, stemming from the work of three pioneers, namely George Lakoff, Ronald W. Langacker, and Leonard Talmy. Scholars working within the Cognitive Linguistics framework hold the view of language as an instrument for organizing, processing, and conveying information. As a consequence, formal structures of language are seen as reflections of general conceptual organization, categorization principles, processing mechanisms, and experiential and environmental influences. As Geeraerts and Cuyckens (2007b) point out, cognitive linguists are prominently interested in topics such as the structural characteristics of natural language categorization, the functional principles of linguistic organization, the conceptual syntax-semantics interface, the experiential and pragmatic background of language-inuse, and the relationship between language and thought. Indeed, Cognitive Linguistics is based on the assumption that our interaction with the world is mediated by informational structure in the mind: language is thus seen as a structured repository of meaningful categories to help us to deal with new experiences and store information about old ones. Geeraerts and Cuyckens (2007b) assert that Cognitive Linguistics shows three foremost characteristics: • the primacy of semantics in linguistic analysis: meaning is the central linguistic aspect, since categorization is the main function of language; • the encyclopedic nature of linguistic meaning: language is a system for the categorization of world knowledge; • the perspectival nature of linguistic meaning: the categorization function of the language imposes a structure on the world rather than mirroring objective reality. The two scholars also observe that “language is a way of organizing knowledge that reflects the needs, interests, and experiences of individuals and cultures” (Geeraerts and Cuyckens (2007b: 5), underlining the importance of embodied (both individual and collective) experience in human reason (see §2.1.1). Evans and Green (2006), following some work of Lakoff's, emphasize the fact that there are two widely shared commitments which underlie the Cognitive Linguistics paradigm: • the Generalization Commitment: a commitment to the characterization of general principles that are responsible for all aspects of human language; • the Cognitive Commitment: a commitment to providing a characterization of general principles for language that accords with what is known about the mind and brain from other disciplines. 32 2. From Cognition to Grammar The first commitment refers to the fact that cognitive linguists assume that there are common structuring principles that hold across different aspects of language, and that an important function of linguistics is to identify them. There has always been a methodological tradition in modern linguistics to focus on distinct aspects of language one at a time. This policy is especially pursued by proponents of formal approaches to the study of language, which posit mechanical devices operating on theoretical primitives in order to produce the complete set of linguistic possibilities in a given language. Within this academic tradition, it is often argued that different areas of language deal with different kinds of structuring principles operating over different kinds of primitives. The most influential of these formal approaches, namely Chomsky's Generative Grammar, takes this conviction one step further, arguing for a modular view of language: according to generative linguists, phonology, morphology, syntax, semantics etc. are different modules concerned with different objectives28. While cognitive linguists admit that it is often useful to treat different areas of language as notionally distinct, they reject the idea of the existence of different encapsulated language modules; instead, in line with the Generalization Commitment, they believe that “apparently distinct language components can be seen to share fundamental organisational features.” (Evans and Green 2006: 28). Evans and Green argue that there are at least three areas in which this fact emerges rather clearly: categorization, polysemy, and metaphor. With regard to the former, studies in cognitive psychology demonstrate that categories frequently seem to be fuzzy rather than clear-cut, with some members of a category appearing as more central and others more peripheral. Moreover, the degree of centrality often depends on the way we interact with a particular category at any given time (see §2.1.1 above). Categories exhibiting degrees of centrality, with members being more or less similar to other members of the category rather than sharing a single defining trait, are said to exhibit family resemblance. Fuzziness and family resemblance are not limited to physical objects, but they also apply to more abstract entities such as linguistic categories, which – regardless of the specific aspect of language they are related to – all appear to exhibit these phenomena. While advocates of the formal linguistics tradition tend to see categories as exhibiting uniform behavior, members of linguistic categories often show quite divergent behavior. With regard to morphology, Evans and Green show that, for instance, the Italian diminutive does not have a single meaning associated with it, but instead represents a category of meanings which behave in a variety of different, though related ways. While a common meaning associated with this form is “physically small” (e.g. paese “village” → paesino “small village”), this is not the only possibility: Evans and Green (2006: 3031) point out that the diminutive can also be used to convey affection (e.g. mamma “mum” → mammina “mummy”). When it is applied to abstract nouns, the diminutive takes a meaning of short temporal duration, reduced strength or reduced scale (pioggia “rain” → pioggerella “drizzle”). When applied to adjectives or adverbs, it works as a sort of downtoner (bene “well” → benino “quite well”), whereas when added to verbs, it signals a process of intermittent or poor quality (dormire “sleep” → dormicchiare “snooze”). The diminutive in Italian does not have a single meaning, but represents a category of different, but related meanings: the category shares a related form and a 28 It is possible to notice the analogy between this modularity of language and the modularity of mind mentioned in §2.1.1: indeed, Generative Grammar and Mainstream Cognitive Science are deeply interrelated, as we shall be arguing below. 33 Grounding Meaning in Everyday Experience in the World related set of meanings, i.e. a reduction in size, quantity, or quality. Thus, they show family resemblance. A similar phenomenon can be observed in English syntax when dealing with parts of speech, traditionally identified according to the morphological and distributional behavior of words: an examination of the grammatical behavior of nouns and verbs shows that categories such as “noun” and “verb” are not homogeneous, but instead that certain nouns and verbs are “nounier” or “verbier” than others. Therefore, parts of speech are fuzzy categories. For instance, a transitive verb is a verb that can take an object. However, while transitive verbs can often be nominalized (i.e. made into agentive nouns), some cannot. Compare the following two sentences: (1) (a) (b) John imports rugs → John is an importer of rugs John knew the fact → *John was the knower of the fact Moreover, verbs can often be substituted by the “be-V-able” construction, this is not always the case. See the two sentences in (2) below: (2) (a) (b) His handwriting can be read → His handwriting is readable The lighthouse can be spotted → *The lighthouse is spottable Finally, while the majority of transitive verbs undergo passivization, some do not: (3) (a) (b) John kicked the ball → The ball was kicked by John. John owes two pounds → *Two pounds are owed by John. Nevertheless, these verbs do share some common “verbish” behavior (e.g. they can all take the third person present tense suffix -s). Therefore, while certain verbs fail to display some “typical” verb behavior, they are still part of the category “verb”. On the other hand, this variation shows that there is no fixed set of criteria serving to define what it means to be a verb. Consequently, the linguistic category “verb” contains members that are broadly similar but exhibit variable behavior. These two examples (drawn from Evans and Green: ch. 2) from morphology and syntax suggest that the nature of the linguistic categories found in these areas can be described in similar terms: in terms of categorization, we can generalize across what are often considered completely distinct kinds of phenomena (this observation is also shown to hold for phonology in Evans and Green 2006: 34-35). Polysemy is the phenomenon where a single linguistic unit displays several distinct but related meanings. It is to be distinguished from homonymy, where two words have the same form but have different and unrelated meaning. While the study of polysemy is traditionally restricted to the area of lexical semantics, according to cognitive linguists polysemy is a fundamental feature of human language; therefore, it is considered as a key to generalization across a range of distinct phenomena, revealing commonalities between different levels of linguistic organization. An example of polysemy in morphology can be found in some particular bound morphemes. Evans and Green (2006: 36-37) illustrate the case of the English agentive -er suffix in a group of nouns such as those listed in (4) below: (4) (a) (b) (c) teacher villager toaster 34 2. From Cognition to Grammar (d) best-seller In each of these words, the suffix adds a slightly different meaning to the “matrix” word. In the first case it denotes somebody who performs the action expressed by the verb teach; in the second, it describes somebody who lives in a particular village; in the third example, it relates to an artefact able to perform the action expressed by the verb toast; in the fourth case, -er denotes a quality associated to an artefact. Each of these usages is distinct; nevertheless, they are related because they share a common property: they describe a functional ability (to teach, to toast, to sell well) or attribute (to live in a specific place). In syntax, a similar situation holds for several constructions, as extensively discussed in Goldberg (1995). For instance, the ditransitive constructions, which shows the syntactic structure illustrated in (5) below, is associated to a range of abstract meanings which Goldberg describes as reported in (6) (these examples are slightly adapted from Evans and Green 2006: 37-38): (5) (6) Subj V Obj1rec Obj2 (a) SENSE INSTANTIATED BY: 1: AGENT successfully causes RECIPIENT to receive PATIENT verbs that inherently signify acts of giving (e.g. give, pass, hand, serve, feed) e.g. [Subj Mary] [V gave] [Obj1 John] [Obj2 a cake] (b) 2: conditions of satisfaction imply that AGENT causes RECIPIENT to receive PATIENT verbs of giving with associated satisfaction conditions (e.g. guarantee, promise, owe) e.g. Mary promised John the cake SENSE INSTANTIATED BY: (c) 3: AGENT causes RECIPIENT not to receive PATIENT verbs of refusal (e.g. refuse, deny) e.g. Mary refused John the cake SENSE INSTANTIATED BY: (d) 4: AGENT acts to cause RECIPIENT to receive PATIENT at some future point in time verbs of future transfer (e.g. leave, bequeath, allocate, reserve, grant) e.g. Mary left John the cake SENSE INSTANTIATED BY: (e) 5: AGENT enables RECIPIENT to receive PATIENT verbs of permission (e.g. permit, allow) e.g. Mary permitted John the cake SENSE INSTANTIATED BY: SENSE (f) INSTANTIATED BY: 6: AGENT intends to cause RECIPIENT to receive PATIENT verbs involved in scenes of creation (e.g. bake, make, build, cook, sew, knit) e.g. Mary baked John the cake Cognitive linguists assert that polysemy is a phenomenon common to distinct areas of language (Evans and Green 2006 also ofter an instance of polysemy with regard to the lexicon). Therefore, both fuzzy categories and polysemy are considered as unifying characteristics of human language, enabling generalizations. Several cognitive linguists see metaphor as a central feature of human language. Metaphor, the phenomenon whereby a conceptual domain is systematically structured in terms of another, is considered to play an important role in meaning extension (Lakoff 35 Grounding Meaning in Everyday Experience in the World and Johnson 1980, 1999; Sweetser 1990; see also §2.1.1 above), which can be identified across different linguistic phenomena, supplying further evidence in favor of generalization across distinct areas of language. Lakoff (1987) argued that metaphor has an important role in the extension of lexical meaning, arguing for instance that the “control” meaning of the English preposition over (e.g. she has a strange power over me) derives from the “above” meaning by metaphorical extension (e.g. the picture is over the sofa). This result is achieved by the application of the metaphor “control is up”, according to which control is understood in terms of greater elevation of the vertical axis, whereas lack of control is conceptualized in terms of reduced elevation. Compare the two examples in (7) below: (7) (a) (b) I'm on top of the situation He is under my control. Then, as Evans and Green assert: By virtue of the independent metaphor CONTROL IS UP, the lexical item over, which has an ABOVE meaning conventionally associated with it, can be understood metaphorically as indicating greater control. Through frequency of use the meaning of CONTROL becomes conventionally associated with over in such a way that over can be used in non-spatial contexts (...), where it acquires the CONTROL meaning (Evans and Green 2006: 39) The ditransitive construction, mentioned above to illustrate polysemy in syntax, can also be exploited as a vehicle to observe meaning extension in syntactic constructions. Goldberg in fact observes that the distransitive construction requires a volitional agent in subject position, for its meaning is one of intentional transfer. However, ditransitive sentences in which the subject is not a volitional agent can also be found, as in the two clauses below: (8) (9) The rain gave us some time. The missed ball handed him the victory. According to Goldberg, examples along these lines are extensions of the ditransitive construction motivated by the metaphor “causal events are physical transfers”. Since we typically understand abstract causes in terms of physical transfer, we recognize the convention within our language system of understanding causal events (metaphorically) in terms of physical transfer. Consequently, Goldberg asserts that the “causal events are physical transfers” metaphor licenses the extension of the ditransitive so that it can occur with non-volitional agents. Though above we have provided a rather minimal description, it should be sufficient to illustrate the Generalization Commitment adopted by cognitive linguists, according to which different areas of human language share certain basic organizing principles. As will be seen in §2.3.1, this approach achieved a remarkable success in providing a unified theory of grammar and lexicon, according to which lexical and syntactic items are units of a single repository of (both lexical and syntactic) constructions, often labeled “constructicon”29. The Cognitive Commitment represents the view that language and linguistic 29 The label “constructicon” was introduced by Goldberg (2003). 36 2. From Cognition to Grammar organization should reflect general cognitive principles, rather than cognitive principles specific to language. The modular theory of mind (already mentioned in §2.1.1 above) holds that the human mind is organized into distinct modules of knowledge, one of which is language, that serve to “digest” raw sensory input in such a way that it can be processed by a central system. Cognitive Linguistics strongly rejects the claim for the existence of a specific language module. According to cognitive linguists, linguistic structure and organizations are not markedly distinct from other aspects of cognition. Evans and Green (2006: §2.1.2) outline three lines of evidence underlying the view that linguistic organization reflects more general cognitive function: attention, categorization, and metaphor. Language provides way of directing attention to certain particular aspects of the scene being linguistically encoded. This ability is generally labeled profiling by cognitive linguists. Language exhibits profiling in the range of grammatical constructions it has at its disposal, each of which serves to profile different aspects of a given scene (see §2.3.2 below). For instance, given a scene in which two boys beat a man making him become unconscious, distinct aspects of the scene can be linguistically profiled. Consider the following four sentences: (10) (a) (b) (c) (d) Two boys beat the man. The man was beaten. The man was driven unconscious. The man is unconscious. The first example is an active sentence, in which the boys are the agent and the man is the patient. This sentence profiles the whole process of energy-transfer from the agents to the patient, i.e. the fact that two boys were acting upon the man, hitting his body. The second sentence is a passive. Here, the energy-transfer and the patient are given prominence, while the agent is not mentioned. However, though the agent is not profile, it must be understood as part of the background, since an energy-transfer event requires an agent to get the transfer off. The third sentence profiles the change of state of the man, the fact that he became unconscious as a consequence of somebody else's action. This is achieved using a subject-auxiliary-verb-complement construction. Here, the complement unconscious completes the meaning of the expression was driven, giving prominence to the internal change of state of the man. Finally, the fourth sentence uses a subject-copula-complement construction, which is specialized for encoding a particular state. As a matter of fact, the last example describes the state which represents the result of the energy-transfer process. Each of the grammatical constructions profile particular aspects of a process. Linguistic structure reflects our ability to attend to distinct aspects of a scene, showing that linguistic organization reflects a more general cognitive ability, namely attention30. We have seen that categorization is not an “all-or-nothing” affair, but instead categories seem to be fuzzy, containing members which are more or less representative of the category and, therefore, categories include a set of members related by family resemblance rather than a single criterial feature, or a limited set of necessary and sufficient conditions possessed by every member of the category. It was also mentioned that linguistic categories are also rather fuzzy, since their members often display quite 30 Evans and Green (2006: 43) highlight the fact that constructions of this kind are not restricted to the encoding of actions, but can also applied in cases where no action is involved, for instance with stative verbs. 37 Grounding Meaning in Everyday Experience in the World distinct behavior, together with important similarities. The same principles that hold for categorization in general seem to hold also for specifically linguistic categorization (see Taylor 2003). Finally, cognitive linguists consider metaphor as a conceptual rather than merely linguistic phenomenon (see §2.1.1 above). Evans and Green mentions the fact that institutions, governments, and businesses are often conceptualized in terms of a hierarchy, arguing that diagrams of such institutions place the person with the higher rank at the top, while the person with the lowest rank is placed at the bottom. Hierarchies are thus represented nonlinguistically in terms of the conceptual metaphor “control is up” mentioned above. Besides the Generalization Commitment and the Cognitive Commitment, Evans and Green list embodiment (§2.1.1 above) as the third central notion in Cognitive Linguistics. At the same time, in the previous subsection we saw that theorists of Embodied Cognition often look at the research carried out by cognitive linguists as a source of evidence for embodiment. Indeed, the Cognitive Linguistics enterprise and the Embodied Cognition research program can (at least to a certain extent) be considered as mutually informative. Indeed, it appears to be no accident that they had a quite parallel development. Rather than repeating or summarizing what already said in §2.1.1, where we outlined Embodied Cognition following Gibbs (2005), it is convenient here to say something more on the (rather tight, in our view) relationship between Embodied Cognition and Cognitive Linguistics, comparing it with the equally strong link between Mainstream Cognitive Science and Generative Grammar. In the previous subsection, we mentioned the fact that Embodied Cognition developed in the last decades as an alternative to the dominant paradigm in cognitive science, which we labeled “Mainstream Cognitive Science”. We pointed out that scholars working within that framework view the human mind as a sort of computational system which manipulates abstract, amodal symbols in order to carry out its complex cognitive processes, which are considered as functional states resulting from the manipulation of symbols following an effective method for solving problems expressed as finite sequences of steps, independent of the physical system in which they are implemented. We also mentioned in passing the fact that one of the key figures of Mainstream Cognitive Science is to be found in the American linguist Noam Chomsky. Chomsky in the late 50s began to develop Generative Grammar, a formal theory of language whose philosophical roots can be traced back to Cartesian rationalism and whose basic assumptions go hand in hand with those of Mainstream Cognitive Science. It seems relevant to summarize the basic theoretical observations which led Chomsky to develop such a model of grammar. Chomsky was strongly critical toward the dominant paradigms in both linguistics and psychology of language, i.e. structuralism and behaviorism, respectively. He was especially concerned with the fact that neither the analysis of the structural properties of languages and the relationships between its components nor a stimulus-response theory are adequate to explain how language works. In particular, he focused on the problem of how speakers of a language can understand and produce a vast amount of sentences they have never heared before. Moreover, native speakers of a language can also distinguish acceptable from unacceptable expressions in their language without negative evidence. The key problem can be summarized by the following question: “how can speakers know what expressions are unacceptable, since they do not encounter them during learning and then their unacceptability is not explicitly stated?”. This absence of 38 2. From Cognition to Grammar negative evidence represents the basis of the poverty of the stimulus argument, according to which natural language grammar is unlearnable given the relatively limited data available to children learning a language. The reader should remind that Mainstream Cognitive Science posits that the mind is a computational system independent of the body and complex cognitive processes result from the manipulation of abstract symbols. Furthermore, it was said in the previous subsection that several theorists of Mainstream Cognitive Science also adopt the idea that the mind comprises separate, independent modules each developed for a different functional purpose. As a result, thus far we have the following picture: the mind is independent of the body, and language is an autonomous module in the mind. Since the mind can be studied independently of the body, it follows that the properties of the human body and experience are irrelevant to cognition and language. Language, then, can be simply studied as a computational system. Consequently, it is straightforward to conclude that social aspects of language and cognition are uninteresting and can be conveniently ignored. As Geeraerts and Cuyckens point out, if the source of language is not social, it can only be an innate and universal genetic endowment. Because semantics and the lexicon “constitute the variable, contextual, cultural aspects of language par excellence” (Geeraerts and Cuyckens 2007b: 12), and social interaction is mediated through the meaning of linguistic expressions, it is implausible that the universal aspects of language can be found in meaning; as a consequence, semantics and the lexicon are peripheral. Then, the core of language must be composed of a syntactic, formal rule system. This distinction between core and periphery suggests the fact that the language module mentioned above comprises separated submodules, with the syntactic submodule being the one linguists should be concerned with. A logical conclusion from these premises is that actual usage of language is uninteresting, since it just consists of the application of the rule system, leading to the distinction between competence (the internal grammar of the language user), which is the object of the studies carried out by generative grammarians, and performance (the individual activity of combining the elements that are present in the code), seen as a mere epiphomenon. We saw in §2.1.1 that in recent decades Embodied Cognition arose as an alternative to Mainstream Cognitive Science, rejecting the view of the mind's complex cognitive processes as the result of the manipulation of amodal symbols performed by a computational system. We mentioned the fact that scholars working within the Embodied Cognition research program claim that our physical body and our interaction with the environment are instead the primary source of human cognitive processes, implying that the mind is shaped by the interaction of brain, body, and the external world. Roughly in the same period, Cognitive Linguistics arose as a reaction against Generative Grammar, which was the dominating paradigm in the wide range of linguistic theory. The two commitments underlying the Cognitive Linguistics framework sketched above can be seen as tightly related to the Embodied Cognition view of the mind as the result of an interplay between the whole organism and the environment. Indeed, this view does not require for the mind to be composed of a series of encapsulated, domain-specific modules. On the contrary, a more “distributed” conception of cognition - which does not confine the mind to the computational activity of some neural circuitries, but claims for the involvement of the organism as a whole and its experiences in the external world - makes the case for the supposition of domain-general cognitive principles, rejecting the theory of the modularity of mind. From this perspective, any claim for distinct “submodules” for the various “areas” of 39 Grounding Meaning in Everyday Experience in the World language makes little sense, since no such clear-cut distinction between phonology, morphology, syntax, semantics, etc. is supposed to exist31. The interaction of organism and (physical as well as sociocultural) environment in shaping cognition highlights the fact that everyday experience in the world is basic for cognition, contrary to the claims of mainstream cognitive scientists: there is simply no human mind without the interplay of a human brain, a human body, and the environment. From this point of view, social interaction is dramatically important, since most human experiences involve, to great extent, our being embedded in a definite sociocultural context. It follows that social aspects are of paramount importance for language, since a certain language represents a common code shared by a community. As a consequence, meaning plays a crucial role in language: as Geeraerts and Cuyckens (2007b: 12) assert, “... social interaction, the exchange of ideas, and changing conceptions of the world are primarily mediated through the meaning of linguistic expressions...”. Equally, actual usage of language is all-important in Cognitive Linguistics, since language units are supposed to arise from language use in a bottomup manner, rather than being created in a top-down fashion by means of the application of a series of formal rules. It follows that there is no principled distinction between “competence” (language knowledge) and “performance” (language use) in Cognitive Linguistics: as Evans (2007: 217) points out, “knowledge of language is knowledge of how language is used”. Furthermore, cognitive linguists do not distinguish a “core” of language from a “periphery”: all linguistic expressions are symbolic units which deserve to be taken into consideration and need to be accounted for32. Accordingly, while generative linguists argue for the “autonomy of syntax”, cognitive linguists do not draw any sharp boundary between syntax and the lexicon, which are seen as the two poles of a continuum: lexical units and syntactic units make part of the “constructicon”, a unified repository of form-meaning pairs. Since grammatical units, as much as lexical units, are form-meaning mappings, it also follows that grammar is inherently meaningful rather than purely formal. On the basis of the observations made in the few lines above, a further comment seems worth to be made. It regards the fact that the Cognitive Commitment, together with Embodied Cognition (and, though only indirectly, the Generalization Commitment), may be said to constitute the basis for the opulence of the substrate argument, which is often claimed by cognitive linguists to be an adequate reply to Chomsky's “poverty of the stimulus” argument. Indeed, evidence for the “opulence of the substrate” is often considered to undermine the Chomskyan claim for linguistic nativism. As a matter of fact, if language reflects domain-general cognitive principles acquired through embodied experience, then there is no need for there to be any genetic endowment specifically tailored for language learning: language can be better viewed as an emergent ability, learned through experience like other cognitive capabilities33. A synthetic characterization of the “opulence of the substrate” argument is provided by Feldman (2006) in the following chunk: The child comes to language learning with a rich base of conceptual and embodied experience as 31 For recent empirical evidence against a modular view of language, see Hilpert (2008). 32 Indeed, a substantial amount of work have been carried out in Cognitive Linguistics on conventional expressions (idioms, figures of speech), considered as important to language as productive patterns (see Lakoff 1987; Goldberg 1995). 33 A concise survey of the arguments against the “poverty of the stimulus” argument and linguistic nativism is offered in Lombardi Vallauri (2009). 40 2. From Cognition to Grammar well as a supportive social environment. Words and rules that describe this experience can be learned without formal training, although not without years of focused effort on the child's part. As the child expands his or her scope and deals with abstract concepts and others that are not directly experienced, language coevolves, always maintaining the grounding in direct experience. (Feldman 2006: 317-318) In summary, though both approaches look at language as a mental phenomenon and attempt to build psychologically and neurophysiologically plausible models of language, Generative Grammar and Cognitive Linguistics are two very different frameworks, whose theoretical origins and historical developments are interwoven with those of two competing theoretical paradigms in cognitive science, namely Mainstream Cognitive Science and Embodied Cognition, respectively. Actually, as Evans and Green (2006: 752) point out: “While the formal model views language as an innate, encapsulated, and computational system, the cognitive model views language as an emergent system, inextricably linked with general processes of communication and conceptualisation, with meaning at its core.” Geeraerts and Cuyckens (2007b) also highlight the fact that these two paradigms hold different views of the epistemological role of natural language in the relationship between subject and object: while cognitive linguists see language as the intermediate link between subject and object, generative linguists take language as the object itself of the relationship. These different beliefs have remarkable consequences on the interests of the scholars working within the two paradigms: cognitive linguists focus on our knowledge of the world and how language contributes to it, whereas generative linguists are concerned with our knowledge of a language and attempt to explain how it can be acquired given a cognitive theory of learning. As a consequence, it seems clear that the basic properties of these two approaches to the study of language can hardly be seen as independent from the assumptions of the two corresponding enterprises in cognitive science. The Cognitive Linguistics enterprise is often seen as divided into two branches: cognitive semantics and cognitive approaches to grammar (see e.g. Evans and Green 2006; Evans 2007). The former is considered, in a sense, to be foundational for the second: since grammar is inherently meaningful, a theory of meaning must be delineated before it is possible to develop a theory of grammar. Cognitive semantics is concerned with the study of the relationship between experience, the conceptual system, and the semantic structures encoded by language. Therefore, cognitive semanticists focus on knowledge representation and meaning construction. Language is then employed as the tool through which cognitive phenomena can be investigated. With regard to cognitive approaches to grammar, Evans and Green draw a distinction between the work of Ronald W. Langacker, whose Cognitive Grammar focuses on the study of the cognitive principles which give rise to linguistic organization, and the work of researchers who attempt to provide an inventory of the units of language, who developed a cluster of theories collectively referred to as “construction grammars”, or “constructionist approaches to grammar” (as will be seen in §.2.3, we shall adopt the second label)34. Cognitive approaches to grammar encompass the entire inventory of linguistic units defined as form-meaning pairings; therefore, meaning and grammar are 34 The distinction between Cognitive Grammar and constructionist approaches to grammar is not clearcut: some authors consider Cognitive Grammar as a particular constructionist approach to grammar (see Croft and Cruse 2004: ch. 10; Croft 2007). On the other hand, when constructionist approaches to grammar began to spread, Cognitive Grammar was already well-established. This is not a major matter of controversy, however. 41 Grounding Meaning in Everyday Experience in the World seen as two sides of the same coin: the study of grammar deals with the units of language, while the study of semantics focuses on the relationship between these linguistic units and the conceptual system, which is linked to embodied experience. Cognitive semantics, unlike traditional truth-conditional semantics, considers linguistic meaning to be a manifestation of the nature and organization of mental representation in all its richness and diversity, and this is what makes it a distinctive approach to linguistic meaning (see Talmy 2000). Evans and Green (2006: 157) list four assumptions which play a crucial role in cognitive semantics, reported below: (11) (a) (b) (c) (d) Conceptual structure is embodied. Semantic structure is conceptual structure. Meaning representation is encyclopedic. Meaning construction is conceptualization. The first assumption refers to the importance in cognitive semantics of the relationship between conceptual structure and the external world of sensory experience: scholars investigating cognitive semantics aim to build a theory of conceptual structure which is consistent with the ways in which we experience the world. Once more, Embodied Cognition intersects Cognitive Linguistics, since cognitive semanticists believe that “the nature of conceptual organization arises from bodily experience, so part of what makes conceptual structure meaningful is the bodily experience with which it is associated” (Evans and Green 2007: 157). To illustrate this point, Evans and Green suggest that we imagine a man in a locked room. A room has the structural properties associated with a bounded region: an interior, a boundary, and an exterior. As a consequence, the bounded landmark has the property of containment: the man cannot leave the room. This instance of containment is partly a consequence of the properties of the bounded landmark and partly a consequence of the properties of the human body. In other words, containment is a meaningful consequence of a particular type of physical relationship experienced in interaction with the external world. The concept associated with containment is an instance of an image schema, i.e. a concept whereby (as seen in §2.1.1 above) bodily experience gives rise to meaningful concepts. While the concept “container” is grounded in the experience of interacting with bounded landmarks, image-schematic conceptual structures can give rise to abstract kinds of meaning. For instance, several metaphorical expressions are licensed by the metaphorical projection of the “container” image schema onto the abstract conceptual domains of “states”, which include “love” (e.g. He is in love), “trouble” (e.g. We are out of trouble), and “health” (e.g. He is coming out of the coma). Then, meaningful structure from bodily experience gives rise to concrete concepts like the “container” image schema, which in turn serves to structure more abstract domains like “states”. This is the embodiment of conceptual structure. The second assumption asserts that language refers to concepts in the mind of the speaker, rather than to objects in the external world. Semantic structure, the meaning conventionally associated with linguistic units, can thus be equated with conceptual structure35. This principle is vital for Cognitive Linguistics, since semantic structure relates to all linguistic units, not just words. A linguistic unit may be a word (e.g. cat), a bound morpheme (e.g. -er) or a larger conventional pattern, like the structure of an 35 It is relevant to point out that this does not mean that the two are identical; on the contrary, “we have many more thoughts, ideas and feelings than we can conventionally encode in language” (Evans and Green 2006: 159). 42 2. From Cognition to Grammar active sentence or a passive sentence. Cognitive linguists claim that these patterns are inherently meaningful, though schematic: an active transitive sentence (e.g. Mary baked a cake) differs from the corresponding passive transitive sentence (e.g. A cake was baked by Mary) in terms of focal prominence, for in the former we focus on the causer participant, whereas in the latter we focus on the undergoer. Cognitive semantics aims to provide a unified account of lexical and grammatical organization. Proponents of cognitive semantics claim that we have concepts either because they are useful ways of understanding the world or because they are inevitable ways of understanding the world, given our cognitive architecture and physiology. The third assumption states that words serve as points of access to vast repositories of knowledge relating to a particular concept. The meaning associated with words often rest on complex and elaborated bodies of knowledge. This is not to neglect the fact that words do have conventional meanings associated with them. According to cognitive semanticists, the conventional meaning associated with a word serves as a prompt for the process of meaning construction: the selection of an adequate interpretation against the context of the utterance. Evans and Green illustrate the case by presenting three sentences (reported below) including the word safe, which has several meanings, and instruct the reader to consider them against the context of a child playing on the beach. (12) (a) (b) (c) The child is safe. The beach is safe. The shovel is safe. The first sentence means that the child will come no harm, the second means that the beach is an environment in which the risk of the child coming to harm is minimal, while the third means that the shovel will not cause harm to the child. There is no single fixed property that safe assigns to the word child, beach, and shovel: in order to understand what the speaker means, we rely on our encyclopedic knowledge relating to children, beaches, and shovels, along with our knowledge relating to the notion of “safety”. Then, the meaning of the sentence is built selecting the meaning which is more appropriate in the context of the utterance. The fourth assumption associated with cognitive semantics is that meaning is not simply encoded by language. Rather, it is constructed at the conceptual level: meaning construction is equalized with conceptualization, “a dynamic process whereby linguistic units serve as prompts for an array of conceptual operations and the recruitment of background knowledge.” (Evans and Green 2006: 162). Therefore, meaning is not a static entity, but rather a construction process which draws upon encyclopedic knowledge and involves inferencing strategies connected to different aspects of conceptual structure, organization, and packaging. This quality of meaning construction has been extensively modeled by Gilles Fauconnier, who emphasizes the role of mappings, i.e. local connections between distinct mental spaces, conceptual packets of information build during the online processing of meaning construction36. Evans and Green provides the following example (from Taylor 2002: 530). (13) In France, Bill Clinton wouldn't have been harmed by his relationship with Monica Lewinsky. Following Fauconnier and Turner (2002), Evans and Green posit that the sentence 36 Details on Mental Space Theory can be found in Fauconnier and Sweetser (1996); Fauconnier and Turner (2002). 43 Grounding Meaning in Everyday Experience in the World above prompts us to set up a mental space for reality, in which Clinton is the U.S. President, Lewinsky is his intern, they have an affair, are found out, causing a scandal to break. Then, we have to set up a second space for another aspect of reality, containing the President of France and our knowledge that French culture is less rigid on this topic than American culture. Then, we have to blend the two spaces into a third, where Clinton is the French President, Lewinsky is his intern, they have an affair and are found out, but no scandal stirs up. Owing to the conceptual mapping relating the first and the second spaces to the third space, we get something more than the original input spaces, learning that cultural and moral sensitivities toward extramarital affairs between politicians and members of their staff are different in the United States and France. This meaning is constructed on the basis of complex mapping operations between different reality-based scenarios, which combine to give rise to a new counterfactual scenario. The blended space gives rise to a new meaning, not available from encyclopedic knowledge. Now, we can spend some time to illustrate slightly more in detail the relationship between the first two assumptions listed in (11), focusing on the notion of “image schema” (which will be crucially important for the case study illustrated in ch. 3). In the previous subsection, we cited the definitions of “image schema” provided by Johnson (1987) and Gibbs (2005), but perhaps it is relevant to our current purpose to also mention Oakley's (2007: 215), according to whom “an image schema is a condensed redescription of perceptual experience for the purpose of mapping spatial structure onto conceptual structure.” Indeed, image schemas are derived from sensory and perceptual experience as we interact with and move about in the world. As an example, Evans and Green mentions the fact that the vertical orientation of the human body exerts a strong influence on our interplays with the environment, giving rise to the “up-down” schema: our (functionally asymmetrical) vertical axis, which interacts with gravity, gives rise to meaning as a result of how we interact with our environment. It follows that image schemas are not innate but rather emergent: this kind of experience results from our bodily interactions with the world; consequently, it arises in conjunction with our physical and psychological development during early childhood. Since the label “image schema” is composed of two words, which have been long exploited in philosophy, it is important to underline the meaning of the terms image and schema. As pointed out by Evans and Green (2006: 178), the term “image” in “image schema” is equivalent to the use of this term in psychology, where imagistic experience derives from our experience of the external world, which comes from sensoryperceptual mechanisms that include, but are not restricted to, the visual system. Therefore, in spite of the fact that the term “image” is restricted to visual perception in everyday language, its application in Cognitive Linguistics encompasses all types of sensory-perceptual experience. With regard to the term “schema”, it means that image schemas are abstract (rather than detailed) concepts consisted of patterns arising from iterated instances of embodied experience. The relation between “images”, “schemas”, and “image schemas” is captured by Oakley (2007) in the following few lines: To summarize thus far, a schema has been historically defined as a fixed template for ordering specific information, whereas an image has been defined as a representation of specific patterns capable of being rendered schematically. As a composit notion, images schemas are neither fixed nor specific, even as they manifest characteristics of each. Many image schemas have “topological” characteristics, insofar as they constitute “spaces” sectioned into areas without specifying actual magnitude, shape, or material. Lack of specificity and content makes image schemas highly flexible preconceptual and primitive patterns used for reasoning in an array of 44 2. From Cognition to Grammar contexts. (Oakley 2007: 216-217) Evans and Green also list several properties which cognitive linguists usually associated with image schemas. The first (and perhaps foremost) regards the fact that image schemas are considered to arise from sensory experiences in the early stages of human development that precede the formation of concepts. Conceptual representation is seen as arising only when the recurrent patterns of sensory information and have been extracted and stored as an image schema. This means that image schemas are thought to be the first concepts to arise in the mind and thus to represent the foundations of the conceptual system. Given the sensory-perceptual nature of image schemas, it follows that they are particularly schematic, and that we are often unaware of them. Nevertheless, image schemas underlie several much more specific concepts. For instance, the “container” schema, which consists of the structural elements interior, boundary, and exterior, represents the structural basis for any specific kind of container (e.g. a pencil-case, a strongbox, a house, a bag, etc.). Deriving from embodied interaction with the world, image schemas are inherently meaningful. Asserting that embodied experiences are inherently meaningful means that they have predictable consequences. For instance, if we hold a cup of coffee in our hands, we expect that moving the cup, the coffee moves with it. This is because the cup contains the coffee, and containment is defined by boundaries, constraining the location of any entity within these boundaries. Then we know that the cup exerts force-dynamic control over the coffee. This kind of knowledge (which we tend to take for granted) is acquired through our interaction with our physical environment. Another important property of image schemas is that they take a form in the conceptual system that mirrors the sensory experience being represented: Evans and Green point out that while we can describe image schemas using symbolic units such as words and/or diagrams, they are represented in the mind as sensory-perceptual experiences rather than as symbols37. Since they derive from sensory experience, image schemas are represented as summaries of perceptual states recorded in memory. Nevertheless, they give rise to consciously accessible concepts, then structuring linguistic concepts. Image schemas are often internally complex, i.e. they comprise aspects which can be analyzed separately. For instance, the SPG schema (already mentioned in §2.1.1) consists of a starting point (the source), a final destination (the goal), and a series of continuous locations in between. Like all complex image schemas, SPG represents an experiential gestalt, a coherent whole with an internal structure. Consequently, distinct components of SPG can be highlighted and referred to (as will be seen in ch. 3). It seems appropriate to specify that image schemas differ from mental images: the latter are rich of details and are the result of a demanding and at least partly conscious activity which involves visual memory, whereas the former have a schematic nature, emerging from ongoing embodied experiences across distinct types of sensory experience; therefore, image schemas are not specific to a particular sense but instead are abstract patterns arising from a vast range of perceptual experiences and as such are hardly available to conscious introspection38. Lakoff (1987) affirms that image schemas can also undergo transformation. Since 37 This is the reason why we cannot learn to perform certain activities (such as driving a car, playing an instrument, doing a sport etc.) by simply reading a manual, but instead we need to practice a lot. 38 Evans and Green remark the important fact that blind people have access to image schemas, showing the multimodal (or perhaps intermodal) roots of these schemas. 45 Grounding Meaning in Everyday Experience in the World embodied experience, the source of image schemas, is ongoing, an image schema can transform into another. A widely cited example of this phenomenon is the transformation between the mass schema (which relates to a homogeneous entity) and the count schema (which relates to a grouping of individual entities). Oakley (2007: 218) offers a straightforward instance: suppose you want to select a book from a large shelf in a library. From a distance, you will first perceive the shelf as a homogeneous mass but, as you move closer, you will then recognize it as a set of individual items. Image schemas can also appear in a network of related image schemas. Indeed, Johnson (1987) distinguishes seven different “force” schemas, consisting of a series of related schemas which share a bundle of common characteristics. The first of these schemas is labeled “compulsion”, and emerges from the experience of being moved by an external force (e.g. being pushed along in the crowd). The second “force” schema is the “blockage” schema, which derives from encounters in which an obstacle resist force (e.g. when we bump against a wall). The third schema related to force is labeled “counterforce”, and derives from the experience of two entities meeting with the same amount of force (e.g. when we hit somebody in the street). The fourth force-related schema is the “diversion” schema, and it occurs when an entity meets another entity, resulting in a change in trajectory (e.g. when a football hits a player). The next of these schemas is labeled “removal of restraint” and describes the case in which an obstruction to force is removed, allowing energy to be released (e.g. when a door we are leaning on suddenly opens, letting us in). The sixth “force” schema, called “enablement”, involves potential, rather than actual force vectors, and derives from our sense of potential energy in relation to the execution of a specific task. Last, the “attraction” schema derives from experiences in which an entity is drawn toward another due to the force exerted upon it (e.g. magnets). Image schemas can serve as the conceptual representation underpinning lexical items. According to Johnson, some “force” schemas are the “root” meanings (grounded in sociophysical experience) of English modal auxiliaries. Consider the examples in (14) below39: (14) (a) (b) (c) You must move your foot or the car will crush it. You may now kiss the bride. John can throw a javelin over 20 metres. Johnson argues that the root meaning of must, expressing physical necessity, derives from the “compulsion schema”. The root meaning of may, expressing permission is seen as deriving from the “removal or restraint” schema. The root meaning of can, denoting physical capacity is though to be related to the “enablement” schema. Johnson thus claims that the meanings associated to modal auxiliaries have an image-schematic basis, being ultimately grounded in embodied experience. Finally, as mentioned with regard to metaphor in §2.1.1, some cognitive semanticists argue that abstract thought has bodily basis, with image schemas providing the concrete basis for metaphoric mappings. Evans and Green (2006: 191) provide an example of the (very general) images schema “object” being exploited for the conceptualization of inflation. The “object” schema is a schematic representation emerging from embodied experience which generalizes over common properties of objects (they occupy a bounded region of space, they have certain attributes such as shape, weight, etc.). These properties are then mapped onto an abstract object like “inflation”, which lacks such 39 (14a) and (14b) are drawn from Johnson (1987: 50), (14c) is from Evans and Green (2006: 189). 46 2. From Cognition to Grammar physical properties. As a result, this abstract entity is understood in terms of a physical object, allowing us to quantify it and talk about its effects (e.g., If there is much more inflation, we'll never survive; Inflation makes me sick). Then, image schemas (which are grounded in preconceptual embodied experience) can serve to structure abstract concepts. Though the account of the notion of “image schema” provided here is certainly sketchy and does not do justice to the extensive literature available on this topic, it should nonetheless be helpful for the reader to understand the tightness of the connection between embodied experience, conceptual structure and semantic structure and, as a consequence, between Embodied Cognition and Cognitive Linguistics. We can now conclude the present section with a brief outline of the theory of Frame Semantics, an approach that has been highly influential in Cognitive Linguistics (both in the study of semantics and grammar), which in a sense is not directly related to the scope of the present study, but may be useful for the reader to remember in the next chapter, with particular reference to §3.3.2. Frame Semantics theory was developed by Charles Fillmore and his collaborators since the 70s, in order to emphasize the continuities between language and experience. In Petruck's (1996: 1) words, a frame is “any system of concepts related in such a way that to understand any one concept is necessary to understand the entire system: introducing any one concept results in all of them becoming avaliable”. In Frame Semantics, the meaning associated with a symbolic unit can be understood only against the frame with which it is associated. For instance, the meaning of a word is represented by a lexical concept and its background frame. A widely exploited example of frame is the “Commercial Transaction Frame” (see Fillmore 1982), which includes elements (or “participant roles”) such as a buyer, a seller, goods, and money. There is a large set of verbs linked to this frame, each of which evokes a different aspect of the frame. For instance, the verb buy focuses on the buyer and the goods, backgrounding the seller and the money, whereas the verb sell foregrounds the seller and the goods, backgrounding the buyer and the money. Verbs like pay, spend, cost etc. will show a different combination of “in focus” and backgrounded elements. Quite obviously, knowledge of what happens in a commercial event is a prerequisite in order to know the meaning of any of these verbs: knowledge and experience provide the motivation for the categories represented by the words. Petruck points out (1996: 2) that a complete description of these verbs must also include information about their grammatical properties and the various syntactic patterns in which they occur, providing the example reported below: (15) Carla bought the computer from Sally for $100. In this sentence, the subject Carla is the buyer and the direct object the computer is the goods. Neither of these elements can be omitted. The oblique objects from Sally (the seller) and for $100 (the money) are optional. Petruck correctly underlines the importance of the prepositions used with these obliques objects: from is the only preposition allowing the interpretation of Sally as the seller, while for is not the only possibility for the money. Grammatical information about the syntactic-semantic valence description is not specified in the frame, but can be deduced from the description of the elements. Frame semantics often makes use of the concept of prototype (see §2.1.1), applying it to linguistic analysis. For instance, it is claimed that the understanding of the word 47 Grounding Meaning in Everyday Experience in the World breakfast relies on the comprehension of the institutions and practices of the culture in which the category exists: the habit of eating three meals a day at about fixed times, and the fact that the first meal of the day, after a period of sleep, has a particular menu and is labeled breakfast. The conditions defining the prototype do not need to be all present at once in order for native speakers to use the word appropriately. Petruck points out that Americans may have different sleeping / eating habits, but in the various conditions the first meal of the day may still be called breakfast, independent of the time, menu, etc. Thus the word breakfast defines a category which can be used in a variety of contexts determined by the word's prototypic use. Another important concept in Frame Semantics is perspective. The sentence in (15) above describes a commercial transaction from the perspective of the buyer, which is profiled. The example below takes the perspective of the seller, instead: (16) Sally sold the computer to Carla for §100. Different words assume different perspectives on the same scene, and understanding the adoption of certain words for talking about that scene requires appealing to the history of events leading up to it. Petruck illustrates this phenomenon mentioning the fact that, in English, a traveler spending time on land is understood as interrupting a see voyage, whereas a traveler spending time on the ground is understood as interrupting a flight. Since Frame Semantics approaches meaning as an experience-based schematization of the speaker's world rather than in terms of necessary and sufficient conditions, it contrasts with truth-conditional semantics. Furthermore, defining the meaning of a lexical unit on the basis of its background frame rather than in relation to other units, Frame Semantics also contrasts with theories of semantic fields. According to Frame Semantics, meaning relies on its conceptual underpinning, whose knowledge is vital for the adequate use of a unit. This claim is clearly consistent with the Cognitive Linguistics view of semantic structure. There is also a second kind of frames, which provides a means of framing the communication context, the speech event frames. These frames schematize knowledge about types of interactional contexts which contribute to the interpretation of particular grammatical constructions. These frames contain schematic knowledge about registers of language use. Each of these provide a means of framing a particular type of linguistic interaction. For instance, the speech event frame for fairytales will differ from that for academic lectures or newspaper reports. As Evans and Green (2006: 228) point out, there are many lexical items which index a specific speech event frame: for instance, an English text beginning with the expression once upon a time will evoke a fairytale and the expectations connected to this particular type of frame. Speech event frames serve as organizers of culturally embedded knowledge structures. Finally Frame Semantics is also consistent with Cognitive Linguistics sharing the view that all linguistic units are meaningful. As a matter of fact, Frame Semantics encompasses both open-class and closed-class units of language: while the former provide the content of a grammatical expression, the latter provide a configuration of the scene. For instance, as mentioned above, the active and passive constructions of a certain proposition structure the same scene from different perspective, thus carrying a slight change in meaning. 48 2. From Cognition to Grammar 2.2. Insights from the NTL paradigm The present section is meant to work as a sort of link between the general theoretical roots of the present study (outlined in §2.1) and the specific linguistic model our work is going to focus on (see §2.3 below), providing some information on the research program within which such a model was carried out. This section is divided into three subsections: in the first, as stated in the title, we are going to provide a shallow outline of the topics the NTL paradigm is concerned with, following the organization of Feldman's book; in the remaining subsections, we are going to delve a bit more into two of the main pillars of the NTL paradigm, which are also crucial for the ECG model. In §2.2.2, we shall consider the notion of simulation mentioned in §2.1.1 in some more detail, pointing that it stands at the very heart of the semantic system of ECG and summarizing a series of studies carried out by scholars associated to the NTL program. The third section, instead, will be devoted to the explanation of the importance of computer simulations for the development of the projects undertaken by NTL scholars, again recurring to the help of some specific models designated and implemented by academics working within the NTL paradigm 2.2.1. The NTL paradigm in a nutshell The NTL paradigm is an interdisciplinary research program developed by scholars and alumni at the International Computer Science Institute and several departments of the University of California at Berkeley. This program was set up by Jerome A. Feldman (the director of the enterprise) and George Lakoff in the late 1980s, with the intention to devote the research efforts of a team of scholars to the task of uncovering the bridging links between human mental activities (with special reference to language) and the underlying neural correlates, integrating insights from various scientific fields, looking to Embodied Cognition and Cognitive Linguistics as their main theoretical background. During the two decades following the early days of NTL, several scholars involved in the enterprise have published a considerable number of academic papers addressing the various topics tackled by the several projects undertaken within the NTL paradigm; nevertheless, the only introductory monograph on NTL currently available is Feldman's (2006) book, a slim volume which provides a necessarily brief but comprehensive and highly accessible description of the assumptions, the goals, and the methods of the research program, together with an outline of the main projects undertaken by NTL scholars, illustrating the importance of the connections between the different levels at which the overall enterprise works and stating the rationale for the adoption of an interdisciplinary approach in order to develop a cognitively plausible theory of language. Feldman approaches the problem adopting a clear starting point: thought is a structured neural activity and language cannot be separated from thought and experience. As a consequence, he explicits the necessity to bridge neural structure and meaningful language. In order to deal with this issue, Feldman states the necessity to work on three related centerpieces: neural computation (applying the general theory of 49 Grounding Meaning in Everyday Experience in the World computation to the neural circuitry of the brain)40, the embodied nature of thought and language (using neural computation to account for the embodiment of thought and language), and the integral organization of language (positing multifaceted constructions which cover all subfields of linguistics). Feldman's work is divided into nine parts. The first part introduces the idea (fairly widespread among theorists of Embodied Cognition) according to which when we think about an action or experience, we exploit the same neural structure we use in thought and language about such an action or experience (see e.g. Gentilucci 2003; Gallese and Lakoff 2005; Gallese 2006, 2008). The second part shows the biological basis of the neural theory of language. Feldman explains in a highly simplified but effective way how neurons process information and how neural networks work, often recurring to unsophisticated but powerful metaphors. The third part illustrates how specific neural networks can represent meaning, relying on the core, generally accepted idea that “mental connections are active neural connections” (Feldman 2006: 91). Feldman states that conceptual systems, based on categorization, are tightly linked to language, and he explains the role of computational and structured connectionist models in bridging the gap between the underlying neurobiological level with the manifest language level (see e.g. Feldman and Narayanan 2004). In the fourth part, Feldman deals with the actual implementation of some computational models in order to simulate the learning of concrete words. In this part, Feldman shows the role of cognitive structures massively exploited in Cognitive Linguistics studies, such as schemas and frames, in a child's understanding of experiences in the world and his subsequent learning of words to label them. Then, he briefly outlines the pros and cons of a particular connectionist model of learning spatial relation words developed within the NTL paradigm (Regier 1996; see §2.2.3 below). Part V is dedicated to learning words for actions, an activity in which the embodied knowledge of action plays a crucial role, stressing the importance of executing schemas, dynamic representations based in part on the motor and perceptual systems. In this part of the book, Feldman introduces a model for learning action words carried out within the NTL program (Bailey 1997). In the sixth part, Feldman deals with abstract concepts, highlighting the role of mapping in understanding abstract ideas: abstract concepts are mapped to and activate brain circuits involved in embodied experience, exploiting again the idea of categorization and in particular of basic-level categories (Lakoff 1987). The crucial importance of mapping is showed in the discussion of metaphor (Lakoff and Johnson 1980). Feldman then provides some evidence for the fact that understanding comes through mental simulation: “we understand a story by imagining ourselves in it or observing it”41 (Feldman 2006: 212). The existence of simulation is supported by results of experiments carried out in the field of neuroscience (see e.g. Gallese et al. 1996; Gallese et al. 2004; Gallese and Lakoff 2005; Rizzolatti and Craighero 2007). Part VII deals with the structure of action and events, what linguists call aspect, illustrating the simuation of dynamic concepts and stressing the importance of our 40 Neural computation is inherently different from the disembodied manipulation of amodal symbols posited in Mainstream Cognitive Science: it is an attempt to explain the role of the brain in the constant embodied interactions between organism and environment in computational terms, rather than an assertion that the mind actually works as a computer. 41 Feldman (2006: 233) reminds us that we can also simulate different scenarios and then blend them into a single mental space (see Fauconnier and Turner 2002). 50 2. From Cognition to Grammar ability to make inferences, based on our encyclopedic knowledge and our beliefs about the present situation. He also illustrates belief networks and outlines a computational model for understanding news stories (Narayanan 1997). In the eighth part of the book, Feldman turns his attention to grammar, highlighting the necessity for grammar to take into consideration all the different types of formmeaning mappings, in order to be plausible. He presents a model of neurally plausible grammar, the ECG model, to which the main part of my work will be dedicated (see ch. 3), Being a constructionist approach to grammar, ECG assumes that the construction (i.e. a pairing of form and meaning) is the fundamental linguistic unit. It was named “embodied” because in this formalism the semantic part of a construction is composed various kinds of embodied schemas. Feldman exploits some examples in order to show the workings of this formalism, arguing that on the one hand, grammar includes meaning and so it is closely related to other knowledge, but on the other hand, it is not simply a product of universal learning, taking a stand which is clearly at odds with the nativist stance embraced by most generative linguists, but at the same time is not as extreme as unconstrained emergentism, the position generally advocated by PDP connectionist modelers. The last part of Feldman's book shows how ECG provides Cognitive Linguistics with a formal notation, prominently by means of two main cognitive structures, schemas and constructions. Again, the author also makes use of an example of linguistic analysis carried out using the ECG formalism. Also, Feldman outlines a model of language learning based on ECG (Chang 2008)42. At the very end of his book, the scholar addresses two remaining mysteries, the origins and evolution of language43 and the nature of subjective experience, and makes a summary of the strong points and the shortcomings of the neural theory of language. As the reader will have noticed, the NTL program covers a rather wide range of related projects, and it would never be possible for us to cover them all here. For our current purposes, it is sufficient to underline three key core assumptions listed by Feldman in the introduction of his monograph. First of all, as repeatedly asserted in §2.1.1 above, the scientific explanation for human language and cognition needs to be based on our bodies, brains, and experiences. Second, it is also important to stress that most projects developed by NTL researchers aim at the formulation of explicit theories and/or models linking language to neural computation on the basis of what is known from linguistics on the one hand, and neurobiology on the other, even though there is not sufficient amount of evidence for the explanation posited by the researcher: Feldman reminds us that it is quite common in contemporary science, when a certain knowledge at both ends of a causal chain is available, to build and test theories to explicate the bridging links. Finally, it seems relevant to emphasize that the attempt to coordinate results from different fields of study, though not particularly widespread yet (but see e.g. Ellis and Larsen-Freeman 2010a), is motivated by the need to avoid to build theories that might seem perfectly adequate from a perspective but contradict evidence from another. While this requirement may frequently be missed, the attempt to combine key findings from different theories should nevertheless help to constrain the possible explanations to a narrower range of possibilities. 42 A concise outline of this model can also be found in Chang (2004). 43 However, some theories of linguistic evolution have recently been proposed, bringing together hints from usage-based linguistics, neuroscience, and connectionist modeling: see e.g. Chater and Christiansen (2010). 51 Grounding Meaning in Everyday Experience in the World 2.2.2. Simulation semantics In §2.1.1, we exploited the notion of simulation more than once, but we did not delve into this topic, preferring to procrastinate a discussion of this phenomenon to the present subsection. Contrary to what observed for the definition of “embodiment”, the authors working on the simulation hypothesis tend to provide rather equivalent definitions of “simulation”. Since this notion is vital for the theories of Embodied Cognition and represents a cornerstone of the NTL program, we find it useful to begin the present subsection providing a concise definition of the phenomenon. The one proposed in Evans' glossary seems to summarize the concept adequately: The ability to mentally activate or rehearse perceptual images such as particular sensations or experiences in the absence of the external perceptual stimulus which gives rise to the images. This is achieved, in part, by activating those regions of the brain responsible for processing the sorts of perception being simulated. (Evans 2007: 199) The ability to simulate a certain situation in the absence of the associated perceptual stimulus is inherently related to the nature of human embodiment: it is, to a great extent, dynamically carried out intracranially by the same neural structures which are active in the perception of that situation, but it is nevertheless grounded in our embodied experience in the situated environment. In §2.1.2, we saw that Cognitive Linguistics emphasizes the centrality of the semantic aspect of natural language, awarding a prominent role to meaning. We also mentioned the fact that semantic structure is seen as overlapping with a fraction of conceptual structure: meaning construction is seen as conceptualization. Because our conceptual structure derives from our bodily experience in the world, it follows that conceptual structure is inherently embodied. As a result, the connection between linguistic meaning and mental simulation is not difficult to spot. Indeed, it is unsurprising that a growing number of cognitive linguists (together with some cognitive psychologists) advocate for a cognitive theory of linguistic meaning labeled simulation semantics. Simulation semantics is one of the pillars of the NTL program, which aims to work out the nature of the relation between human high-level cognitive processes to the biological activity of our brains. In particular, during the last decade simulation semantics has been the subject of the research work carried out by Benjamin K. Bergen and his collaborators. According to proponents of simulations semantics, language users build mental simulations of perceptual and motor experiences described by the language they comprehend, and then produces inferences on the basis of these internal simulations. Mental simulation is supposed to be active during the processing of both literal and figurative language (see. Bergen 2005a). This approach to meaning posits the activation of perceptual, motor, social, and affective knowledge that characterizes the content of utterances. As a consequence, language users learn to pair linguistic forms with the original (perceptual, motor, social, and affective) experiences, in a broadly constructionist fashion. In subsequent instances of language use, when the original stimuli are not contextually present, their experience is recreated through the activation of neural structures responsible for experiencing them. As Bergen (2005a: 257) points out, “meaning depends on an individual having had experiences in their body in the actual world, where they recreate those experiences in response to linguistic input, and use them to produce meaningful linguistic units.” 52 2. From Cognition to Grammar Bergen asserts that simulation semantics shows remarkable theoretical assets. First of all, it offers an account of what meaning is and how it could be acquired: a simulation performed in response to language can be learned on the basis of previous associations between experiences of the world and the language used to describe them. Understanding language, from this embodied perspective, is similar to the (perceptual, motor, social, affective, and even subjective) experience it denotes. It is relevant to remark that simulation is not limited to the activation of representations of past experiences, otherwise it would not be possible to understand unexperienced meanings, resulting in the impossibility to learn anything new. On the other hand, Bergen (2005a: 262) underlines, “mental simulation involves the active construction by the conceiver of novel experiences, on the basis of previous percepts, actions, and feelings, and while it is constrained and informed by these experiences, combinatorial and other capacities allow departures from them.” (see also Bergen and Feldman 2009). Therefore, understanding language about the world makes use of the same content of previous embodied experiences to create a new, similar mental experience. We can thus say that understanding a chunk of language can imply performing mental simulation of its content. It follows that the meaning of linguistic units corresponds to the contribution those linguistic elements make to building mental simulations. According to Bergen, simulation semantics is “the study of how different aspects of language contribute to the construction of mental imagery, and the corresponding theory of linguistic meaning as linguistic specifications of what and how to simulate in response to language.” A simulation-based view of meaning in the process of language understanding is made up of three components: • a semantic specification: the activation and combination of parameterized representations of the simulative content that linguistic constructions evoke, exploiting both the words' meaning parameters and constraints specified by the grammar of the utterance (see e.g. Bergen et al. 2004; Gallese and Lakoff 2005)44; • the subsequent performance of a dynamic mental simulation, on the basis of the semantic specification; • the production of inferences on the basis of the understander's knowledge of the experiences activated by the simulation just performed. Consistent with a view largely shared by cognitive linguists, the content of simulation is seen as contributed by different types of linguistic units: content words as well as function words and phrasal patterns have repercussions on the simulation process (see Goldberg 1995; Bergen and Chang 2005). As Bergen (2005a) points out, the link between linguistic constructions and mental simulation is not necessarily direct; on the contrary, an intermediary stage of representation is posited: putting words together in an acceptable configuration during language production or recognizing the set of words in an utterance and their relations during language processing does not require access to the encyclopedic meaning of the constructions involved. On the other hand, words are combined in part on the basis of parameterization about their meaning. For instance, as we saw in §2.1.2 when illustrating Evans and Green's example of the polysemy of the English ditransitive construction, in order to occur in this kind of construction a verb must simply be interpretable as conveying a type of intended successful (literal or metaphorical) transfer of an object to a recipient. Apart from this (rather loose) 44 More information on the notions of semantic specification and parameterized representation will be supplied in §2.3.2 below, when we shall be dealing with ECG. 53 Grounding Meaning in Everyday Experience in the World constraint, which leaves us with the possibility to use a wide range of verbs, the verb constituent is not subject to more specific semantic limitations. This is because it would be problematic for the language user to access the detailed perceptual and motor content of each possible partial interpretation of an utterance, whereas linguistic units seem to encode just abstractions over aspects of the content they trigger in simulation, which in a sense form the “purely linguistic” representation of meaning: on the one hand, they are general enough to allow for operations of combination of linguistic elements; on the other hand, they simplify simulation by being tightly linked to the simulative details they are schematized over. Bergen also mentions another piece of evidence to keep linguistic representations separate from the perceptual and motor content they refer to: the possibility of understanding sentences that involve novel simulation content. Indeed, if sentence understanding were strictly constrained by the actual motor and perceptual possibilities afforded by our real world experience, then we would not be able to understand language which contradicts this experience, but in fact we are. Nevertheless, not all combinations are acceptable. In order to make this important point explicit, Bergen (2005a: 264) mentions the fact that the English verb to eat can be interpreted and simulated when it combines with direct objects which denote physical objects (even if these are not edible) or metaphorical objects and other classes, but the detailed affordances of whether an object can be eaten or not are not relevant until simulation: the act of preparing to perform a simulation is neither directly contingent on existing detailed perceptual and motor knowledge, no exclusively based on purely abstract criteria. Rather, linguistic constructions like words and clausal patterns combine on the basis of generalized characteristics of the simulations that the word evokes. If the meanings of linguistic constructions are parameterized representations of aspects of simulation, then comprehending an utterance involves setting up the meanings of the constructions that make that utterance up. When understanding language, the language user must assemble a set of constructions that fit together and best account for the utterance to be understood. This process is labeled constructional analysis (see. e.g. Bergen and Chang 2005; Feldman et al. 2009; Dodge and Bryant forthcoming)45 and it binds together the meaning of the words, phrasal constructions, etc. that make up a sentence46. These must fit together on both formal and semantic levels. The result of this process of analysis is the semantic specification mentioned above, which provides the parameters for a simulation to be run. The simulation process is then followed by the propagation of the inferences it generates. Apart from giving rise to an internal experience of the situation denoted by language, an important asset of simulation is that it provides rich perceptual, motor, and affective detail about the content of an utterance, which the understander can then use to update their beliefs about the world. Nevertheless, Bergen clarifies that inference propagation could in principle proceed independently of the simulation mechanism. Through associative learning, a language user could acquire that whenever a certain language chunk is uttered, they should simply update their beliefs in some determinate way. Even though simulation is important in language understanding, it does not follow that this process is always necessary for the comprehension of every linguistic unit. 45 As in the case of semantic specification and parameterized representation, the concept of constructional analysis will be deepened in §2.3.2. 46 We are just considering the hypothesis of an analysis limited to syntactic and lexical constructions. When morphology is taken into consideration as well, such a binding process will also have to take morphemes into consideration. 54 2. From Cognition to Grammar The simulation semantics approach considers meaning as based on parameterizations over aspects of simulation, together with the simulations itself that those schematization generalize over, and the following propagation of inferences. From this embodied perspective, rather than being simply static representations, the meaning of linguistic units combine static, parameterized knowledge with the dynamic performance of the simulations they specify. Besides the fact that there are theoretical reasons to see meaning as tightly linked to the internal recreation of embodied experience, simulation semantics is also supported by a substantial amount of empirical evidence 47. As a matter of fact, both behavioral and brain imaging studies seem to validate the perspective that language understanding is based on the (partially unconscious) simulation of previous embodied experiences, using brain structures dedicated to perception and action (cfr. Glenberg 2007; Rizzolatti and Craighero 2007). With regard to neuroimaging research, whose role is crucial to evaluate the overlapping between brain areas responsible for acting or perceiving and those involved in language understanding, several studies denote that certain brain regions are reliably associated with both perceiving objects in particular places of the visual field and performing actions with particular effectors (see Bergen 2007: 296-298). The most common techniques employed for imaging the living human brain are PET (Positron Emission Tomography) and fMRI (functional Magnetic Resonance Imaging), two non-invasive methods which function through the detection of metabolic changes in particular regions of the brain, since blood flow correlates positively with neural activity. PET studies consist in the introduction of a substance which emits positrons into the subject's bloodstream and blood flow to particular regions is measured by the intensity of the positron emission in those regions. fMRI studies measure changes in the magnetic resonance of regions of the brain, which are caused by changes in blood flow, making use of magnetic fields and radio waves. As Bergen (2007: 297) highlights, this kind of method allows a snapshot of the brain at a given time, providing information on the regions being active. Brain scanning studies are clearly important to the study of simulation, since motor areas (regions of the primary, supplementary, and secondary motor cortices, and regions of the cerebellum) are somatotopically organized (i.e. distinct body regions map to distinct regions of the given brain area) and visual cortices are retinotopically structured (i.e. parts of the retina are mapped spatially onto parts of the visual cortex). In other words, motor imagery activates the same parts of the motor cortex responsible for performing actions using the same effectors and visual imagery involves brain regions for perceiving similar images. Imagery thus appears to be selectively executed by the same motor and perceptual areas responsible for the real-world correlates of the particular imagery performed. This line of empirical evidence appears to be matched by the evidence shown by a remarkable amount of psycholinguistic experiments on language comprehension. It is this kind of evidence we are going to address in the next few pages, focusing on some experiments undertaken at the University of Hawai'i at Manoa and the University of California at San Diego and reported in a series of papers by Bergen and other members of the Language and Cognition Lab48. Narayan et al. (2004) provide an account of two experiments on verbal semantics. 47 Bergen (2007) points out that, on the other hand, “There is very little evidence for the symbolmanipulation view of language understanding.” Evidence for simulation semantics may thus be considered as a point on which Embodied Cognition overwhelms Mainstream Cognitive Science. 48 For a more detailed survey of neuroscientific evidence broadly converging with the results obtained by these psycholinguistic studies, the reader is referred to Pulvermüller's (2007) contribution. 55 Grounding Meaning in Everyday Experience in the World In the first one49, subjects were shown an image of a figure performing an action. Then, they were presented with a written verb. They were required to decide quickly whether the verb matched the image or not. Three different categories of verbs were used: those which matched the image shown to the subjects, those which did not but involved the same effector, and those involving a different effector.50 The hypothesis behind this study was that subjects would have longer reaction times when comparing nonmatching stimuli involving the same effector than when comparing non-matching stimuli involving different effectors, as there would be a greater level of interference between the two concepts. The results were consistent with this hypothesis. In the second experiment, subjects were shown a verb, followed by a near-synonym, a nonmatching verb involving the same effector, or a non-matching one involving a different effector. They had to decide whether the two verbs had approximately the same meaning or not. Again, the experimenters' hypothesis was that the reaction time for mismatches involving the same effector would be longer than otherwise. The hypothesis was corroborated by the results of the experiment . Then, Narayan et al (2004: 11) conclude that “language and cognition are ultimately grounded in sensorimotor experiences”, finally asserting that the organization of the sensorimotor system turns out to be of crucial importance for theories of language understanding. Bergen and Wheeler (2005) go a step further in the description of the language understanding process. In their paper, the authors outline two experiments aimed at verifying what kinds of language trigger simulation. On the basis of the results of these experiments, Bergen and Wheeler arrive at the claim that processing sentences describing both actions performed on or by the hearer and actions performed on or by third persons engages the activation of motor control circuitry. In the first experiment, subjects were presented with sentences denoting motion either away from or toward the body, and they had to state as quickly as possible whether such sentences made sense or not by pressing a “yes” button or a “no” button on a keyboard, which was rotated in order that it was perpendicular to the subject, allowing the “yes” button, a blank middle button and the “no” button to be in a straight line with four keys separating them from each other. All referents in the sentences were third-person. At the beginning of the experiment, response direction ordering was fixed with the “yes” button being farther from the subject than the “no” button, and then it was reversed halfway through the experiment. The results showed that subjects took longer to respond to sentences when the direction of their response was incompatible with the direction of the described motion. In the second experiment, again subjects had to decide quickly whether them sentence made sense or not. The critical sentences described an action performed either with an open palm or a clenched fist handshape. Subjects had to push a pedal with their hand, either using an open palm or a closed fist. The authors' hypothesis was that subjects would take longer to push the pedal when they had to do so with an incompatible handshape. Actually, subjects were quicker when the handshape required to respond was compatible with the handshape implied by the sentence. Observing the results of these two experiments, the authors assert that motor action is facilitated even when language describes actions involving only third-person participants, showing that motor imagery is more pervasive than was previously 51 49 This experiment was first outlined in Bergen et al. (2003). 50 The term “effector” identifies hands, feet, and mouth. 51 By saying that the results of an experiment corroborate a hypothesis, we mean that tests of statistical significance either rule out or at least considerably reduce the possibility that such results occurred by chance. 56 2. From Cognition to Grammar thought. Therefore, they draw the following conclusion: Both of these findings serve to strengthen an embodied theory of meaning wherein experiences of action and perception provide the basis for the subsequent mental recreation of these experiences such that they can be recruited for the purposes of interpreting and acting on the basis of upon linguistic input. (Bergen and Wheeler 2005) While the two studies outlined above dealt with sentence understanding, made use of written English material, and involved right-handed52 subjects who were native speakers of English, Tseng and Bergen (2005) focus on lexical processing and exploit signs from the American Sign Language (ASL henceforth). In this experiment, two different groups of subjects were involved: a group of ASL signers and a group of non-signers. Subjects were presented with twenty-two signs denoting literal motion of the arm and hand, twenty-two signs denoting metaphorical motion, and twenty-two signs “phonologically”53 directional but whose semantics is unrelated to that motion. Signs were shown in movies, and subjects were asked to determine whether two ASL signs (which they saw in sequence) were the same sign or different signs. They had to press a “go” button to see the sequence of signs, and to release it when they were ready to answer. Then, they had to press a “same” key or a “different” key on a keyboard, rotated in a way that the long axis extended outwards from the subject, so that in the first half of the experiment, subjects had either the “same” or the “different” button closer to them, and the situation was reversed halfway through the experiment. The “go” button was between the two. The dependent variables were the release time (i.e. the time it took for the subjects to release the “go” key) and the response time (i.e. the time it took for them to press the “same” button). The non-matching sign pairs showed movement in the same direction in order to rule out the possibility that the subjects distinguished matching from nonmatching signs just on the basis of direction cues. If interaction between sign direction and response direction correlated for the literal motion and metaphorical motion signs as well as for the “phonologically” directional signs, then this fact would confirm the possibility that the effect was due to the “phonology” and not the semantics of the signs, whereas if this effect held for signs encoding literal or metaphorical motion but not for those having only “phonological” motion, this would indicate that the effect is caused by semantic motivations. The results for ASL signers showed a significant interaction of sign direction and hand motion for literal and metaphorical motion signs, but no interaction for merely “phonological” signs, whereas with regard to non-signers, results showed no interaction in either case. This fact negates the possibility that the effect originates simply from the visual motion encoded in the signs, underlining the role of semantic processing in triggering such an effect. Tseng and Bergen then assert that processing words denoting motor actions activates motor imagery, a fact which strongly supports an embodied view of meaning. Bergen (2005b) investigates the role of simulation in spatial language processing, undertaking three different experiments, in order to attain three distinct but related scopes: determining whether lexical association with perception and motion is sufficient 52 In all of the experiments accomplished by Bergen and his colleagues, subjects were instructed to use their right hand only. 53 Tseng and Bergen in their study use the term “phonology” to refer to the “form” of the signs. Of course, this word should not be understood in the sense it normally assumes in linguistics. 57 Grounding Meaning in Everyday Experience in the World to yield simulation, or the integration of lexical elements into larger linguistic units is required; distinguish what linguistic elements can yield simulation; understand how detailed visual simulations are. In the first experiment, native English speakers heared an intransitive sentence over headphones and then they were presented with either a circle or a square in one of the four quadrants of a computer screen and had to identify the shape as quick as possible by pressing a button on the keyboard (“z” for circle, “x” for square). Subjects' reaction times were the dependent variable. In order to ensure that they attended the meaning of the sentences, some filler sentences were followed by a comprehension question. The hypothesis was that if subjects performed visual imagery in processing critical sentences, then we should observe interference when the direction of the sentence coincided with the location of the object. The results confirmed this hypothesis, showing a significant level of interaction between sentence direction and object location, thus allowing Bergen to conclude that sentences denoting upward and downward motion interfere with the categorization of objects in the same part of the visual field. The second experiment was analogous to the previous one, but presented subjects with vertically neutral verbs and nouns canonically associated with upness (e.g. ceiling) or downness (e.g. cellar). Again, the results show a significant interaction between object location and sentence direction: sentences with subject nouns canonically associated to upness or downness selectively interfere with the visual processing of objects in the same part of the visual field. The final experiment was aimed at discovering if sentences including verbs expressing motion drive visual imagery even when they do not denote actual physical motion or if the contribution words make to the construction of a simulation is mediated by an interpretation of the larger linguistic unit in which it is embedded. The subject nouns of the sentences exploited in these experiments denoted entities that could not physically move up or down, yielding sentences that made metaphorical use of the motion verbs. By contrast with the result of the previous two experiments, there was no significant interaction between sentence direction and object location with the metaphorical sentences. Bergen posits four possible explanations for these results. First, metaphorical sentences about changes in quantity did not drive visual imagery in the given part of the visual field. Second, visual imagery was triggered with metaphorical subjects, but it was weaker. Thrid, visual imagery was performed but in a different time scale54. Fourth, the metaphorical sentences did not trigger any visual imagery at all (from the perspective of simulation semantics, this explanation is quite unlikely, since it is inconsistent with the fact that metaphorical language is deeply understood). In summary, Bergen's study highlights the fact that processing sentences that denote events that tend to take place in a certain part of the visual field triggers inference on actually using the same part of the real visual field, both when the location of the event is denoted by a verb or a noun. However, having an up- or down-associated word in a sentence is not sufficient to yield interference, but the sentence must literally encode a scene involving the relevant location in the visual field; therefore, it is not lexical priming that produces the interference but rather the performance of mental imagery corresponding to the meaning of an utterance. A further important study on the role of motor imagery in language understanding is Wheeler and Bergen (2010). This clear and accurate study addresses a major issue: the level of detail at which motor information is simulated. Subjects (who again were native speakers of English) had to read English sentences and distinguish those of them which 54 It seems relevant to underline that these three explanations are not mutually exclusive. 58 2. From Cognition to Grammar made sense from the ones which did not. They had to express their judgment by pushing a button using a pre-assigned handshape. The response pattern (either with fist meaning “yes” and open palm meaning “no” or the other way round) was casually assigned to each participant and reversed halfway through the experiment. The experimenters used two different types of stimuli: some sentences contained a verb clearly implying the use of a determined handshape, whereas others had a verb whose action could be performed with one handshape or the other depending on the grammatical context. This discrimination was useful to distinguish the cases in which the compatibility effects resulted from the lexical meaning of the verb from those situations in which they resulted from the simulation of the event described by the sentence as a whole. Wheeler and Bergen's hypothesis was that if subjects were simulating the relevant handshape, they would take longer to react when the physical response expected them to perform an action incompatible with the one they had just simulated. The interaction between response handshape and sentence handshape proved to be significant and confirmed the authors' hypothesis: subjects pressed the button faster when the handshape required to respond was compatible with the handshape implied by the sentence. According to Wheeler and Bergen (2010), “this result indicates that understanders are incorporating fine motor detail into their internal simulations of motor events when understanding action language.” Therefore, this experiment suggests that not only do we simulate motion when we are engaged in language understanding, but our embodied simulation also includes a level of motor detail which is more precise than previously thought, even though still schematic. Bergen and Wheeler (2010) performed two experiments aimed at the investigation of the role of grammatical factors in mental simulation, focusing on aspect. Consistent with one of the main tenets of Cognitive Linguistics, according to which grammar is inherently meaningful, they hypothesized that grammatical factors can affect the content of mental simulation in their own right, rather than simply assembling the contributions of lexical items. In particular, Bergen and Wheeler predicted that progressive sentences would drive the motor system to perform mental simulation of the core of the described action, while perfect sentences would not. If the results confirmed this hypothesis, it would follow that different grammatical constructions produce systematically different uses of the motor system, then the motor system is only activated when the words and the grammar of an utterance not only indicate a motor action event but deal with the processes of that motor action as well. Subjects were asked to press a button placed in the middle of a keyboard in order to be presented with an English sentence and to release the button when they had read it. Then, they had to press a second button, located either closer or farther from the body of the participant, in order to indicate if the sentence was meaningful or not. The key independent variable regards the compatibility or incompatibility of the response direction with the sentence direction. In the first experiment, right-handed native speakers of English were presented with eighty critical transitive sentences, constituted by forty pairs with a sentence describing movement toward the body and the other denoting movement away from the body. Twenty pairs included sentences characterized by different verbs, while twenty pairs included sentences with different object noun phrases. All sentences involved thirdperson referents. Each subject had to use a computer keyboard rotated 90° counterclockwise, so that it laid in front of the participant along their sagittal axis. In the first half of the experiment, the “yes” button was farther from the participant's body, whereas the “no” button was closer to it. The condition was reversed in the second half of the 59 Grounding Meaning in Everyday Experience in the World experiment. Sentence direction was therefore crossed with response direction. Significant compatibility between action and sentence (i.e. faster button presses to indicate meaningfulness judgments when the direction in which participants had to move their hands was the same as the direction of motion implied by the sentence) was expected if the progressive yielded detailed mental simulation of event-internal actions, then this effect should be present in response to progressive sentences about concrete hand motions. Three independent variables were involved: not only sentence direction (toward vs. away from the protagonist's body), but also response direction (toward vs. away from the participant's body), and sentence type (noun-manipulated vs. verbmanipulated). The first two variables were found to interact significantly. Thus, progressive sentences produced a reliable compatibility between action and sentence, regardless of whether the sentence encoded motion through contributions by the verb or the object noun, in line with the predictions made by simulation semantics. The second experiment differs from the first in the aspect of the stimuli. If perfect sentences focus mental simulation on the end state of an event, then they should not show significant effect on action-sentence compatibility on response time with these perfect sentences. As predicted, the interaction between sentence direction and response direction was not significant, confirming that perfect aspect blocks mental simulation of the core of described events. Thus, Bergen and Wheeler conclude that the results of the two experiments suggest that grammatical structures affect how language understanders engage their perceptual and motor system to perform mental simulations of described content. From this point of view, grammar influences mental simulation in three ways: providing content, binding together the contributions that content words like nouns and verbs make to the simulation, and finally supplying information such as what factors to focus on or what perspective to adopt. Finally, we shall review a recent experiment undertaken by Dennison and Bergen (2010). Given the fact that motor action is subject to culture-specific prescriptions (the authors mention the fact that Korean people use both hands to handle a small object to someone of higher social status, but one hand with peers or inferiors, while many other cultures do not share this convention), this experiment investigates whether sociallycontingent prescriptions for motor action reach into other aspects of cognition as well, focusing on language-understanding. Native Korean speakers born and raised in Korea were presented with critical Korean sentences describing transfer of small objects, conventionally transferred to people of higher status using both hands and to people of equal or lower status using just one. Participant performed meaningfulness judgment task. They were seated in front of a computer with two keyboards aligned side by side in front of them, oriented in such a way that the long axis of the keyboard projected out directly in front of the participant. Subjects were asked to press two keys located on the edge of the keyboard close to their bodies with their thumbs to begin auditory presentation of a sentence. They were then asked to release these buttons and to press the “yes” or “no” buttons on the keyboard to indicate their decision. The response buttons were positioned to require the participant to use either both hands (so they were quite far from each other) or only their right hand (and then they were close to each other). The time from the release of the first button to the “yes” or “no” button press was measured in order to evaluate the possible influence of socially expected action on subsequent motion performance. Dennison and Bergen predicted that two-handed responses would be faster when participants had just heard a sentence about transferring an object to someone of lower 60 2. From Cognition to Grammar status, whereas one-hand responses should be faster when the preceding sentence described transfer of an object to someone of higher status. This prediction is due to the fact that priming effects of action language on motor control are quite sensitive to timing. When there is a delay (more than half a second) the between the word denoting the action and the action itself, language about action facilitates broadly similar motion actions, such as the direction of motion or handshape. On the other hand, when the critical action word and the motor action are temporally aligned (within half a second), similar but not identical actions will undergo inhibition. Since in Korean the verb occurs at the end of the sentence, the action subjects are asked to perform falls within the half a second window (contrary to what happens in languages like English). The results of the experiment seems to confirm this hypothesis. Dennison and Bergen highlight that the finding hints that motor simulation reflects the “computation” of socially appropriate action, which must take into account not only merely low-level physical properties of a mentioned object, but also social variables, but they also underline that the presence of honorifics, which encode the speaker's sociocultural regard towards the referent, may play a crucial role in the determination of these results, emphasizing the necessity to carry out follow-up experiments. In summary, Dennison and Bergen show that the motor simulation people engage during language comprehension includes the number of hands one would use to perform a described action. Moreover, they suggest that cultural-specific rules for motor control may enter into these mental simulations. Last, word order appears to have a certain role in the direction of the interaction between action and sentence. Therefore, even though Dennison and Bergen call for carefulness, since other variables may also come into play, these findings seem to support a view of language understanding as a dynamic process which engages heterogeneous cognitive systems, using our comprehension of physical actions and the surrounding environment. Here, we have focused on some of the many experimental studies carried out by Benjamin Bergen and his collaborators because Bergen has long been involved in the activities of the NTL program and he represents a key figure in the development of the ECG model we are going to introduce in §2.3.2, but several other scholars have contributed to the studies on the nature of simulations. Consequently, the experiments outlined above represent only a small fraction of the number of studies on mental simulation carried out in the fields of psycholinguistics, neuroscience, and cognitive psychology. Nevertheless, their description should be sufficient to provide the reader with an idea of the kind of empirical evidence in favor of the adoption of an embodied approach to language and cognition in general, and a simulation perspective on linguistic meaning in particular, is nowadays available. The concept of mental simulation in language understanding, which is consistent with the basic assumptions of Embodied Cognition and Cognitive Linguistics, was foundational for the development of the ECG formalism which, as we shall see in the next section, represents a computational interface between the relatively discrete units used in language and the vast amount of conceptual knowledge they reflect, and may also be understood as a sort of potential intersection point between the intrasubjective aspect of language and its social dimension. 61 Grounding Meaning in Everyday Experience in the World 2.2.3. The computational dimension It was already pointed out that Embodied Cognition and Cognitive Linguistics represent two of the main cornerstones of the NTL paradigm. Nevertheless, a third dimension have always represented a pillar of this research venus as well: computational modeling. As a matter of fact, since the early days of the NTL program in the late 80s, all the projects undertaken within this enterprise have always involved the design of computational models developed in order to be suitable for machine-implementation. This fact is hardly surprising, for two main reasons: first of all, computer science and artificial intelligence have always been one of the main concerns of cognitive scientists; second, most scholars who are part of the NTL research group are themselves computer scientists and/or engineers (including the founder and director of the effort Jerome A. Feldman). The strong interest for computer simulation of human cognitive functions on the part of NTL scholars is not at odds (as perhaps it may seem) with the tenets of Embodied Cognition: although advocates of Embodied Cognition reject the view of the human mind as a computer, it does not follow that digital computers cannot be helpful for the simulation of cognitive processes. On the contrary, from the perspective of the NTL group, computer models are a valuable tool which allows scholars to test their theories, providing real-time feedback and supplying insights to correct and improve models55. Moreover, in several cases it is the only way to make and test hypothesis on how empirically observed cognitive processes may be realized at the biological level, since it is generally not possible to directly observe their inner working mechanisms. In the present subsection, we shall be outlining three projects undertaken within the NTL paradigm during the last fifteen years, all of which are (to a greater or lesser extent) related to the constructionist model of grammar we are going to introduce in §2.3.2 and exploit in the next chapter. First of all, it seems relevant to begin our survey briefly outlining Terry Regier's model of word learning, which is generally considered one of the major outcomes of the NTL paradigm. In his book, Regier (1996) undertakes the enterprise of building a computational model able to simulate children's acquisition of terms related to spatial relations in the most realistic manner possible. This project is rooted in the author's interest in the search for linguistic universals, a topic which has always been the source of a heated debate among the linguistic community, especially between functional typologists and generative grammarians (see e.g. Croft 2009 for an overview). In particular, Regier is interested in semantic universals and the definition of constraints to what he calls the human semantic potential, i.e. the possible structures of meaning generated by the human linguistic system given the Cognitive Linguistics perspective on semantics mentioned in §2.1.2. The starting point of Regier's work is the assumption that, if meaning derives from human perceptual, sensory, and motor experience in a determined physical and sociocultural environment, then it will be constrained by the cognitive structures of the human body. In other words, the limits of the cognitive system will represent the cut-off point between plausible and implausible semantic structures. In a sense (as Regier himself admits), this view is loosely inspired by the Chomskyan argument of the innateness of language; nevertheless, Regier's perspective diverges from that of generative linguists, viewing the human genetic endowment as a set of independently motivated perceptual structures, rather than a set of formal rules. 55 This belief is by no means a peculiarity of the NTL research group. See e.g. Plebe (2004) for a critical survey of (neuro)computational models of language. 62 2. From Cognition to Grammar In order to fruitfully explore this working hypothesis, Regier decided to build a computational model of word learning, focusing on the acquisition of spatial terms. In order to realize his computer simulation, he looked at two traditions in connectionist modeling, noticing that a “hybrid” methodology combining properties of both would fit perfectly with his search for semantic constraints motivated by the limits of our perceptual and motor system (Regier 1996: ch. 3). The first of these methodology is known as parallel distributed processing (PDP from now on). PDP connectionism is a modeling methodology which exploits massive parallelism in processing (many computations proceed in parallel) and the distributed nature of representation: several computing units are organized in networks, and each unit participates in a number of different functions and is itself best characterized at a deeper conceptual level than that conveyed by the symbols of natural language. A typical PDP network is a multi-layer perceptron with full connectivity between layers: this kind of network includes an input layer, a hidden layer whose units develop a distributed representation of the input in the process of producing the desired output, and an output layer. Parallel representations are computationally efficient, allowing to carry out a complex task in a few time steps, and they allow the representation of similar things to be similar forms. As a result, it allows generalizations to be made on the basis of representational form, which captures the intuitive notion of conceptual similarity. Moreover, damages to representations cause graceful degradation of performance (rather than abrupt loss of information), for no single unit is responsible for a given symbolic entity. Furthermore, representations can develop as a result of the learning process (rather than being explicitly coded by hand), exploiting learning tools such as the backpropagation algorithm: backpropagation is first given a training set of input patterns paired with desired output patterns, then it trains the network to produce the input-output association specified in the training set. Then, it is possible to take a test set of input and desired output patterns (never presented to the network during training), and test the network's generalization to the test set after training. If it has captured the regularities in the training set, it should respond well to new items in the test set. When being trained under backpropagation, a network takes a training set input vector as input and produces an output vector at the output nodes. The output pattern is then compared with the desired output given in the training set for the input pattern. The goal is to reduce the difference between desired and actual outputs for all patterns. The backpropagation algorithm reduces error by performing gradient descent in an error measure represented by a mathematical formula (not shown)56. In order to compute the gradient, the network has to backpropagate an error term backward through the net. Since the 1980s, when connectionism began to spread, PDP networks have been used to model several different cognitive tasks, especially in the field of psychology. Regier (1996: 40) asserts that the appeal of PDP models in psychology derives from the ability to inspire reinterpretations of old findings and to provide a unified account of several different empirical results. PDP models are to a large extent empirically based: once a network is trained to model a particular cognitive function, our understanding of its working will be largely based on empirical observation of the network's behavior during simulation, since carrying out an analysis of the network's working is tremendously difficult and exacting to be performed offstage, being PDP models completely 56 An account of the mathematical apparatus used to actually implement models on machines goes far beyond our present scope. For details on the algorithmic “machinery” exploited for implementation, the reader is referred to the sources cited in this subsection. 63 Grounding Meaning in Everyday Experience in the World unstructured and data-driven. The second strand of connectionist modeling taken into consideration by Regier is structured connectionism, whose proponents claim that unstructured networks like those used in PDP models can hardly be adequate to solve problems in inherently structured domains (e.g. language); instead, they build their networks including knowledge of the target domain in their architecture. As a result, the structure of these networks reflects the cognitive/neural structure the modeler is positing. At the same time, structured connectionist models are unlikely to be used to accomplish more than one task, since the inclusion of structures dramatically limits their flexibility. Some light on the functioning of structured connectionism will be shown below when dealing with the Shruti project. For the time being, it is important to keep in mind that this kind of models are (relatively) easy to understand, so their analysis is not empirically based: the inclusion of structures in connectionist networks makes it possible to analytically determine constraints on the model's operation that are translatable into psychological terms, without running the network. Analyzability, as Regier remarks, is an important asset in cognitive modeling, for it allows the modeler to understand exactly why the network behaves as it does and thus supports credit (and blame) assignment: the scholar can tell which aspects of the model are responsible for which aspects of the observed behavior. Regier proposes constrained connectionism as an attempt to combine the benefits of both approaches sketched above, in order to retain the flexibility of PDP models while gaining the analyzability of structured connectionist architectures, designing PDP-like networks with built-in structures which constrain their operation. To put it in the simplest way, Regier's networks may be seen as having five layers: an input layer, followed by a hidden layer characterized by psychophysically and neurophysiologically motivated structural devices which constrains the operations of the network. Then, there is an unstructured hidden layer similar to those found in PDP models, which supports the formation of distributed representations, followed by another layer endowed with structural devices. Finally, we can find the output layer. The network as a whole is trained using backpropagation. Regier claims that constrained connectionism shows three main assets: First, constrained learning devices such as these lead to superior generalization from the training set. Second, if the structures built in are independently motivated in some manner, this helps to motivate the model as a whole. Third, (...), these devices provide ease of analyzability coupled with the advantage of distributed representations. (Regier 1996: 46) The first benefit derives from the fact that prior structuring of the network reduces the dimensionality of the search space, then reducing the number of solutions being considered. This in turn reduces the number of false solutions. Clearly, the way in which structural devices are designed is crucial to the success or failure of such a network: well-designed structures, accurately incorporating knowledge of domain into the network architecture, will constrain the hypotheses considered during the learning process to those consistent to the knowledge of the domain the designer was trying to capture. Building structures into a network can also help to motivate the network as a cognitive model, provided that there is independent motivation for the structure. For instance, in Regier's model the structural devices are motivated by neurophysiological 64 2. From Cognition to Grammar and psychophysical evidence concerning the human visual system57. Finally, analyzability is important for cognitive theorizing. Indeed, PDP models have often been said to be theoretically empty for not letting themselves to clear analysis: there is no straightforward way to translate the model into a set of principles governing the cognitive process under study. According to critics, it is not important whether the network accurately models the data as long as it is not possible to say why it does. A related point regards the fact that unconstrained PDP networks are said to be too powerful to be enlightning. Indeed, not only should an accurate cognitive model match human data, but it should also avoid to produce data that a human being would not produce. This is often not the case with PDP models. Including specific constraints, constrained connectionism avoids both these problems and also probably results more in line with the tenets of Embodied Cognition: indeed, PDP connectionists implicitly claim that regularities in human behavior are entirely due to the structure of the environment, whereas advocates of constrained and structured models implicitly award an important role in these regularities to the structure of the organism as well58. Regier's work represents a tentative attempt to partially determine the possible extent of spatial semantic variation examining underlying perceptual mechanisms. He draws inspiration from previous work pursued in the domain of focal colors (for a characterization, see Lakoff 1987): a well-known crosslinguistic study showed the existence of semantic universals in this cognitive domain, organized in an implicational hierarchy; subsequent research provided a partial explanation for this phenomenon, reducing the psychological percepts of colors to the neurophysiology of the visual system on the basis of neurobiological data59 and computationally based category combination. Though later studies showed that this is not exactly the case, for not always the responses obtained psychophysically match those obtained neurophysiologically, Regier underlines the importance of this kind of study for three main reasons: it proved the viability of determining perceptual constraints on semantic potential; it tried to provide an explanation for a psychological phenomenon combining neurobiological evidence with computational ideas; it failed to provide a complete explanation of the semantics of color, highlighting that the reduction of psychological percepts to neurophysiological structures is most often problematic. In his study, Regier confines himself to examining closed-class spatial terms, “in the expectation that this skeletal subportion of the domain as a whole will reveal whatever fundamental structure may be present” (Regier 1996: 16). The choice of spatial terms is due to a cluster of reasons. First of all, spatial relations are concrete and hardly contestable: they are objectively measurable and thus accessible to scientific investigation. Furthermore, space serves as a conceptual structuring device in language: spatial relations are often expressed by closed-class forms, linguistic forms that generally show core conceptual content; what is more, spatial terms are often extended, through metaphor, to nonspatial domains. The second phenomenon is particularly 57 Regier (1996: 47) emphasizes that the inclusion of nonlinguistically motivated structures is consistent with the “spirit of inquiry” into issues of semantic universality and relativity, potentially providing insights for a comparison between the roles of crosslinguistic variation and universally shared perceptual constraints on semantics. 58 It does not follow that PDP models cannot say anything interesting for us. On the contrary, we shall be glancing at some of these models when discussing the results of our analysis in §3.3.3. 59 It was noted that there is an observable neural coding for the categories red, yellow, green, and blue in terms of opponent cells of the geniculate nucleus and that there is also a distinguishable neural coding for white and black in terms of the responses of other cells. The peak response for each neural coding coincides to the best example of the corresponding color. 65 Grounding Meaning in Everyday Experience in the World pervasive in language at all levels: spatial organization is so central to human cognition that a great number of concepts are conceptualized in spatial terms (see e.g. Lakoff and Johnson 1980). In short, the fact that spatial relations often express core conceptual content and the metaphorical use of spatial relations throughout other parts of the semantic system of the world's languages seems to entail the prominence of space in human cognition, thus explaining Regier's interest in undertaking the enterprise of building a computational model of the acquisition of spatial terms. The great amount of crosslinguistic variation in spatial structuring is another important motivation for Regier's work, since it prompts the question of what kind of model could adapt itself to so many different structurings of space. Several studies show that the adoption of different systems by language communities to express spatial relations is not simply influenced and constrained by cognitive experience in the world, but the reverse condition holds as well: language affects cognition. In other words, our shared perceptual apparatus does not dictate what our experience of space is, but only constrains it, and the language spoken by members of a particular community has a role in shaping the nature of such an experience within the constraints set by the perceptual apparatus (see e.g. Chan and Bergen 2005). Therefore, the influence of the speaker's language may percolate through the semantic system to cognition60. Since even closely related languages can differ remarkably in their spatial systems, Regier sets himself to building a model able to learn spatial closed-class terms in English and four more languages with different degrees of relatedness to English. Regier's model works as follows: it is shown a set of movies of two-dimensional objects moving relative to one another (a trajector and a landmark, in Langackerian terms), such that each movie has been correctly labeled as a positive instance of a certain closed-class spatial term from a particular language. Then, the model learns the association between words and the spatial relations they describe. After the model has successfully accomplished its task, it should be able to distinguish which of the spatial terms learned would be accurate for describing previously unseen movies. In a few words, the model's task is learning how to perceive simple spatial relations in order to label them as a native speaker of a certain language would. The model is trained using error-backpropagation, so that when a movie is shown, only the output nodes corresponding to closed-class terms that properly describe the event will be activated. It is trained to do so without negative evidence. This is a crucial point, because it is not clear how to generalize from the positive examples seen: without negative evidence, the system will have no way to recognize when to stop generalizing. Nevertheless, humans do learn to generalize correctly, i.e. they avoid both undergeneralization and overgeneralization. The solution adopted by Regier is to make use of a heuristic suggested in the literature on child language learning, positing that positive instance for a concept is implicit negative instance for all other concepts (i.e. every movie in the training set is taken as a desired output of 1.0 for the output node relative ot the relevant concept, and a desired output of 0.0 for all other nodes). In spite of the fact that this “mutual exclusivity” principle is accompanied by the problem of false implicit negatives, which inevitably arises when this principle is applied to domains with overlapping concepts, the heuristic can still be used, considering implicit negative evidence as providing only weak evidence, less compelling than explicit positive 60 With regard to this aspect, the reader is also advised to see Clark's (2006) article on the influence of language use on thought, whose approach is somewhat different from (but at least some of his observations are consistent to) NTL scholars'. 66 2. From Cognition to Grammar evidence. As a result, implicit negative evidence is exploited as a violable bias, which is not seen as strong as explicit positive evidence (the weakness of negative evidence compared with positive evidence is included in the model through a mathematical formula, not shown). Regier's model was thus trained to learn all the concepts in a training set in parallel, and then it was tested on a test set. The model performed very well, showing an extremely low error rate. Prior knowledge is included in the model using three structures which reflect three general architectural principles. The first of the principles is orientation combination, which is based on the idea that spatial concepts involve a combination of evidence from various sources. The first of these sources is the direction of potential motion, which arises from experimental evidence that humans viewing static scenes mentally represent the forces that are acting on the objects, keeping them where they are; when they are shown a situation in which a previously present force is removed, they mentally represent the potential motion enabled by the removal of the counterbalancing force. The other sources of orientation combination are proximal orientation, the orientation of an imaginary directed line segment connecting the landmark to the trajector when two objects are closest, and center-of-mass orientation, the orientation of an imaginary directed line segment connecting the center of mass of the landmark to that of the trajector. The combination of these sources is considered to supply a characterization of the overall orientation of a trajector relative to the landmark. The second principle structured in the model is map comparison, based on the idea that the topological features of contact and inclusion may be detected by observing the trajector boundary and seeing if it or portions of it lie within or adjacent to the landmark. More precisely, it is suggested that these topological features can be detected by attending to the boundary of the trajector and determining the relation of each point along it to the interior of the landmark (i.e., the region including both its outline and its interior). The third principle assumes that the linguistic categorization of motion involves a particular structuring of motion trajectories: the source-path-destination schema (corresponding to the SPG schema mentioned in §2.1.1), whereby the path of an object through space is portrayed in terms of a starting point or source, an end point or destination, and the path from the source to the destination. These three structures bias the model, making some things easy to learn, and other difficult, thus putting constraints on its learning ability. It is relevant to stress that the postulation of exactly these structures (rather than others) is due not only to the fact that they are computationally convenient, but also to their level of independent motivation61. The adequacy of the model was tested on a set of spatial terms from five different languages: English, German, Russian, Japanese, and Mixtec. In all these languages, Regier's model performs excellent generalization from the trained set to the test set, showing that it can adapt to a range of quite distinct spatial systems, and that it is able to learn without being provided with negative evidence. The model also tells us something about semantic universals in the domain of space. Indeed, the constraints placed on the model's operation can then be translated into predictions about the human acquisition of spatial semantics. Regier explicitly argues for three specific predictions: • graded response predictions: children receive and are sensitive to linguistic 61 Regier (1996: 122) highlights that “if we have independent evidence for particular computational structures, these will be preferred over other equally efficacious structures, since we are being asked to take less on faith – there is already good reason to believe that structures of this nature exist.” 67 Grounding Meaning in Everyday Experience in the World input that indicates how central a member of a spatial category a particular spatial configuration is. As a matter of fact, an interesting property of the model is that it exhibits prototype effects, an aspect of its behavior it was not explicitly trained for, showing sensitivity not only to binary all-or-nothing categorizations, but also to aspects indicating that a certain configuration is a less-than-perfect instance of a specific term. The map comparison structure was designed to be able to respond in this fashion and the units in the rest of the model also have graded activation functions, but the structure of the model does not predetermine this kind of results, it simply allows it. This fact suggests that this graded structure is to be found in the input rather than in the perceptual mechanisms that process it; • intermediate sequentiality prediction: the only sequentiality that is relevant in closed-class forms for spatial events is the tripartite structure of beginning configuration, path taken, and ending configuration. No language will distinguish two events that differ only in the sequence in which configurations within the path occur: the path representation for the intervening frames confines itself to recording what has occurred over the whole path, it is not sensitive to exactly when particular events occurred. The description of complex paths in which path-internal sequentiality is significant will require the composition of closed-class forms; • endpoint configuration prediction: any language that has a closed-class form denoting motion out of or motion through some configuration will also have a closed-class form denoting either motion into that configuration or static location in it. The model learn to categorize events on the basis of the spatial features that occur at the end of some event it has seen, only62. For instance, Regier (1996: 157) asserts that “the existence of a form denoting through in a language implies the existence of forms like in or into, since the model predicts that through would be unlearnable otherwise.” (italics original). This prediction stands on the idea that closed-class linguistic expressions of motion through some configuration is linguistically relevant; this recognition in turn depends on the existence in the language of terms denoting either location in that configuration or motion into it. As Regier (1996: 205) himself observes in the closing lines of his book, the results of his work as a whole “are preliminary and perhaps best viewed as suggestive.” Indeed, Regier's computational model represents an attempt to investigate the possible constraints on the human semantic potential, addressing the problem of how closedclass terms expressing spatial relations can be learned. The model is never claimed to represent a fully adequate explanation of the complex phenomenon taken into consideration, nor is it meant to directly reflect neural structures in the brain. Rather, Regier specifies that he simply aimed to ground the acquisition process in a perceptually based model which incorporates structures similar to those in the visual system: the perceptual mechanisms posited in the model can then be seen as motivated by, rather than identical to, known neural structures. Nevertheless, Regier's model is considered as a major outcome of the NTL paradigm, since it provides some hints on the possible nature of language learning and the human semantic potential, which are also in line 62 This is consistent with experimental psycholinguistic data showing that children tend to linguistically categorize actions on the basis of their results rather than other aspects (e.g. the means by which the actions were effected). See Regier (1996: ch. 6). 68 2. From Cognition to Grammar with some work developed within the field of Cognitive Linguistics: the model succeeds in learning terms denoting spatial relations in five different languages characterized by different spatial systems, thus showing substantial flexibility. Furthermore, it is able to learn without explicit negative evidence, simply adopting a “mutual exclusivity” principle, which is treated as a violable bias rather than as a strict rule; finally, it matches some data available with experiments on child language acquisition. Moreover, it highlights the importance to place some constraints on computational models, in order to increase their explanatory power (the pertinence of this methodology is also reasserted in some of Regier's later work, see e.g. Regier 2003a, 2003b). Regier's work is important for the present study because it represents one of the first attempts made to relate the embodied nature of human cognition and its manifest linguistic realization, tackling the the problem from the perspective of the acquisition of spatial terms by children as his particular topic. Moreover, Regier's work exploits computational resources to posit perceptually motivated constraints to the human semantic system, combining the rigor of computational formalisms to the fuzziness of cognitive semantics, paving the way to subsequent projects, including the ECG model itself. It is relevant to highlight that Regier makes massive use of image schemas, trying to (tentatively) relate them to neural structures: this is especially salient from our perspective, since schemas constitute an all-important semantic resource in ECG. In particular, the methodology whereby Regier's computational model learns to couple form and meaning of spatial relation terms (applying a linguistic label to a movie representing the spatial relation between a trajector and a landmark) seem to strengthen the ECG view of schemas as conceptual gestalts which function as semantic primitives. Another strand of computational work developed within the NTL paradigm (and relevant to the present study) is Narayanan and Jurafsky's (1998, 2001) adoption of a probabilistic model in order to deal with ambiguous sentence processing. The two authors focus on the long-termed studied phenomenon of the local ambiguity between main verb and reduced relative in English, exemplified in (17) below (from Narayanan and Jurafsky 2001). (17) (a) (b) The cop arrested the forger. The cop arrested by the detective was guilty of taking bribes. Narayanan and Jurafsky's (1998) model is based on the assumptions that linguistic knowledge is represented probabilistically and that multiple interpretations are maintained in parallel, along with the conviction that the probabilities of these interpretations can be computed using a belief net63. The authors assume that linguistic knowledge comprises a collection of constructions, each of which is associated with a prior probability value, computed from relative frequencies from corpora or norming studies. Belief nets are data-structures representing probability distributions over a collection of random variables, and specific constructions are seen as values of latent variables that render top-down (syntactic, lexical, and argument structure) evidence and bottom-up (phonological or graphological information) evidence conditionally independent. This approach to human sentence processing allows the authors to a) quantitatively evaluate the impact of different independence assumptions in a uniform 63 A belief net is a probabilistic graphical model that represents a set of random variables and their conditional dependencies via a direct acyclic graph. See Ben-Gal (2008) for a concise characterization. 69 Grounding Meaning in Everyday Experience in the World framework, b) directly model the impact of highly structured linguistic knowledge sources with local conditional probability tables, while algorithms for updating the belief net can compute the global impact of new evidence, and c) develop an online interpretation algorithm, where partial input correspond to partial evidence on the network, and the update algorithm appropriately marginalizes over unobserved nodes. The model is applied to online disambiguation assuming the existence of a set of constructions consistent with the input data. Given top-down and bottom-up evidence, the posterior probabilities of the different intepretations are computed at different stages of the input. Then, a specific algorithm is used to cut off all constructions whose posterior probability is inferior to a certain ratio of the best construction. The overall posterior ratio requires propagating the conjunctive impact of syntactic and lexical/thematic sources on the model. The syntactic and thematic influences only depend on the value of the specific construction (either the main verb or the reduced relative). The conjunctive impact of these sources is computed using a model where it is assumed that whatever mechanism inhibits a specific source from providing support for a construction, is independent of mechanisms that inhibit other sources from providing support for the same construction. Online disambiguation studies seem to show that Narayanan and Jurafsky's model is able to account for garden-path disambiguation effects, but what is important for us here is that this computational model implicitly provides support for the view that language is processed incrementally, with syntactic and lexical cues interplaying in the process of selecting the correct interpretation of a sentence. In the case of the phenomenon analyzed by Narayanan and Jurafsky, there is generally a “momentum” (corresponding to the phrase which follows the main verb/reduced relative) when an interpretation wins over other competitors. Narayanan and Jurasky (2001) then extended the model to predict reading time given the probability of different interpretations, testing the model's results against experimental psycholinguistic data on the processing of sentences along the lines of those shown in (18) below: (18) (a) (b) The cop arrested by the detective was guilty of taking bribes. The crook arrested by the detective was guilty of taking bribes. The experimental data show that the probabilistic relationship between the noun and the head verb biases the thematic disambiguation decision. Indeed, sentences with good agents for the verb in question (like cop in (18a)) cause longer reading times for the following phrase than sentences with good themes for that verb (like crook in (18b)). Therefore, at the initial noun phrase, reading time is lower for good-agent sentences than good-patient sentences, but at the NP after the word by, reading time is lower for good-patient sentences than good-agent sentences. Narayanan and Jurafsky's model predicts longer reading time whenever an input word causes the best interpretation to drop in probability enough to “switch in rank” with another interpretation. The model consists of a set of probabilities placing constraints on sentence processing and a network that represents their independence relations. It is applied to online disambiguation by assuming the existence of a set of interpretations consistent with the input data. At different stages of the input, the posterior probabilities of different interpretations are computed. The model results are consistent with the online disambiguation studies with human subjects: the model predictions show close correspondence to the human judgments about whether a specific ambiguous verb was 70 2. From Cognition to Grammar used in the main clause or reduced relative constructions64. Moreover, the human time reduction effects (reduced vs. unreduced interpretations) increase for good agents but decrease for good patients in the “ambiguous region” (i.e. between the subject NP and the by-phrase). The model predicts this larger effect from the fact that the most probable interpretation for the good-agent case flips from the main verb to the reduced relative intepretation in this region. No such flip is observed for the good-patient case. On the basis of these results, Narayanan and Jurafsky suggest that this flipping of preferred interpretation may hint that reading time correlates positively with surprise. The reliability of Narayanan and Jurafsky's statistical approach is relevant to our study because, even though we are not going to implement any probabilistic model, in the next chapter we shall illustrate an ECG analysis of Italian caused-motion constructions claiming that the comprehension of a sentence is driven by the constant interaction of top-down and bottom-up cues during online processing, and Narayanan and Jurafsky's results, together with other computational work (carried out by scholars not directly associated to the NTL paradigm), represent a source of substantial support for our arguments (see §3.3.3). We can now briefly outline Shruti, a structured connectionist model developed by Lokendra Shastri and his collaborators in order to model the performance of inferences to establish explanatory and referential coherence. Basically, Shruti represents a neurally plausible system able to express causal knowledge involving n-place relations, limited quantification, and type-restrictions. It encodes specific events and also contextsensitive priors over events. It expresses dynamic bindings via the synchronous firing of appropriate node clusters and performs inferences via the propagation of rhythmic activity over node clusters. This propagation corresponds to a parallel activation of the underlying causal graph, thus resulting in fast reasoning. Shruti makes use of weighted links and activation combination functions at nodes, encoding soft rules and performing evidential inference. The model supports supervised learning which allows to fine-tune its causal model in a data-driven manner (see Shastri and Wendelken 2000). Shastri and Grannes (1996) describe a model designed to reach the goal of explicitly dealing with negation. For this model to be cognitively plausible, the authors posit a set of conditions that it must satisfy: it should allow inconsistent facts and rules to co-exist in long-term memory, but should also be capable to detect contradictions whenever inconsistent beliefs that are within a certain inferential distance from each other become co-active during an episode of reasoning. Shastri and Grannes' paper illustrates the model's encoding of the simple sentence Mary loves Tom, which can be seen, in firstorder logic terms, as the predicate love with two roles, a lover and a lovee (which are to be filled by the Mary value and the Tom value, respectively). Such a predicate is encoded by a cluster of nodes: besides the two nodes which represent the two roles, there is an enabler node, whose activation means that the system is trying to determine whether the currently active dynamic instance of the predicate is supported by the knowledge in the memory; moreover, we can find two collector nodes, one of which is activated when the answer to the previous question is positive, and the other is activated when such an answer is negative; these nodes are provided with mutually inhibitory links. The levels of activation of the collectors of a predicate results from the activation 64 Both human and model predictions were conducted at four different stages of the input sentence: verb (the crook arrested), by (the crook arrested by), the (the crook arrested by the), and agent NP (the crook arrested by the detective). 71 Grounding Meaning in Everyday Experience in the World incident on the collectors from the rest of the network and the mutual inhibition between the two collectors. When either collector is active, the answer to the search question is uncertain, whereas when both collectors receive a degree of activation which goes over a threshold specified in an additional node, then a contradiction is detected. The weighted links between the collectors and the enabler of a predicate convert a dynamic assertion into a query about the assertion, so that the system can constantly evaluate incoming knowledge against the background of existing knowledge (the higher the weight, the more skeptical and “conservative” the system is), allowing us to detect potential and contradictions. The role of these links is to create positive feedback loops of spreading activation and then creative stable coalitions of active nodes under approapriate circumstances. Dynamic bindings are represented by the synchronous firings of appropriate role and filler nodes. For instance, in the dynamic fact love(Mary, Tom), the lover node is firing in synchrony with the Mary node, and the lovee node is firing in synchrony with the Tom node. The enabler node is also firing, so the system is asking if it believes that Mary loves Tom. A long-term fact behaves like a temporal pattern matcher, which is activated when the static binding it encodes matches the dynamic bindings represented in the system's state of activation. For instance, the long-term fact love(Mary, Tom) is encoded in a node which is linked to the enabler node of the associated predicate and sends a link to the positive or negative collector of the predicate depending on whether the fact encodes a positive or negative response. Rules are encoded by (i) linking the roles of the antecedent and consequent predicates so as to reflect the correspondence between these roles specified by the rule; (ii) connecting the enabler of the consequent predicate to the enabler of the antecedent predicate, and (iii) connecting the appropriate collectors of the antecedent predicates to the appropriate collector of the consequent predicate. The encoding of rules exploits weighted links between predicates. These weights distinguish categorical from soft rules and also lead to a gradual weakening of activation along a chain of inference. Eventually the chain of inference terminates when activation falls below a threshold. For instance, Shastri and Grannes invite us to consider that the long-term memory of the system includes the rule that a bachelor is not married: (bachelor(x) → ˥married(x, y)). The predicate bachelor is represented in a node cluster including an enabler node, two collector nodes and a role node, while the predicate married is represented in a cluster node including an enabler node, two collector nodes, and two role nodes (one labeled husband, and the other wife). Bachelor's positive collector fires in synchronization to married's negative collector, whose activation is propagated to its enabler collector. Bachelor's role node fires simultaneously with married's role node (filled by the husband value). Suppose also that the system knows that John is a bachelor (bachelor(John)). If the system is then told that John is married to Susan, the positive collector of married is activated, dynamically binding husband to John and wife to Susan. The activation propagates to the enabler node of married and then to the enabler role of bachelor. Now, the fact that John is a bachelor matches the dynamic binding at bachelor, activating the corresponding positive collector node. The activation propagates to the negative collector of married. Therefore, both collectors of married are activated, detecting the contradiction between the agent's existing beliefs and the new information. Contradictions can of course also arise between existent knowledge of different facts stored in long-term memory. Moreover, in order to be consistent with the human nature, 72 2. From Cognition to Grammar the model must sometimes overlook relevant information and make some mistakes, due to the limitedness of its resources. Despite these limitations, an appropriate cue may make the necessary information available and lead to a correct response. The model we have just briefly sketched is a (relatively) simple one, consisting in an extension of the Shruti system to negative evidence, but the system can be further extended in several ways to cover a broader range of phenomena (see e.g. Shastri and Wendelken 2000; Wendelken and Shastri 2005). Nevertheless, our goal here was just to give an idea of how Shruti works and we do not need to delve into the functioning of the system any deeper. For the scope of the present study, the reader is just required to remember the importance of bindings in this computational model of inference, in particular the connection between existent knowledge stored in long-term memory and incoming information. Indeed, even though we shall not be particularly focused on inferences in the present study, when applying ECG to the analysis of Italian utterances they will be mentioned frequently. As a result, it seemed appropriate to destinate a few words in this section to a brief sketch of how inferential reasoning is considered to work by scholars involved in the NTL program, including proponents of ECG65. 2.3. Embodied Construction Grammar: setting the scene In the present section, we are going to introduce the constructionist model that will be used in our analysis in ch. 3. This section is divided in two subsection: in §2.3.1, we are going to briefly outline the most important characteristics of cognitive and constructionist approaches to grammar (especially in comparison with the mainstream generative model), also briefly introducing three among the most developed and well known models which had a particular importance for the theoretical foundations of ECG. Then, we shall also briefly summarize the main commonalities and divergences between these three models. In §2.3.2 we shall concisely situate ECG in the context of cognitive approaches to grammar and provide the reader with a brief explanation of the functioning of the ECG model, along with a simple illustration of such a model at work. 2.3.1. Cognitive approaches to grammar The label “cognitive approaches to grammar” serves as a very large “umbrella” that covers a number of approaches to the scientific study of language which share some basic tenets. First of all, it must be stressed that all of these approaches were developed within the Cognitive Linguistics enterprise, then representing a reaction against mainstream Generative Grammar66. Therefore, scholars working on these approaches see language as an integrated branch of cognition rather than an encapsulated autonomous faculty: language is considered to be governed by the same general cognitive principles which govern other facets of human cognition67. 65 The importance of this point is also emphasized in Bergen and Chang (2005). 66 For a fine-grained yet highly accessible criticism of the Chomskyan paradigm, the reader is advised to see Dąbrowska (2004: chs. 1-9). A concise exposition of arguments against Generative Grammar (with a special focus on linguistic nativism) is also offered in Lombardi Vallauri (2009). 67 However, recent developments seem to show that an interaction between cognitive and “autonomous” 73 Grounding Meaning in Everyday Experience in the World Here, we are going to briefly address some of the main differences in the theoretical assumptions and working mechanisms between generative and cognitive approaches, some of which were briefly addressed in §2.1.268. First of all, an important point of divergence between these two traditions regards the fact that while generative linguists posit the existence of an innate computational system that derives a well-formed grammatical structure without recourse to meaning, cognitive linguists posit an inventory of symbolic units containing schematic templates, which emerge as a result of regular use. A speaker interpreting new structures will compare the new structure with existing template, taking into account the communicative goals, the context, etc. While Generative Grammar captures generalization and defines well-formedness as the output of the application of rules, Cognitive Linguistics captures generalizations and defines well-formedness as the results of a categorization process. Then, generative linguists make use of derivational rules which precede and thus determine the specific expressions that instantiate them. As pointed out by Evans and Green (2006), within the generative framework, lexical items are stored in the lexicon together with information about their phonology, semantic, and core syntactic properties (e.g. word class). As a result of its interaction with generalized syntactic principles, this information gives rise to deep structures: syntactic structures in which the core requirements of the lexical items are satisfied in accordance with the syntactic principles. Deep structures typically correspond to unmarked active declarative sentences, which are often viewed as the basic syntactic structures within a given language. Less basic clause types, such as interrogatives, are then derived from deep structures by means of syntactic transformation, giving rise to surface structures. Consider the following simple examples of an interrogative and a declarative transitive sentences, respectively: (19) (a) (b) Is Penny dating someone? Penny is dating someone. Generative linguists consider the structure in (19a) as a surface structure derived from the deep structure in (19b) by means of the application of a rule which raises the auxiliary verb to a position in front of the subject (see e.g. Graffi 1994), which might be formulated in (20a) below, given the deep structure in (20b): (20) (a) (b) Derive interrogative from declarative: Move AUX in front of SUBJ SUBJ AUX V-ING OBJ On the other hand, cognitive linguists conceive language as a structured (i.e. not random) inventory of conventional pairs of form and meaning (or form and function), called constructions in the various versions of Construction Grammar (CxG henceforth), and symbolic assemblies in Cognitive Grammar69. These symbolic units views of language may be possible and, to a certain extent, fruitful. For an update outline of the situation, the reader is referred to Taylor (2007). 68 It may be relevant to emphasize the fact that several different approaches to grammar exist in the generative enterprise. Here, we shall just make reference to the transformational model for two main reasons: first, as underlined by Evans and Green (2006: 747), it is the most prominent generative model; second, it is the model against which cognitive approaches to grammar defined themselves in the early stages of their development. 69 Langacker also exploits the notion of “construction”, but with a narrower scope. When describing the basic properties of Cognitive Grammar, we shall stick to Langacker's terminology, while in the rest of 74 2. From Cognition to Grammar are monostratal (i.e., they do not imply the existence of a deep structure opposed to a surface structure) and they include information regarding all aspects of language (from phonology to pragmatics). Constructions are not obtained via syntactic derivation, but are “stored whole” by the learner. Indeed, virtually all cognitive and constructionist approaches to grammar subscribe to the usage-based thesis (see e.g. Tomasello 2003; Goldberg 2006; Evans and Green 2006: ch. 14), that is, they claim that language is learned by the abstraction of constructions from real instances of language. As a consequence, in a cognitive perspective, the two sentences in (19) above are seen as instantiating two different schematic patterns previously stored whole as an effect of repeated use, each of which is associated with different semantic and pragmatic functions. These emergent patterns are illustrated in (21) below: (21) (a) (b) Interrogative pattern: AUX SUBJ V-ING OBJ Declarative pattern: SUBJ AUX V-ING OBJ Another important aspect in which generative and cognitive linguists hold diverging views regards their different views of redundancy, which is stigmatized in the Chomskyan tradition, while it is taken to be an essential feature of language in Cognitive Linguistics70. Indeed, scholars working within a generative framework distinguish between “regular” forms, which can be derived from the application of a generalized rule, and irregular forms, which need to be explicitly listed in the grammar: this distinction follows from their commitment to the idea that language must be a maximally economic system, in order to be acquired and manipulated rapidly. Cognitive linguists, on the contrary, avoid the notion of rule, preferring to recur to schemas that follow from instances: schemas represents patterns emerged from entrenched units as a consequence of usage. Consequently, generalizations are the outcome of recurring patterns of usage that allow the speaker to infer a higher-order schema. As a consequence, both the schema and instances of that schema are listed in the grammar, and the schema represents an expression of the generalization which emerges from patterns of usage. Thus, with regard to the example sentences seen above, we can say that generative linguists will consider the deep structure in (19b) as stored in the grammar, while the surface structure in (19a) will not be listed in the grammar, since it can be derived by the application of the rule in (20a). On the other hand, cognitive linguists will argue that both sentences in (19) will be stored in the grammar, together with the schematic patterns they instantiate. A further difference which is related to the one we have just sketched, regards the fact that generative linguists focus on the statement of general rules that account for well-formedness in language. As a consequence, generative grammarians are usually not concerned with conventional expressions such as by and large, all of a sudden, etc. since they often fail to conform to general patterns of syntactic structures. As Evans and Green (2006: 755) highlight, “These structures are considered peripheral and uninteresting because they do not reveal general and productive patterns. Instead, the formal model focuses upon 'core' phenomena.” On the contrary, cognitive linguists consider conventional expressions as a central part of language knowledge and use: all “regular” and “irregular” expressions are part of the speaker's inventory of symbolic the present study we shall use the term “construction” in a broader sense. 70 “Redundancy is not to be disparaged, for in one way or another every language makes extensive use of it.” (Langacker 2008: 188). 75 Grounding Meaning in Everyday Experience in the World units and so deserve to be accounted for. The difference between the two kinds of units stands on the fact that while “regular” expressions like those in (19) above show a high type-frequency, and thus their entrenchment is followed by the rise of a higher, more schematic pattern which will be productively used to create novel expressions, irregular expressions are stored but do not give rise to any schematization. In other words, they show a high token-frequency only. Finally, we can mention the fact that generative linguists view linguistic elements as having a componential structure: elements are seen as having a complex internal structures and been built from scratch. On the contrary, cognitive linguists hold that entrenched instances give rise to schemas. Nevertheless, this does not mean that they reject the view that speakers recognize complex structures as having compositional structure. Simply, Langacker proposes that component structures are immanent in the complex grammatical construction, regardless of whether the compositionality is recognized by the speaker. According to Langacker, entrenchment decreases the salience of compositionality. The compositional structure of a grammatical construction may be essential to the initial creation or construction of that expression, but once the construction is entrenched and gains the status of a unit, this compositional scaffolding is no longer required. Despite this, the compositional structure remains immanent: we may still recognize the compositionality of well-entrenched units, but it does not follow that people build them from scratch each time they use them. While we have to admit that this can just represent a concise (and by no means complete) list of the main properties shared by the members of the family of cognitive approaches to grammar, in our opinion the brief characterization just supplied above, together with the information supplied in §2.1.2, should be sufficient to provide the reader with an idea of the common assumptions which underlie the approaches developed within the Cognitive Linguistics framework and distinguish them from mainstream generative models. The identification of shared assumptions between different models within the world of cognitive approaches to grammar is important because this reality is rather fragmented and variegated. As a matter of fact, it is possible to find several different cognitive models of grammar, most of which are broadly compatible, but differ from each other for small differences in the perspective from which they deal with grammatical phenomena. To make things even more complicated, Cognitive Linguistics is generally considered very close to Functional Linguistics71; as a result, it is often not clear when to consider an approach as functional, cognitive, or both. Indeed, Cognitive Linguistics and Functional Linguistics are usually both considered as part of the broader family of functionally oriented approaches to the study of language, in a (more or less watertight) opposition to formal approaches. Basically, functionally oriented approaches share the assumption that linguistic structure cannot be analyzed independently of the uses to which it is put and, at a very general level, it is often difficult to keep them separate from each other. The basic tenets which characterize Cognitive Linguistics, in most cases, also apply to Functional Linguistics and interactions between these two fields are not rare. In particular, the field of functional typology seems to be particularly compatible with Cognitive Linguistics (see Croft 2003). To cut a long story short, we may follow Evans and Green saying that functional approaches tend to be “less concerned with the psychological representation 71 It is not chance that several scholars work in both frameworks and may be defined functional linguists as well as cognitive linguists. To mention a few (in alphabetical order): Joan Bybee, Sonia Cristofaro, William Croft, Talmy Givón, Francisco Gonzálvez García, Willem Hollmann, Paul Hopper, Silvia Luraghi, William McGregor, Jan Nuyts, Sandra Thompson, Elizabeth Closs Traugott. 76 2. From Cognition to Grammar of language as a system of knowledge and more concerned with its use.” (Evans and Green 2006: emphasis original). However, this is an oversimplification, since the boundaries between functional and cognitive approaches can hardly be drawn with a sufficient degree of certainty72. In consideration of the above, it appears obvious that a selection will be necessary, in order to avoid to engage in the discussion of dozen approaches most of which have very little to do with ECG and have exterted virtually no influence on the present study. Therefore, in the remainder of the present subsection we shall confine ourselves to summarizing the main points of three among the most developed and influential cognitive approaches to grammar, which represent the linguistic part of the theoretical background the ECG model stands on: Langacker's Cognitive Grammar, Goldberg's Construction Grammar, and Croft's Radical Construction Grammar73. Of course, our short overview of these models is not meant to do justice to the work of those scholars who have been working on them for years or even decades: we are fully conscious of the fact that several important notions will have to be oversimplified, and many significant concepts will unfortunately be completely left out from the exposition. Nevertheless, the goal of the outline presented below is just to provide the reader with a general idea of how these approaches work and to point their attention to some aspects of these cognitive models which will probably be useful to remember while pursuing the reading of the present study74. 2.3.1.1. Cognitive Grammar: exploring the semantic basis of grammar Langacker's Cognitive Grammar (CG henceforth) is the oldest, and by far the most detailed and influential model in the world of cognitive approaches to grammar. While Langacker (among with other scholars) spent several decades working on the development of CG, in the present exposition we shall primarily refer to his recent introductory book (Langacker 2008, to which the reader is advised to make reference for a deep but accessible illustration of the mechanisms of CG), for the sake of both clarity and space. Consistent with the tenets of Cognitive Linguistics, CG is designed to be a natural and psychologically plausible theory of language. In order to satisfy such requirements, Langacker claims that a language needs only three kinds of structures: • semantic structures, conceptualizations exploited as the meanings of expressions, e.g. the concept “cat” (felis catus); • phonological structures, symbolic representations of meanings (i.e. sounds, gestures, or orthographic signs), e.g. the phonological string /kæt/; • symbolic structures, bipolar structures which incorporate a semantic structure and a phonological one, e.g. the word cat, which pairs the phonological string /kæt/ and the concept “cat”. 72 For a brief outline of the relation between functional and cognitive approaches to grammar, the reader is referred to Nuyts (2007). For a comparison between several functional and cognitive models, see Gonzálvez García (2009a, 2009b). An example of a hybrid model is concisely presented in Ruiz de Mendoza Ibáñez (2010). 73 The reader is also advised to see Steels and De Beule (2006) for an overview of Fluid Construction Grammar, a cognitive model of grammar supported by engineering technical support. 74 A more detailed illustration of the models briefly outlined below can be found in Evans and Green (2006: chs. 14-20). See also Croft and Cruse (2004: ch. 10); Broccias (2006). 77 Grounding Meaning in Everyday Experience in the World It is possible to combine two or more symbolic structures to form a higher-level one. Both lower-level and higher-level structures constitute a symbolic assembly75. CG considers the lexicon as the set of fixed (i.e. familiar and conventional) expressions in a language, which is not equivalent to the set of its words 76. An important characteristic of CG is the absence of a clear-cut distinction between lexical and nonlexical items, since familiarity and conventionality are a matter of degree. Langacker claims that is possible to notice in the lexicon the following basic phenomena observable in other facets of cognition: • association, the establishing of psychological connections with the potential to influence subsequent processing; • automatization, the mastering of complex structures through repetition, to the point that using them requires little conscious monitoring: through progressive entrenchment, a structure eventually becomes established as a unit. Unit status does not entail the absence or unimportance of components, merely the routinized nature of their execution; • schematization, the process of extracting the commonality inherent in multiple experiences to arrive at a conception representing a higher level of abstraction. Schematization plays a role in the acquisition of lexical unit, since their conventional forms and meanings are less specific than the usage events on the basis of which they are learned. For instance, Langacker points out that the basic sense of the English word ring (“circular piece of jewelry worn on the finger”) is schematic relative to the conception of specific rings in specific contexts. Schematization can be carried to different degrees, depending on the diversity of the elements it is based on. Because ring is also used for adornments in other places than the finger, it is possible to posit a more schematic value (e.g. “circular piece of jewelry worn on the body”). The latter meaning represents an elaboration or instantiation of the former. Ring can then also assume still more abstract meanings (e.g. “circular object”, “circular entities”, etc.); • categorization is most broadly describable as the interpretation of experience with respect to previously existing structures. A category is a set of elements considered equivalent for some purpose. If a structure A belongs to a category, it can be used to categorize another structure B, which may then become a category member. If A is schematic for B, i.e. B elaborates A, categorization is straightforward: for instance, if “cat” is a member of the category “feline”, and “Maine Coon” is a member of the category “cat”, then “Maine Coon” is a member of the category “feline”. However, B may also conflict with A's specifications, and still be assimilated to the category on the basis of an association or perceived similarity. A is then a prototype, and B is an extension from it (e.g. the word ring can be applied to rectangular arenas, as used in sports such as boxing, wrestling, and so on). In CG, there is no clear demarcation between lexicon and grammar. Symbolic assemblies are supposed to vary along three parameters: • symbolic complexity, e.g. moonless is more complex than moon; • schematicity / specificity, depending on the precision and detail of their 75 The morpheme will therefore represent the simplest kind of symbolic assembly. 76 As Langacker (2008: 16) points out, “there are fixed expressions larger than words (like moonless night), and there are possible words – such as dollarless – that are novel rather than familiar and conventionally established” (italics and emphasis original). 78 2. From Cognition to Grammar characterization: dog is more specific than animal, but less specific than pitbull; • entrenchment / conventionality, the extent to which they achieve the status of unit and become conventional within a speech community. According to Langacker, specificity is the key parameter to draw a distinction between lexicon and grammar: “lexicon can be characterized as residing in fairly specific symbolic assemblies, grammar in more schematic ones.” (Langacker 2008: 22, emphasis original). Therefore, lexicon and grammar constitute a continuum along which the various items are distributed according to their degree of schematicity / specificity; toward the two extremes we expect to find cases easily identifiable as lexical (e.g. dog) or grammatical (e.g. the English interrogative pattern), while there will be many structures lying in between. Then, a speaker's knowledge of grammatical patterns resides in a structured inventory of symbolic assemblies, abstracted from real occurrences, and once they are established as units they can function as models guiding the formation of new expressions following the same pattern. For instance, Langacker asserts that, once abstracted from common expressions such as moonless night, childless couple, or hopeless situation, the schematic assembly N1+less N2 is available to license novel expressions, e.g. dollarless surgeon. The analyst working in the CG framework is subjected to the severe constraints posited by the content requirement, which Langacker sums up in the following few lines: ...the only elements ascribable to a linguistic system are (i) semantic, phonological, and symbolic structures that actually occur as parts of expressions; (ii) schematizations of permitted structures; and (iii) categorizing relationships between permitted structures. The thrust of the content requirement is that the linguistic knowledge we ascribe to speakers should be limited to elements of form and meaning found in actually occurring expressions, or which derive from such elements via (...) association, automatization, schematization, and categorization. (Langacker 2008: 25, emphasis original) Langacker highlights the fact that provision (i) allows us to posit specific elements sufficiently frequent to become entrenched as units77, while provision (ii) allows schematizations (which can distinguish different levels of abstraction) of these elements, and provision (iii) permits the analyst to posit categorizing relationships, such as those between schemas and their instantiations (see below). The content requirement applies to both phonological and semantic structures. Langacker states that “grammar resides in vast networks of symbolic assemblies (...) with varying degrees of abstraction and symbolic complexity” (Langacker 2008: 26). The content requirement is very important because it rules out the use of any kind of generative devices used in formal linguistics, allowing the analyst to stick to the central assumption of CG according to which languages are learned on the basis of real occurrences. In CG, the long-termed distinction between linguistic meaning and contextual knowledge is avoided, since Langacker considers such a division to be artificial and untenable. CG views meaning as consisting of some content and a way to construe that content (i.e., the ability to conceive the same situation in different ways). The distinction between content and construal is not clear-cut. Meaning (especially content) is defined and analyzed on the basis of a series of notions developed within the Cognitive Linguistics enterprise, by scholars (including Langacker himself) working in the field of 77 We suppose that this “sufficient frequency” will be different depending on the single symbolic assembly. 79 Grounding Meaning in Everyday Experience in the World cognitive semantics: frames, prototypes, schemas, networks, metonymies, metaphors, blends, conceptual hierarchies, and so on78. Moreover, seeing semantics and pragmatics as a continuum, CG analyzes meaning (in particular construal) making use of some notions developed by scholars working on pragmatics and information structure, such as focus, topic, foreground, background, givenness, etc.79 Exploiting notions from these two branches of linguistics, CG is able to adopt a rich, dynamic, and psychologically plausible notion of meaning. As an example, we shall now sketch the way in which CG fruitfully exploits the notion of prominence. According to Langacker (2008: 66), language structure “displays numerous kinds of asymmetries that are reasonalbe considered matters of prominence”. In particular, he focuses on two specific kinds of prominence: profiling and trajector/landmark alignment, which both involve the focusing of attention. As the basis for its meaning, an expression selects a certain body of conceptual content, which Langacker labels conceptual base, which can be seen as the general locus of attention. Within this onstage region, attention is direct to a particular substructure, called the profile, which stands on as the specific focus of attention within its immediate scope. Consider for instance the English kin term aunt. The essential content of this lexeme is the kinship relation between a female and a reference individual. It is this relationship that is critical for characterizing the female in question; nevertheless, the word does not profile the relationship but the female person it serves to identify. When a relationship is profiled, varying degrees of prominence are conferred on its participants. The most prominent participant, called the trajector, is the entity construed as being located, evaluated, or described: the primary focus within the profiled relationship. Often, some other participant is made prominent as the secondary focus: this is labeled the landmark. Expressions can have the same content, and profile the same relationship, but differ in meaning because they make different choice of trajector and landmark. Langacker provides an example of this situation illustrating the contrast between the prepositions above and below. Consider the two sentences below: (22) (a) (b) The lamp is above the table. The table is below the lamp. The relationship profiled by the two clauses is referentially the same, but there is a semantic contrast between (22a) and (22b) residing in the degree of prominence conferred on the relational participants: X above Y specifies the location of X (the higher participant), while Y below X locate Y (the lower participant). In both cases the other participant serves as a spatial landmark for that purpose80. CG, overtly challenging a traditional belief largely shared among modern linguists81, 78 A detailed description of all these notions (and many more) can be found in Evans and Green (2006: part I). The reader is also advised to see Ungerer and Schmid (2006). 79 Langacker makes particular reference to Lambrecht's (1994) influential piece of work. An accessible introduction to pragmatics can be found in Huang (2007). 80 It is important to stress the fact that the notions of trajector and landmark are defined in terms of primary and secondary focal prominence, not in terms of any specific conceptual content. Consequently, these notions are not limited to the domain of spatial relationships, but they can be applied to any kind of cognitive domain. 81 A view for which Langacker (2008: 98) blames what he calls “objectivist semantics, the identification of meaning with objective features of the situation described”. This dominant position in studies on semantics seems to be an instance of the objectivist philosophical viewpoint radically criticized by 80 2. From Cognition to Grammar considers that basic grammatical classes can be defined semantically. The traditional hostility towards a meaning-based definition of word classes is based on the rejection of inaccurate proposal for a semantic categorization of parts of speech. Langacker asserts that ruling out the possibility to claim that nouns denote “objects” and verbs denote “actions” does not imply that a semantic definition of parts of speech is impossible. According to Langacker, such an argument against a conceptual approach to grammar is simplistic, for it considers only a limited class of concepts representing a particular level of generality. Labels such as “objects” and “actions” are general, but not the most schematic conceptions possible. Langacker (2008: 94) points out that these notions represent “experientially grounded conceptual archetypes” (emphasis original), being appropriate as the prototypes for linguistic categories82: objects and actions, for instance, can be seen as prototypical for the noun and verb categories, but it is necessary to find schematic definitions satisfied by all members of these categories, rather than central members only. Definitions adequate for all class members should be more abstract than the archetypal notions mentioned above. According to Langacker, this goal can be reached taking cognition into consideration. For instance, he claims that it is necessary to recognize the role of conceptual reification, our conceptual capacity for construing events as abstract objects. Langacker claims that the verb explode and the noun explosion can both refer to the same event. Nevertheless, this does not make the two expressions semantically equivalent: they differ in meaning with regard to how they construe it: explode directly reflects the event's processual nature, whereas explosion construes it as an abstract thing derived by conceptual reification. This conceptual contrast makes the two expressions belong to different categories. Nevertheless, Langacker also underlines that this schematic characterization which subsume the large array of entities designated by a grammatical class is possible only provided that such a class can be considered universal and fundamental83. As Langacker (2008: 96) points out, “A pivotal issue concerning grammatical categories is how they relate to grammatical constructions.” Some approaches developed within the Cognitive Linguistics enterprise propose that constructions (rather than categories) are the basic units of linguistic structures (see e.g. Croft 2001). From this perspective, every construction in a language defines a category, specific to the language, consisting of just those elements that occur in it. As a consequence, a descriptive framework must allow one to specify the range of elements that appear in a given construction. In order to reach this goal, CG accomodates the classes impicitly defined by occurrence in particular constructions. Across languages, a certain array of notions are especially prevalent in characterizing grammatical behavior and contributing to the semantic cohesiveness of constructionbased classes. Particular conceptions have sufficient cognitive salience that they are often invoked for grammatical purposes, inducing classes to coalesce around them. Their degree of cognitive salience determines to what extent the corresponding classes are universal and fundamental. Langacker claims that the most universal and fundamental categories coalesce around a highly salient conceptual archetype, as well as a basic cognitive ability, initially manifested in the archetype and responsible for its emergence. The former functions as category prototype, while the latter provides its schematic characterization. With regard to nouns, for instance, the archetype is the Lakoff and Johnson (1999). 82 On prototypes in Cognitive Linguistics, see Taylor (2003: especially chs. 3 and 4). 83 These two parameters refer to how many languages and how many constructions figure in, respectively. 81 Grounding Meaning in Everyday Experience in the World conception of a physical object, which emerges due to the basic ability Langacker refers to as conceptual reification. Because cognitive salience is a matter of degree, CG does not posit any fixed inventory of universal categories. Instead, universality is seen as a scale whose choice of a cut-off point would be arbitrary. We have already mentioned the fact that, according to Langacker, a semantic categorization of grammatical classes is possible taking into consideration the central role of cognition (with particular reference to construal) in the process – how situations are apprehended and conceptualized. Especially relevant to this goal are two aspects of construal: profiling and level of specificity. The latter is important since, as already highlighted, general definitions can only be found at a high level of schematicity. Profiling is critically important because “what determines an expression's grammatical category is not its overall conceptual content, but the nature of its profile in particular.” (Langacker 2008: 98, emphasis original). The profile represents what an expression designates, it is the focus of attention within the context evoked. Therefore, it should be awarded a determining role in categorization. In order to define basic categories, Langacker exploits a term which is maximally general in its application, the word entity. This term applies to anything that might be conceived of or referred to in describing conceptual structure and is not reserved to discrete and separately recognized entity. Entities can be divided into things and relationships. “Thing” is a technical term, not limited to physical objects but extended to any product of conceptual reification and, as pointed out by Broccias (2006: 91), in general corresponds to the semantic pole of the noun class. The technical term “relationship” is also used in an abstract technical sense, and it denotes connections between entities. Langacker emphasizes that various kinds of relationships can be distinguished and used to characterize basic categories. Most fundamental is the distinction between processes and atemporal relations. The former develop through time and the different facets of a scene are conceptualized successively, and they represent the semantic pole of verbs, whereas atemporal relations denote configurations fully manifested at a single instant, which can be of two types: stative relations, which involve a stable configuration through time and correspond to the semantic pole of adjectives, adverbs, and static prepositions, and complex atemporal relations, which are made up of more than one configuration over time but are viewed holistically, so that their temporal evolution is backgrounded, and they correspond to the semantic pole of dynamic prepositions. Since, for space's sake, we cannot provide a detailed description of conceptual characterization in CG, we shall recur to an example supplied by Langacker (2008: 100101), in order to illustrate the importance of construal in defining grammatical categories: let us consider the verb choose together with the derived nouns chooser and choice. As a verb, choose profiles a process, designating the relationship between a trajector, the one who makes a choice, and a landmark, the entity chosen. The trajector engages in mental activity in order to select the landmark from a range of alternatives. The event unfolds through time. The nouns chooser and choice derive from choose and evoke the process it designates as their conceptual base. Langacker underlines that they are nouns exactly because their derivation consists in shifting the profile from the process per se to a thing characterized in relation to it. With regard to chooser, the profiled thing is the one who does the choosing (the verb's trajector). Choice, on the other hand, has three basic meanings, illustrated by the sentences in the example in (23) below (from Langacker 2008: 101): 82 2. From Cognition to Grammar (23) (a) (b) (c) Unfortunately, their top choice proved incapable of doing the job. They offer a wide choice of investment options. She made her choice in just a few seconds. While (23a) profiles the thing chosen (the processual landmark), the second clause profiles the range of options, and the third sentence profiles an abstract thing obtained by conceptual reification of the base process. Finally, there is another important notion in Langacker's model we find relevant to briefly outline here, the notion of construction. Langacker (2008: 161) points out that constructions are complex symbolic assemblies, to some degree analyzable into smaller symbolic elements. Thus, grammar consists of the patterns for constructing such expressions, and both the expressions and the patterns are referred to as constructions. A symbolic structure is a bipolar structure constituted by a phonological structure and a semantic structure. Two or more symbolic structures combine with each other to form more elaborate structures. Both component structures and the composite structure are symbolic assemblies. Langacker mentions the example of the NP jar lid, where the component expressions jar and lid are integrated to give rise to a composite expression. The component and the composite structures each consist of a semantic and a phonological pole, connected by a relationship of symbolization (i.e., the one stands for the other). At each pole, the component structures participate in a relationship of integration with one another and relationships of composition with regard to the composite structure. Following Langacker, we can first consider the semantic structures “jar” and “lid”. “Jar” profiles a thing, further identified as a physical container open at the top; on the other hand, “lid” profiles another thing, further identified as a cover for an opening in the upper side of a container. “Lid” itself does not specify the kind of container, evoking it schematically. The integration between the semantic structures “jar” and “lid” indicates that the container profiled by “jar” corresponds to the schematic container profiled by “lid”. Similarly, the phonological pole /ʤɑ:r/ and the phonological pole /lɪd/ integrate with each other occurring at a certain point in speech time: “One aspect of the characterization of a word is the potential for other words to precede or follow it along this axis” (Langacker 2008: 163), hence the merger of the corresponding elements produced the word sequence jar lid at the composite-structure level. Composition consists in the combination of component structures consistent with the correspondences holding between their elements at the semantic and phonological poles. Component-structure elements that correspond to one another correspond to the same element at the composite-structure level. Langacker then states that the composite semantic structure profiles the cover for a container identified as a jar, rather than schematically as a container. The composite phonological structure is the two-word sequence jar lid, with the first word bearing stress (/ˈʤɑ:r lɪd/). It is important to note that the composite structure is not simply the sum of the component structures, at either pole; on the contrary, it is a symbolic structure in its own right, usually showing emergent properties, not inherited from their components. For instance, the facts that the expression jar lid profiles a lid rather than a jar and that the stress falls on the first word are not predictable from the nature of the component structures. Accordingly, while component structures are seen as motivating the composite structure to varying degrees and usually supplying most of its content, they are not building blocks which are put together to form the composite whole. Grammar has conventional patterns for putting symbolic assemblies together. These 83 Grounding Meaning in Everyday Experience in the World patterns are similar to the complex constructions they characterize, but they are schematic rather than specific. They are labeled constructional schemas. Schemas are acquired through a process of schematization, being abstracted from real expressions as skeletal representations of shared organizational features. Once learned, a schema serves as a template to create novel expressions on the same pattern. For instance, jar lid instantiates a basic compounding pattern of English: at the phonological pole we have two words, while at the semantic pole we have two things being profiled. The pattern also specifies that the first word bears stress and that the second component determines profiling. The specifications are incorporated in the constructional schema the expressions all instantiate. When an expression completely conforms to the schema’s specification, it is said to elaborate the schema, whereas when there are some discrepancies between their specifications, the relationship is labeled as one of extension. In either case, the schema is better seen as lying within its instantiations, rather than being wholly distinct from them. Concluding this subsection, it seems relevant to remind the reader that, despite our necessarily brief characterization presented here, CG represents the most comprehensive theory of grammar developed within the Cognitive Linguistics framework, and to a large extent it has provided the (especially semantic) background for the emergence of other cognitive approaches grammar84. 2.3.1.2. Goldberg’s Construction Grammar: emphasis on argument structure Goldberg’s Construction Grammar (CxGx hereafter) represents the first version of CxG developed adopting a prominently cognitive approach85. While nowadays a lot of CxGx work has been carried out by several scholars, the reference text for this model remains to be found in Goldberg (1995), generally seen as a sort of “birth certificate” for CxGx. In Goldberg (1995), the author proposes an approach to grammar which exploits and elaborates insights from previous theories, especially Lakoff’s (1987: case study 3) cognitive theory of grammatical constructions, and Fillmore and Kay’s Construction Grammar, along with Langacker’s model outlined above. In her book, Goldberg circumscribes the label constructions to cover combinations of syntactic phrases only86. Moreover, this label just refers to those symbolic units whose meaning is not wholly predictable from the meaning of its component parts. For instance, consider the simple clause illustrated below: (24) Sam sneezed the napkin off the table. In CxGx, the meaning of the sentence will be determined by the interaction of the semantics of the verb and the meaning of the construction. It is important to specify that meaning in CxGx is understood in terms of frame semantics (see §2.1.2), therefore, lexical meanings are defined relative to some particular background frame. Goldberg, 84 In particular, in §2.3.2 and in the next chapter, we shall see that ECG often exploits semantic notions from CG. 85 To be precise, Fillmore and Kay’s Construction Grammar model was worked out several years earlier (see for instance Fillmore et al. 1988), but that could hardly be defined as a cognitive model, since it adopted a broadly generative (though not transformational) orientation. 86 Combinations of morphemes were not originally allowed the status of constructions. 84 2. From Cognition to Grammar for instance, illustrates the difference between the English words land and ground: though both of them describe a solid ground, the former is used to denote it as opposed to the sea, whereas the latter is used to denote it as opposed to air. As a result, these terms are distinguished primarily on the basis of the frames in which they are defined. Like nouns, verbs involve frame-semantic meanings, and therefore their designation requires reference to a background frame rich with world and cultural knowledge87. Goldberg points out that many scholars reject this view and use semantic decompositional structures like X CAUSES Y TO MOVE Z to paraphrase the meaning of verbs, claiming that they capture “syntactically relevant aspects of verb meaning” (i.e. linking rules between theta-roles and lexical entries which serve to determine the syntactic expression of arguments). In CxGx, these structures correspond to constructional meaning. Since the mapping between syntax and semantics is done via constructions (and only in limiting cases do verbs show such skeletal meaning, see e.g. give, make, become, receive, do, and so on), what are often referred to as “syntactically relevant aspects of verb meaning” are actually aspects of constructional meaning. Furthermore, Goldberg asserts that the distinction between the semantics of constructions from the verbs which instantiate them and the possibility for verbs to be associated with rich frame-semantic meanings are motivated by the need to account for novel uses of verbs in particular constructions. For instance, in order to interpret the sentence in (24) above, it will be necessary to know that sneezing involves the forceful expulsion of air from the mouth, and this would not be captured by a skeletal decompositional entry88. With regard to constructional meaning, in CxGx a role of paramount importance is assigned to argument structure constructions (A-S constructions from now on), i.e. a special class of constructions which provides the basic means of clausal expressions in a language and specifies the arguments that a verb can be combined with. As an example, the English caused-motion construction requires that the verb be combined with three arguments: a causer, a theme, and a goal. These arguments will be syntactically realized as a subject, a direct object, and an oblique, respectively. Moreover, A-S constructions must clarify how verbs will combine with them, as Goldberg points out: Constructions must specify in which ways verbs will combine with them: they need to be able to constrain the class of verbs that can be integrated with them in various ways (…) and they must also specify the way in which the event type designated by the verb is integrated into the event type designated by the construction. (Goldberg 1995: 49) The meaning of the sentence illustrated in (24) will therefore be provided by the interaction between the semantics of the verb sneeze and the meaning of the causedmotion construction. The rich semantics of a verb includes the delimitation of participant roles, which are instances of the more general argument roles associated with the construction. Verbs lexically determine which aspects of their frame-semantic knowledge are obligatorily profiled. Lexically profiled roles are entities in the frame semantics associated with the verb that are obligatorily accessed and function as focal 87 As an example, consider the verb marry: its meaning includes the knowledge of a ritualized ceremony which involves two partners, results in a change of their legal status, and supposes their intention to engage in conjugal relations and remain with each other until one of them dies. See Goldberg (1995: 27-28) for more instances. 88 More reasons to adopt a frame semantics approach to verb meaning are listed in Goldberg (1995: 2830). 85 Grounding Meaning in Everyday Experience in the World points within the scene, achieving a special degree of prominence89. These profiled participant roles correspond to those participants which are obligatorily brought into perspective. Goldberg (1995: 44) highlights that “profiling is lexically determined and highly conventionalized – it cannot be altered by context.” Sometimes, differences in profiling captures the foremost difference between the meanings of two verbs, as in the case of buy (which profiles the buyer and the goods) and sell (which profiles the seller and the goods). Phrasal constructions profile every argument role which is linked to a direct grammatical relation: Subj, Obj, or Obj2. Direct grammatical functions profile particular roles as being either semantically or pragmatically salient. Goldberg underlines the importance to understand that the profiling of argument roles are of a different kind than the profiling of participant roles: the criterion for deciding which of a verb’s participant roles are profiled in that all and only obligatorily expressed participant roles are profiled, whereas all and only argument roles which are expressed as direct grammatical relations are profiled. If a verb is a member of a verb class that is conventionally associated with a construction, then the participant role of the verb may be semantically fused with argument roles of the argument structure construction. The possibility of roles fusing is determined by whether they are of compatible types. Which participant role is fused with which argument roles is determined by two principles: • The Semantic Coherence Principle: only roles which are semantically compatible can be fused90; • The Correspondence Principle: each participant role which is lexically profiled and expressed must be fused with a profiled argument role of the construction. For instance, let us consider the English caused-motion construction, which shows the syntactic structure shown in (25), the basic meaning reported in (26), and the argument structure illustrated in (27) below: (25) (26) (27) Subj V Obj Oblpath/loc X causes Y to move Zpath CAUSE-MOVE (causer theme path) Only the causer role must be fused with a participant role of the verb: the remaining two can be contributed by the construction. Indeed, if we look back to our example in (24) above, we shall notice that the verb sneeze has only one participant role, sneezer. Sneezer is clearly compatible with the construction’s causer role, so they can be fused and will have the syntactic realization of subject (in the case of our example, Sam), the goal and the theme participant are provided by the construction and will be realized as an oblique (off the table) and a direct object (the napkin), respectively. The interplay between the frame evoked by the verb and the schematic meaning of the A-S construction gives rise to the meaning of the construction as a whole. It is relevant for us to underline two particular features of constructional meaning in a CxGx perspective. As already pointed out in §2.1.2, constructions are considered to display polysemy: a construction is typically associated with a family of closely related senses rather than a single, fixed, abstract sense. For instance, the English caused89 The notion of “profiling” is borrowed from Langacker (1987). 90 Two roles are semantically compatible when one can be construed as an instance of the other. E.g., the kicker participant of the kick frame may be fused with the agent role of the ditransitive construction. 86 2. From Cognition to Grammar motion construction typically imply that the causer argument directly causes the theme argument to move along a path designated by the directional phrase, as in the case of the example sentences in (24). Nevertheless, while this can be considered the central meaning of the construction, less typical examples of caused-motion construction display different kinds of meaning. Goldberg (1995: ch. 7) identifies at least five different senses related with this construction, listed in (28) below: (28) (a) SENSE1: CAUSER successfully causes THEME to move GOAL verbs of force-exertion (e.g. push, kick, sneeze, shove) e.g. Frank pushed it into the box INSTANTIATED BY: SENSE2: INSTANTIATED BY: (b) conditions of satisfaction implies that CAUSER causes THEME to move GOAL verbs of saying with associated satisfaction conditions (e.g. ask, order, send) e.g. Sam asked him into the room (c) SENSE3: CAUSER enables THEME to move GOAL verbs of enablement (e.g. allow, let) e.g. Sam let Bill into the room INSTANTIATED BY: SENSE4: CAUSER (d) prevents THEME from moving GOAL verbs of blocking (e.g. lock, keep, barricade) e.g. Harry locked John into the room INSTANTIATED BY: SENSE5: CAUSER (e) helps THEME to move GOAL verbs of helping (e.g. help, assist, guide, show, walk) e.g. Sam helped him into the car INSTANTIATED BY: The senses listed in (28b-e) are taken to be systematic patterns of extention from the central sense. While the CxGx's claim for constructional polysemy is also adopted in other cognitive approaches to grammar, it is particular important in Goldberg's model. Indeed, CxGx maximizes it, at the expenses of lexical polysemy: whenever a verb shows systematic differences in meaning when it appears in different constructions, such differences are ascribed to the particular construction. As a matter of fact, Goldberg asserts that what is stored (through usage) is the knowledge that a particular verb with its inherent meaning can appear in a particular construction, and it is the composite structure of verb and construction which determines the meaning of an expression. Consider the verb slice, which can appear in a variety of constructions, as shown in the following examples (from Goldberg and Casenhiser 2006; also discussed in Broccias 2006): (29) (a) (b) (c) (d) (e) He sliced the bread. (transitive) Pat sliced the carrots into the salad. (caused-motion) Pat sliced Chris a piece of pie. (ditransitive) Emeril sliced and diced his way to stardom. (way construction) Pat sliced the box open. (resultative) Rather than positing five different verb senses, CxGx argues that the verb slice, meaning “cut with a sharp instrument”, is claimed to combine with five different A-S constructions, “all of which are independently stored in the grammar of English” (Broccias 2006: 91), in the same way outlined above with regard to the example in (24). 87 Grounding Meaning in Everyday Experience in the World The merger between the verb and each different construction gives rise to the five sentences listed in (29) above. In CxGx, A-S constructions are allowed such a prominent role because, on the basis of psycholinguistic evidence, they are supposed to have psychological reality. This kind of constructions are assumed to be directly associated to humanly relevant scenes. The label “human relevantly scenes” applies to scenes that are semantically priveleged in being basic to human experience and serve to subdivide the world into discretely classified event types (something causing something, something moving, something undergoing a change of state or location, and so on). It follows that, as already seen above, the English caused-motion construction will provide an expression with the core meaning X CAUSES Y TO MOVE Z (i.e., something causes something to change location), while the ditransitive construction will contribute the core meaning X 91 CAUSES Y TO RECEIVE Z (that is, someone causes someone to receive something) . These meanings reflect event types that show a high level of abstraction, and the interaction of the construction with the verb allows the expression as a whole to assume its specific meaning. Goldberg concisely summarize this situation in the following few lines: Particular combinations of roles which designate humanly relevant scenes are associated with argument structure constructions, which therefore serve to carve up the world into discretely classified event types. Verbs, on the other hand, Are associated with richer frame-semantic meanings (…) some cross-reference between verbs and constructions is also necessary, so verbs will in effect be annotated with information about which event types they can be associated with. (Goldberg 1995: 40) As Goldberg (1995: 67) highlights, in the CxGx approach “constructions form a network, and are linked by inheritance relations which motivate many of the properties of particular constructions. The inheritance network lets us capture generalizations across constructions while at the the same time allowing for subregularities and exceptions.” According to Goldberg, language organization rests on the four general psychological principles listed below: • the Principle of Maximized Motivation: if construction A is related to construction B syntactically, then the system of construction A is motivated to the degree that is related to construction B semantically. Following Lakoff (1987), CxGx assumes that a given construction is motivated to the degree that its structure is inherited from other constructions in language. From this perspective, the more the properties of a given category are redundant, the more it is motivated and the better it fits in the system as a whole. Therefore, an optimal system is a system that maximizes motivation. Goldberg stresses the fact that several studies of child language acquisition show that children learn the names of subordinate terms (e.g. fep-car) more easily when those terms are compounded with basic level terms that they already know (e.g. car). In other words, children learn new terms for concepts related to other, already familiar 91 On the basis of psycholinguistic evidence, Goldberg (2006: ch. 4) posits that A-S constructions show a high degree of overlapping with certain general purpose verbs. For instance, the constructional meaning of the English caused-motion constructions is related to the basic meaning of the verb move, the meaning of the ditransitive construction is convergent with the meaning of the verb give, and the meaning of the resultative constructions overlaps with that of the verb become (Cfr. Langacker's view of language as structured around conceptual archetypes). A remarkable amount of evidence for the salience of these concepts is provided by studies on first language acquisition (see e.g. Tomasello 2003). 88 2. From Cognition to Grammar concepts more easily when the new terms are systematically related to the terms for the familiar concepts; • the Principle of No Synonymy: if two constructions are syntactically distinct, they must be either semantically or pragmatically distinct (but not both). Conversely, a difference in meaning or pragmatics should lead to a difference in form. For instance, Goldberg (1995: 33) provides us with an example, comparing a sentence like Mary taught Bill French, which shows the syntactic structure Subj V Obj1 Obj2, with a clause like Mary taught French to Bill, which has the structure Subj V Obj Obl. The first is a metaphorical extension of the English ditransitive constructions, which express a successful metaphorical transfer (i.e. Bill actually learned some French), whereas the second is an instance of the dative construction, which does not imply that the transfer was successful. While the two sentences may at first glance seem synonymous, their difference in form is reflected by a difference in meaning; • the Principle of Maximized Expressive Power: the inventory of constructions is maximized for communicative purposes. As a matter of fact, a maximally expressive system would have a distinct label for every distinct item in the user's world, following the property known as isomorphism; • the Principle of Maximized Economy: due to the need for simplification, deviations from isomorphism are allowed (see e.g. the phenomena of polysemy and homonymy), in order to constrain the proliferation of constructions. This principle and the previous one mutually constrain each other. The construction network is conceived as a lattice, with constructions related by specific types of inheritance links: information is inherited from dominant nodes transitively as long as that information does not contradict information specified by nodes lower in the hierarchy, and dominated constructions inherit all the information specified contained in the dominating constructions. Each construction is fully specified, but it is also redundant to the degree that information is shared with dominating constructions. Inheritance links make explicit the particular types of relations that may hold among elements of constructions, capturing the fact that all nonconflicting information between two related constructions is shared. Important enough, in CxGx the inheritance links are treated as objects92, so they are expected to have internal structure and to be related hierarchically. Treating links as objects enables the analyst to represent the fact that extensions may be created productively. Since links are objects in the system, a type of link that recurs often throughout the grammar can be said to have a high type frequency and is therefore predicted to be productively applied to new cases which share the relevant factors associated with existing case. In other words, the existence of a construction will predict the existence of an extension related by the productive link. Goldberg distinguishes four distinct kinds of inheritance links: • polysemy links, posited to capture the nature of the semantic relations between the central sense of a construction and any extension from it. Each sense can be represented by a construction that is minimally different from that of the central sense (see the different senses of the caused-motion construction listed in (28) above); • subpart links, posited when one construction is a proper subpart of another 92 As correctly argued by Broccias (2006: 94), this is one of the important topics on which CxGx diverges from both Langacker's and Croft's models, “where the only elements postulated in grammar are the semantic and phonological poles of a construction plus their symbolic links.” 89 Grounding Meaning in Everyday Experience in the World construction and exists independently. For instance, the intransitive motion construction (e.g. The paper flew away) is related to the caused motion construction by a subpart link. The syntactic and semantic specifications of the intransitive motion construction are a subpart of the syntactic and semantic specifications of the caused-motion constructions (there is a goal role, syntactically realized as an oblique, and a theme role, which in the intransitive motion constructions is realized as a subject, rather than an object); • instance links, posited when a particular constructions is a more fully specified version of another construction. Particular lexical items which only occur in a particular construction are instances of that construction, since they lexically inherit the syntax and semantics associated with the construction. For instance, there is a special sense of drive which only occurs in the resultative construction and constrains the goal argument of the construction to mean “crazy” (e.g. Chris drove Pat over the line)93; • metaphorical extension links, posited when two constructions are related by a metaphorical mapping. Goldberg (1995: 81) asserts that “the way the dominating construction's semantics is mapped to the dominated construction's semantics is specified by a metaphor. By treating links as objects, it is possible to capture relations among systematic metaphors and ultimately relate the metaphors via an inheritance hierarchy”. As an example, the resultative construction is often seen as a metaphorical extension of the caused-motion construction (e.g. Pat hammered the metal flat). CxGx allows for multiple inheritance, i.e. a single dominated construction can inherit from two or more dominating constructions; for instance, the intransitive resultative construction (e.g. The river froze solid) inherits from both the resultative construction and the intransitive motion construction, by subpart and metaphorical link, respectively. Goldberg asserts that generalizations across constructions about links between semantics and syntactic realizations, word order facts, case-marking properties, and links between semantics and grammatical relations can all be captured by stating them at sufficiently high node in the inheritance hierarchy of constructions. Therefore, these generalizations are inherited through dominated constructions, unless in a particular construction they are overridden by construction-specific constraints94. For instance, the fact that English is an SVO language can be captured by specifying a word order constraint on the top of the lattice, at the level of the subject-predicate construction. Nevertheless, certain constructions lower in the inheritance hierarchy can override word order constraint with construction-specific constraints. Finally, it seems relevant to mention that in the years following the publication of Goldberg (1995), CxGx has undergone some refinements, most of which can be appreciated in Goldberg's (2006). The most important revision regards the definition of construction, which is no more limited to totally unpredictable symbolic units, but it is extended to fully predictable patterns, as long as they occur with sufficient frequency (therefore, a frequent expression such as I love you is a construction). Goldberg (2006) also provides a specification regarding the meaning of a clause: the author states that an actual expression involves the interaction of several different 93 Quite obviously, every construction which is dominated by another construction via an instance link at the same time dominates it by a subpart link. 94 As Goldberg (1995: 110) points out, this approach seems able to simultaneously capture generalizations and avoid ruling out exceptional cases. 90 2. From Cognition to Grammar constructions and that, besides the integration of the verb and the A-S, the pragmatic context in which it is uttered also plays a crucial role in determining its ultimate meaning. Furthermore, the author makes a distinction between constructionist theories and approaches which only allow the existence of constructions, asserting that the former share “the idea that the network of constructions captures our grammatical knowledge in toto” (Goldberg 2006: 18). However, describing Goldberg (2006) simply as an update of the author's previous book would probably be inaccurate. On the contrary, it also represents an effort to both describe the mechanisms of her model in action and make the case for the adoption of a constructionist account. Particular care is awarded to the investigation of the process of learning constructions and the nature of (linguistic and nonlinguistic) generalizations, both in child and adult language, also relying on experimental data 95. While a discussion of Goldberg's (2006) piece of work is far beyond our current purpose, it seems appropriate to recall that such a volume represents a reference-point for people interested in cognitive approaches to grammar. At present, CxGx represents the “mainstream” constructionist approach to language: it is the most developed and influential construction-based theory. Indeed, Goldberg's approach has been applied to the analysis of several phenomena in various languages (see e.g. Quochi 2007 for Italian or Perissutti 2008 for Czech), and the work carried out by Goldberg and her collaborators generally serves as reference background also for scholars involved in different CxG models. 2.3.1.3. Radical Construction Grammar: insights from typology Radical Construction Grammar (RCG henceforth) is a model of CxG which is rooted in two different approaches to the scientific study of language. On the one hand, the RCG model is inspired by a number of (both synchronic and diachronic) cross-linguistic surveys carried out in the field of language typology; on the other hand, it is considerably influenced by the work carried out by Langacker and colleagues within the CG framework. The reference text for RCG is Croft's (2001) monograph. Building on his experience as both a functional typologist and a cognitive linguist, Croft argues against both the traditional, distributional analysis of word classes, whereby words are grouped with certain classes on the basis their formal properties, and the independent existence of word classes within any language. The distributional method, which can be led back to the American structuralist tradition, defines syntactic categories on the basis of the occurrence or nonoccurrence of their members in different types of utterance, which are defined in structural terms. Moreover, the occurrence of members of a syntactic category in an utterance type is relative to the role they fill in it (e.g., in a passive sentence such as The bride was greeted by the guests, the NP the bride fills the subject role). To illustrate distributional analysis, Croft (2001: 12) offers the examples reported below: (30) (31) (a) (b) (a) Jack is cold. *Jack colds. Jack is happy. 95 Relying on data collected by linguists, psycholinguists, and cognitive psychologists, Goldberg provides strong arguments in favor of a usage-based approach to language. 91 Grounding Meaning in Everyday Experience in the World (b) (32) (33) (a) (b) (a) (b) *Jack happies. *Jack is dance. Jack dances. *Jack is sing. Jack sings In (a) sentences, the four words occur in the predicate role after the inflected copular verb be. In the (b) sentences, the four words occur in the predicate role without a copula but with the tense agreement inflection. Sentences (30b) and (31b) are not acceptable, but sentences (32b) and (33b) are. Thus, since cold and happy have the same distribution, and contrast with dance and sing which share a different distribution pattern. As a result, two classes of words (traditionally labeled “adjective” and “verb”, respectively) are distinguished, on the basis of their syntactic properties. This explanation is generally considered stronger if the category being observed occurs across many utterance types: the more utterance types appear to include a particular category as a role, the stronger the evidence for that category in the language is. The distributional method is used to identify the basic grammatical units out of which complex syntactic structures are built. Croft (2001: 29-30) argues that there at least two main problems with the distributional analysis and its application across languages. Firstly, some languages lack some of the morphological and/or syntactic features that define the distributional approach. For instance, inflectional criteria are commonly used to distinguish parts of speech, but there are languages, such as Vietnamese, which lack all morphological inflection; as a consequence, this criterion cannot be used to distinguish parts of speech in such a language. Second, even when such features exist in a given language, from a crosslinguist point of view they show too different patterns of distribution to allow the analyst to arrive at plausible distributional criteria. For instance, the author claims that Makah (a native American language) has inflections for agreement, aspect, and mood, which are used as criteria for the category “verb” in English and other European languages, but virtually all semantic classes of words can be inflected for person, aspect, and mood in Makah, including those which correspond to English verbs, nouns, adjectives, and adverbs. At the same time, Croft shows that within a single language, the different criteria used to define a particular category often have different distributional patterns as well. For instance, English direct objects (NPs that immediately follow the verb in the active voice and lack a preposition) are traditionally distinguished from obliques on the basis of their passivizability, i.e. the alleged direct object in an active sentence can become the subject of the corresponding passive sentence, while an oblique cannot. Nevertheless, there are postverbal prepositionless NPs that cannot occur as passive subjects, whereas there are oblique objects of prepositions that can. Compare the examples in (34) and (35) below (from Croft 2001: 35): (34) (35) (a) (b) (a) (b) Jack weighs 160 pounds. *160 pounds are weighed by Jack. Claude Debussy lived in this house. This house was lived in by Claude Debussy. According to Croft, the distributional method combined with the theoretical assumption 92 2. From Cognition to Grammar that the categories and relations defined by constructions are the syntactic primitives used to represent grammatical knowledge leads to circularity: constructions are used to define categories; then, the categories taken to be primitive elements of syntactic representation are in turn used to define constructions. The RCG model represents Croft's attempt to develop a radically alternative approach to the study of grammar. RCG posits that the construction is the only true universal syntactic primitive and that categories and relations are shaped by the construction they appear in, not the other way round: the grammatical knowledge of the speaker is considered to include their knowledge of words, syntactic patterns and the mapping between the two. As a consequence, in RCG syntactic categories are seen as construction-specific: Categories can be defined construction-specifically as the class of fillers of a particular role in a single construction (...) Categories can also be defined cross-constructionally, as the class of fillers that has an identical distribution across the relevant roles for all constructions of the language, or at least some specified set of constructions of the language. (Croft 2001: 46) Croft observes that either definition for syntactic categories can be analytically helpful in some cases, but what is basic are the constructions, which define the categories either individually or jointly. Moreover, since RCG finds its starting point in the typological diversity between different languages and in “the uniqueness of each language's grammar” (Croft 2001: 33), in this model constructions are language-specific. In RCG, the term construction applies to all pairings of form and meaning, from morphemes to sentences. The form pole of a construction is called syntactic structure, and its meaning pole is called semantic structure. Members of the former are labeled elements, while members of the latter are labeled components. RCG postulates the existence, inside a construction, of semantic relations between distinct components of the semantic structure and symbolic relations between an element of the syntactic structure and the corresponding component of the semantic structure. There is also a symbolic link between the whole syntactic structure and the whole semantic structure. Syntactic relations between elements of the syntactic structure are ruled out instead, since the only syntactic relation allowed in RCG, labeled syntactic role, is the partwhole relation between a syntactic element and the construction it makes part of. In a nutshell, as Janda (2006) points out, RCG views syntax as “an epiphenomenon of the semantic structures of grammatical constructions.” Finally, as in CG and CxGx, constructions form a structured inventory of a speaker's knowledge of their language, usually represented as a taxonomic network including very specific constructions as well as more general schemas abstracting over their commonalities. Each construction constitutes a node in such a network, and any specific construction has multiple parents. For instance, consider the following (extremely simple) example of sentence-level construction in English: (36) Harriet sings a blues. From an RCG perspective, the grammatical forms Harriet, sings, and a blues represent the two elements which make up the syntactic structure of the sentence-level construction, while the semantic structure of the construction comprises the semantic components “Harriet”, “sing”, and “blues”. Each element is linked to the corresponding 93 Grounding Meaning in Everyday Experience in the World component by a symbolic relation, while the two semantic components are linked to each other by a semantic relation. There is also a symbolic relation which connects the whole syntactic structure to the whole semantic structure. This specific construction and its elements are linked by taxonomic links to more abstract constructions and their elements. Harriet sings a blues is linked to a more general “sing” construction made up of the argument Sing1 (to which Harriet is related), the verb sing, and the argument Sing2 (to which a blues is related). In turn, this is connected to an even more general transitive construction, which comprises the elements IntrSubj (related to Sing1), IntrVerb (related to sing), and TrObj (related to Sing2). It is the case to highight the fact that verbs like sing can also occur in intransitive constructions. Croft explains this possibility positing the existence of a more abstract category subsuming both transitive and intransitive verbs, motivated by its occurrence in another construction, the morphological construction of the tense-agreement inflection. Given what highlighted above, it follows that RCG does not conceive parts of speech as categories of particular languages. By contrast, Croft relies on a theory of parts of speech he developed in earlier work, the universal-typological theory of parts of speech, positing that the grammar for any language must capture the three propositional act functions: predication, reference, and modification. The relevant roles in propositional acts constructions are divided into the semantic classes of objects, properties, and actions, which represent the typological prototypes of referring, attributing, and predicating, respectively (cfr. Langacker's view of grammatical classes). As Croft points out, the semantic classes of objects, properties, and actions are only a small subset of the semantic classes of words found in human languages, and they are defined on the basis of four semantic properties. The first is relationality, that is whether a definition of a concept inherently requires reference to another concept. The second property is stativity and whether the concept represents a state or a process. Then, the third property is transitoriness, whether the concept represents a transitory state or process or an inherent or permanent state of the entity in question. Finally, the fourth property is gradability, whether the entity is gradable along a scalar dimension. The definition of the three semantic classes of objects, properties, and actions in terms of the four properties just listed above can be observed in Tab. 2.1. Tab. 2.1: semantic properties of prototypical parts of speech (reported from Croft 2001: 87) Relationality Stativity Transitoriness Gradability Objects Properties Actions Nonrelational Relational Relational State State Process Permanent Permanent Transitory Nongradable Gradable nongradable In Croft's (2001: 88) words, “A typological prototype category is a functionally defined category that is typologically unmarked with respect to the relevant constructions.” The relevant constructions for parts of speech are illustrated in Tab. 2.2 below. According to the RCG approach, the typological prototypes for noun, verb, and adjective are crossconstructional categories. Actually, Croft (2001: 92) remarks that “the prototype patterns defined by the relevant constructions are found within languages as well as across languages.” While these prototypes underlie the parts of speech in the world's languages, their boundaries may vary from one language to another. 94 2. From Cognition to Grammar Tab. 2.2: typological prototypes (reported from Croft 2001: 88) Reference Objects Properties Actions UNMARKED NOUNS Modification genitive, adjectivalizations, PPs on nouns UNMARKED ADJECTIVES Predication predicate nominals, copulas predicate adjectives, copulas UNMARKED VERBS deadjectival nouns action nominales, complements, infinitives, gerunds participles, relative clauses In order to integrate the language-particular distribution patterns and the universals of propositional act constructions, RCG makes use of the semantic map model, an approach which has been largely exploited in recent functional-typological studies (see e.g. Haspelmath 2003). The functional properties of parts of speech define a conceptual space, i.e. a structured representation of functional structures and their relationship to each other. The conceptual space constitutes a universal structure of conceptual knowledge for human communication. A map of language-specific categories on the conceptual spaces is labeled semantic map. Croft summarizes the nature of the conceptual space approach and its important role in RCG as follows: The conceptual space model allows us to relate functional categories identifiable across languages – propositional act function and semantic class – to their syntactic expression within particular languages – the constructions used to express the propositional act functions of various semantic classes, and the constructions used to express cross-cutting conceptual distinctions for those classes. The conceptual space represents a typologically valid language universal: a universal conceptual structure and universal constraint that syntactic categories (constructionspecific and language-specific) are mapped onto conceptual categories that form connected regions in conceptual space. (Croft 2001: 97-98) In other words, while the (construction-specific) categories may be different from language to language, they are mapped onto a universal conceptual space, which, according to Croft (2001: 139), represents “the geography of the human mind”. Therefore, we can quite straightforwardly conclude that the basic scope of the RCG approach is the explanation of typological patterns on the basis of the commonality of the conceptual system. The RCG model's most remarkable contribution lies in the fact that it makes the case for a fully usage-based approach, underlining the necessity for the analyst to take typological diversity into consideration in order to carry out a plausible description and explanation of linguistic phenomena. Moreover, RCG represents an effective effort to avoid the circularity which characterizes several theories of grammar (especially but not exclusively in the formal linguistics tradition), whereby constructions define categories, and categories so defined are used to define constructions. Even though to date RCG has not been as influential as CG and CxGx, it has nonetheless been applied to account for various linguistic phenomena, both from a synchronic perspective (see e.g. Pedersen 2005) and with regard to diachronic development (e.g. Barðdal 2006). 95 Grounding Meaning in Everyday Experience in the World 2.3.1.4. A summary Above, we supplied a brief exposition of the main properties of the three cognitive approaches to grammar which can be said to have more directly influenced the ECG model. The reader will have noticed that CG and CxGx were given particular prominence, compared with RCG. This is due to the fact that Langacker's and Goldberg's model are those that effectively represent the linguistic backbone of the ECG enterprise, which exploits several concepts, structures, and ideas developed in these frameworks, and are in a sense those which still exert more influence on ECG theorists. On the other hand, RCG's influence on the development of ECG is less pervasive and it can be seen as particularly related to a couple of specific ideas explicited in Croft's model (namely, the construction-specificity of categories and relations, and the view that the grammar of any language should capture the mechanisms of reference, predication, and modification). In the following lines, we are going to briefly summarize the main commonalities and the main differences between these three approaches. First of all, it is important to keep in mind that the three models briefly outlined above share the same basic tenets and, as often vigorously pointed out by Langacker, the commonalities between them overcome the divergences96. As Broccias (2006) precisely points out, one of the most important similarities regards the common assumption that language cannot studied independently of general cognitive abilities and the agreement on viewing language as a taxonomic hierarchy, while their differences seem to be mostly related to their respective perspectives97: ...despite some important divergences, a coherent picture emerges from the various cognitive approaches examined. The importance of the various theories lies in the fact that they tend to highlight different (although related) facets of the shared conceptualization of language as a taxonomic and diffuse network. (Broccias 2006: 84, emphasis original) Moreover, as highlighted by e.g. Evans and Green (2006: 699-702), these approaches share their commitment to the symbolic thesis, according to which grammatical constructions are seen as symbolic units which are stored whole in the speaker's mental grammar, and thus are readily available as gestalts, rather than being “built from scratch” by means of rules of composition and derivation, in contrast to what held in formal approaches. Furthermore, these models are also commited to the usage-based thesis, which holds that language knowledge arises from language use. However, apart from these very general shared assumptions, when we move to a more specific level, we can notice that there are some important differences between these models, which should not be undervalued. As generally recognized, CG represents the most comprehensive and influential theory of grammar available in the Cognitive Linguistics framework. Langacker began to work on his model in the mid 70s, and by now CG has come to deal with a vast amount of linguistic phenomena (though, as Langacker himself asserts, work on grammar from a cognitive point of view can be seen as still at its very beginning). CG represents the first attempt to ground grammar in 96 On this issue, particularly relevant is the exchange of views between Goldberg (2009a, 2009b) and Langacker (2009). 97 For this reason, here we tend to view the different cognitive approaches to grammar as complementary and mutually informative, rather than competing. 96 2. From Cognition to Grammar conceptualization, i.e. to show that grammar cannot be reduced to a set of purely formal rules which manipulate items stored in the lexicon; the main goal of CG is to show that grammar is inherently meaningful, and to uncover the cognitive principles that underlie linguistic organization. At present, CG is cognitive model which has come closest to develop into an all-encompassing theory of language98. Compared with CG, other constructionist approaches are slightly less concerned with unraveling the principles which underpin the organization of language and are a bit more interested in providing an inventory of the symbolic units (that is, constructions), which constitute the linguistic system, and a description of the relations between the constructions which make up the construction lattice. Moreover, while CG can, to some extent, be considered as a “global” attempt to provide a characterization of all aspects of language as a cognitive ability, the other approaches tend to focus more on specific issues related to language. More precisely, as already outlined above, CxGx emphasizes the importance of the role of A-S constructions in determining the meaning of an expression: one of the major aims of this model is uncovering the meaning of A-S constructions, and the relation between whole constructions and their constituent parts. In particular, since the meaning of a sentence is considered to arise as the result of the interaction between the A-S construction and the main verb, CxGx is particularly interested in the interplay between verbal semantics and constructional semantics. Also, viewing constructions as meaningful entities in their own right, CxGx is interested in exploring the relations between the different senses constructions may take, which are claimed to display polysemy in the same way lexical items do. On the other hand, RCG is mostly overlapping with CG, but it is particularly aimed to account for the typological diversity of the world's languages99. The most important feature of RCG is that it advocates for a fully usage-based approach which takes language diversity as its starting point: grammatical categories are considered to be language-specific and, within the same language, construction-specific; while CG attempts to provide a universal definition for categories by means of abstract, schematic characterization arising from our cognitive abilities, RCG at most offers a definition of categories on the basis of a usage-driven functional prototype distinction. The three approaches outlined above also differ in their definition of constructions. In CG, a construction is a symbolic unit (i.e. a pairing of form and meaning) with a complex internal structure, regardless of whether its structure or meaning can be predicted from the properties of its subparts. CxGx, in the version proposed in Goldberg (1995), used to apply the label to symbolic units which were not predictable in terms of meaning or structure only. For instance, a sentence was considered a construction only if its meaning was not predictable on the basis of the meanings of its lexical components. Recently, however, Goldberg (2006) extended the label construction to include sufficiently frequent units even if their meaning is completely predictable. In RCG, every linguistic unit is a construction, regardless of its complexity and predictability. Another parameter of comparison between constructional approaches relates to the issue of whether they can be considered “reductionist” or “nonreductionist” (see Croft and Cruse 2004: ch. 10), with regard to the directionality of the relationship between 98 For instance, you can find a CG approach to grammaticalization and semantic change. Furthermore, in recent year some scholars (including Langacker himself) have undertaken the enterprise to develop a detailed explanation of the principles of phonology on cognitive basis. 99 Indeed, the similarities between the two approaches initially deceived Broccias (2002) to view RCG as a notational variant of CG applied to language typology. 97 Grounding Meaning in Everyday Experience in the World part and whole: in a nonreductionist model, the whole is the primitive unit, and the parts emerge as properties of the whole, whereas in a reductionist model, the parts are the primitives and the whole is constructed from the parts. CG and RCG are generally seen as nonreductionist, since they posit that the whole is the primitive unit and the parts are defined in relation to the whole. The position of CxGx is less clear: in the reductionismnonreductionism continuum, Goldberg's model lies somewhere in between. Indeed, Goldberg's participant roles can be seen as nonreductionist, for CxGx assumes frame semantics and, therefore, the frame is a primitive unit and the participant roles emerge from that frame. On the other hand, Goldberg's analysis of syntactic roles and relations can be seen as reductionist, since CxGx relies upon atomic primitives such as subject, object, and verb in the description of the syntactic properties of each construction, without positing an independent account of the origins of these primitives (cfr. Langacker's view of grammatical units as emerging from a trajector-landmark organization). Overall, the foremost divergences between these approaches can be seen as the effect of the perspective from which the study of language is approached. While the differences between the three models outlined above should not be underestimated, these approaches are mostly compatible, and satisfactory analyses of linguistic phenomena have been carried out within all the three frameworks. Moreover, often scholars working on language adopting one of these constructionist approaches can offer useful insights to researchers exploiting another, since these different points of view are not mutually exclusive at all. As a matter of fact, a model can provide information on facets which might be neglected or not fully worked out in another. In a sense, we may find it profitable to consider these approaches as parts of a unified enterprise which aims at providing a characterization of the domain-general cognitive principles which give rise to language as a structured inventory of symbolic units, and at an analysis of the constructions comprised in this repository of units. Keeping in mind the fact that the models sketched in this subsection constitute the linguistic part of the background of the present study, in the next section we are going to introduce ECG, another model which is part of the family of cognitive approaches to grammar, which we are going to exploit in the next chapter in order to carry out an analysis of Italian data. 2.3.2. The ECG model 2.3.2.1. What is ECG? As already mentioned, ECG is a model of CxG worked out by cognitive linguists and computer scientists affiliated to the University of California and the International Computer Science Institute. It represents an attempt to benefit from the insights provided by research carried out within some different academic disciplines (mainly linguistics, computer science, psychology, and neuroscience) in order to carry out a reliable, plausible analyses of linguistic events. On the one hand, ECG is strictly related to other cognitive and constructionist approaches to grammar. First of all, it shares with them the basic assumptions mentioned at the beginning of the previous subsection. Moreover, it also exploits notions and concepts developed and commonly used in the 98 2. From Cognition to Grammar cognitive semantics tradition (with particular reference to image-schemas, frames, and prototype-based categorization). Furthermore, ECG also directly borrows and applies ideas and structures from the three models outlined in §2.3.1.1-2.3.1.3. As an example, in ECG Langacker's notion of profiling is awarded paramount importance and is extensively exploited in the analysis of sentence-level constructions. Moreover, ECG makes use of formal and semantic connections between constituents of the constructions. Like CxGx, the ECG model allows a crucial role to A-S constructions and minimizes the level of lexical polysemy, in favor of a proliferation of constructions. Similarly to RCG, ECG applies the label construction to all pairings of form and meaning, takes grammatical relations to be construction-specific, and considers that a grammar of any language should capture the mechanisms of reference, predication, and modification100. On the other hand, there are also several points of divergence between ECG and the other cognitive models of grammar. The root of the main differences is due to the scope of ECG: while the other approaches attempt at finding the most consistent way to describe phenomena observed within and across linguistic systems, ECG aims to model the language-user's online processing of data. As a consequence, the ECG approach shows a certain degree of theoretical and methodological peculiarity with respect to the other models. First and foremost, ECG is more deeply connected to the principles of Embodied Cognition (see §2.1.1); indeed, this model supposes that our linguistic and conceptual knowledge is grounded in our neural structures and arises through bodily experience in active interaction with our social and physical environment101. In other words, Embodied Cognition represents the central pillar ECG rests on. This is a crucial point, because the choice of the ECG approach in the present study is prominently due to the model's epistemological roots. As already seen in the previous paragraphs, theories of embodiment study the connections between the activities performed by the human mind and the interaction among human beings and between human beings and the enviroment. The NTL research program investigates language from this perspective, especially trying to exploit the evidence obtained by research in different fields. ECG represents one of the most developed projects developed within the NTL paradigm, and it aims to describe the online processing of linguistic expression, providing a detailed analysis of the stages involved in the comprehension of sentences and to uncover the nature of the embodied knowledge with which the language system interacts, rather than focusing on the nature of the language system itself. ECG scholars consider meaning to be represented by a series of parameterizations of the speakers' perceptual and sensorimotor representations, which constitute a dynamic continuum (an assumption which is based on the results of psycholinguistic and neuroscientific investigations, see e.g. Bergen et al. 2004). Such schematizations are made possible by a process of generalization which extracts the most important aspects of actions and percepts required to perform a 100These are just a few of the commonalities between ECG and the cognitive models of grammar outlined in the previous subsection, further similarities will arise below when we are dealing with the mechanisms of ECG. As should become apparent -we hope- in §2.3.2.4, ECG can thus align itself with a particular model on certain topics, and agree with another on different subjects (for some remarks, see Feldman 2002; Bergen 2003) 101Not only is this assumption (though perhaps a bit too radical) supported by a considerable amount of studies carried out within the NTL paradigm, but also within other research programs (see e.g. Gallese and Lakoff 2005; Pulvermüller 2007; Gallese 2008). 99 Grounding Meaning in Everyday Experience in the World simulation, leaving details aside102. This important aspects, which also depend on facts such as experience and cultural context, are then expressed in relatively discrete linguistic units (constructions). ECG aims to describe the process of online language comprehension via a process of analysis, which consists in the recognition of all the constructions instantiated in a message. The process of analysis supplies parameters for the simulation of the content of the message, which is performed in the following phase, labeled enactment (see below). The ECG model makes use of a formal language, designed by ECG scholars in order to reach two different though related goals: one is to provide a plausible representation of dynamic language processing, the other is the development of a grammar which can be successfully implemented and tested on computers103. In a sense, ECG may be understood, at least ideally, as a sort of intersection point between the work of scholars (especially psychologists and neuroscientists) who aim to discover and simulate the cognitive processes which underlie human activities (such as language and communication) and the work of linguists who attempt at the development of a functional-cognitive theory able to account for the facts of actual language use. After reading several studies which reported the successful application of this model to account for linguistic phenomena of English, we decided to exploit it in order to build a grammar to account for grammatical phenomena of Italian. In a sense, the decision was due to the curiosity to assess the model's adequacy to account for Italian data, rather than being made on the basis of the belief that an ECG approach could provide the best analysis for our data. The motivation for the choice of ECG was due to the interest in the background on which this model rests. Even though ECG is at present the least developed among constructionist models, it has bee successfully employed to analyze several types of English constructions104 (see e.g. Dodge and Wright 2002; Feldman 2002; Bergen et al. 2004; Bergen and Chang 2005; Feldman et al. 2009; Dodge 2010a, 2010b; Dodge and Bryant forthcoming). In spite of the (quite obvious) fact that there are various remarkable differences between English and Italian, we cannot, in principle, see any reason why the ECG formalism should not work for the Italian grammar as well. Indeed, given its high degree of flexibility, we believe that it can be successfully exploited to carry out an analysis of Italian data. This hypothesis will be tested in the next chapter. We are aware of the fact the adopting the ECG model implies some problems. First of all, being a relatively recent approach, ECG is still in its infancy. Indeed, while the ECG formalism has attained a certain degree of descriptive adequacy for an amount of grammatical phenomena, it has still to be practically tested on a wider range of constructions and on a broader range of languages. Moreover, the ECG formalism shows the disadvantage of being quite complex and its adoption can be exacting for the user. Likewise, its comprehension may be not exactly straightforward for the reader. Moreover, the representation of contextual factors in terms of a formal language may be problematic and, to date, it has been awarded relatively little attention by scholars working on ECG (but see e.g. Chang 2004, 2008) Nevertheless, in our opinion the existence of these limits does not represent a reason 102Such important aspects roughly correspond to embodied schemas' roles and constraints (see below). 103In the present study, we shall be prominently dealing with the description of online language processing, not engaging in a machine-implementation of our grammar. 104Some work has also been carried out on other languages (especially Mandarin Chinese), but currently not much is available. 100 2. From Cognition to Grammar to give up the adoption of the ECG perspective to carry out our linguistic analysis105, for several reasons. First of all, while it is true that the fact that ECG is less developed than the three approaches outlined above can be seen as a limit, it is also true that it should also be considered an incentive. Indeed, any new study carried out within the ECG framework can provide an (at least partial) assessment of the model and some insights for its further development and refinement. Actually, unlike well-established models like CG and CxGx, ECG is still very much under way. Moreover, the flexibility of the ECG formalism allows the analyst to introduce changes in order to give a more precise description of the linguistic phenomenon under consideration. It is also true that the formalism is not as simple as both users and readers may want it to be, but simplicity is a criterion which can hardly be satisfied when dealing with human language (and cognitive abilities more in general). With regard to the little attention payed to pragmatic factors, this problem is (to some extent) shared by other models. As a matter of fact, it is necessary to bear in mind that not always is it possible to come into possession of the relevant information about the context of use. In fact, frequently it is not, so the efforts are generally directed to give the most plausible representation of what is known (which is, however, still an impressive goal to reach). Nevertheless, since the ECG project is lively and in progress, there is no reason to rule out that future work will try to deal with context more precisely. In summary, our choice of ECG is deeply related to its perspective and objectives: ECG is the cognitive approach which more directly tries to deal with the issue of what human beings really do when they process language (i.e. what actually happens in their minds), focusing on the nature of the embodied knowledge with which the language system interacts106. Next, not only does it take into account contributions from linguistics, but it is also particularly sensible to the results attained in other scientific fields. What is more, ECG takes into account the role of cultural factors in the dynamic process of language processing. While we do not want to (in fact, cannot) venture that ECG is the best approach to grammar, we can assert, on the basis of its background and objectives, that ECG represents a somewhat unique enterprise in the Cognitive Linguistics framework and, as such, its adoption to undertake an exploration of linguistic phenomena may also yield useful insights for other cognitive approaches to grammar. In the end, since (as previously underlined) ECG is less developed that the cognitive approches to grammar outlined above, we want to specify that our decision to carry out an ECG analysis of linguistic data from our mother tongue language may be welcome for one more reason, namely the fact that no studies on Italian have been carried out within the ECG framework. As a consequence, the application of the ECG formalism to Italian data may be insightful to assess the descriptive and explanatory power of this formalism, in order to evaluate how far it can go in the analysis of the data chosen, what its strong and weak points are and, if possible, to suggest some ways in which the model could be improved. Below, we are going to provide the reader with some details on the functioning of the ECG model. 105After all, an approach whose adoption is totally unproblematic does not seem to exist. 106Such knowledge is taken to be partially universal (as it derives in part from our human nature) and partially language-specific (being in part a result of our sociocultural imprinting). 101 Grounding Meaning in Everyday Experience in the World 2.3.2.2. An overview of the model ECG is a linguistic formalism for simulation-based language understanding, developed in recent years within the NTL paradigm. Being a model of CxG, ECG deems that linguistic knowledge can be characterized as a repository of constructions, pairings of form (phonological schemas) and meaning (conceptual schemas). Like other CxG models, ECG is monostratal, so it does not recur to rules of syntactic transformations, and it also posits that constructions cover all the subfields of linguistics. Dodge and Bryant (forthcoming) highlight the fact that there are many converging points between ECG and other cognitive approaches to grammar: Consistent with Cognitive Grammar and other construction grammars (…) we assume that a grammar includes both lexical and non-lexical constructions. Furthermore, this set of constructions is structured both by (part-whole) constituency relations, and by the relevant generalization over form, meaning, and distribution that the language learner is likely to learn (…) Our goal is not to try to create the most “compact” grammar possible. In fact, we follow Langacker and Goldberg and posit usage-based grammars that might be better described as maximalist instead of minimalist (…) In other words, because the learning of grammar is datadriven, there will be many small sub-regularities upon which generalizations are built. (Dodge and Bryant forthcoming) Furthermore, as pointed out by Bergen and Chang (2005: 147) in one of the foundational papers on ECG, this model assumes (along with other models of CxG) that “...language users exploit constructions at these various levels to discern from a particular utterance a corresponding collection of interrelated conceptual structures”. Nevertheless, there are some aspects on which ECG diverges from other CxG model. First of all, as Evans (2007: 67) asserts, “instead of concentrating on modeling linguistic knowledge, ECG takes for granted that constructions constitute the basic level of such knowledge and focuses on the online processing of linguistic data and on the nature of embodied knowledge.” Moreover, ECG has also been thought up as a model of “formal cognitive linguistics” (Feldman 2004; Feldman et al. 2009). This is not a paradox, as it may sound, nor is it an attempt to bring Cognitive Linguistics closer to the formalist tradition. Rather, it represents an attempt to provide Cognitive Linguistics with a precise notation in order to facilitate a detailed analysis of language instances (cfr. Lakoff 1987: 486). Actually, ECG aims to allow the analyst to exploit insights from the Cognitive Linguistics literature and at the same time to benefit from the use of computational tools107. In fact, some researchers have exploited the ECG formalism to develop highlevel systems of language understanding and models of language learning (see e.g. Chang's 2008 model of child language acquisition). Another characteristic of ECG is the it takes semantic concepts to be embodied and compositional (see Dodge 2010a). They are embodied, since ECG is consistent with the assumptions of the NTL research program, among which we can mention the belief that language is interconnected with meaning and thought and continuous with other cognitive processes, which have a neural realization (see for instance Feldman and Narayanan 2004; Gallese and Lakoff 2005; Bergen and Feldman 2009). They are also compositional, for ECG's basic primitives, schemas and constructions, are seen as gestalts with a complex internal structure, and the meaning of the whole is considered to be provided, partially but 107Søgaard (2006) also managed to convert the ECG formalism in a layered modal language, proving its computational efficiency. 102 2. From Cognition to Grammar crucially, by the meaning of its parts108. These are the main characteristics which make the ECG model a rather innovative project in the world of cognitive approaches to grammar. The ECG model considers the understanding of an utterance as the internal activation of schemas (i.e. cognitive structures generalized over recurrent perceptual and motor experiences) and their mental simulation in context, which is meant to produce the richest set of inferences possible. In order to provide a dynamic inferential semantics, ECG makes use of the so-called “simulation-based language understanding model”: constructions express generalizations linking phonological and conceptual schemas. Together with the communicative context, the existence of a repository of constructions triggers a process of analysis, which determines which particular constructions are instantiated by an actual utterance. The product of the analysis process is the semantic specification (SemSpec henceforth), which specifies the conceptual schemas evoked by the constructions involved and how they are related. Taking the SemSpec as an input, an enactment process109 is then run, which consists in the internal recreation of previous perceptual, motor, social, and affective experiences, even though such stimuli are not contextually present. The enactment process results in open-ended inferences110, which influence the language user's response and shape later processing. Our study, following most work on ECG, will focus on the process of analysis, for an important reason. ECG lies on the concept of embodiment, according to which meaning is grounded in perceptual and motor simulations, which represent a continuous, dynamic system. Language, on the other hand, is an inventory of relatively discrete linguistic units. Consequently, constructional analysis, which represents the languageunderstander's recognition of the constructions instantiated in an utterance, serves as an interface between language and perceptuo-motor simulations (see Chang et al. 2004), giving rise to SemSpec, which provides parameters for enactment. In other words, the meaning of relatively discrete linguistic units is the result of a process of generalization, which “compresses” the continuous perceptuo-motor simulations, omitting details and capturing just their more important aspects. In language understanding, the constructional analysis process gives rise to the SemSpec, which specifies such important aspects, which are crucial for enactment. Therefore, we can observe that analysis is the subprocess which more directly deals with language, bridging the relatively discrete linguistic units (constructions) and the continuous perceptual and motor simulations, expressing their important aspects and giving rise to their parameterization (Bergen et al. 2004, Bergen 2005a). These parameters, provided by the SemSpec, represent what is needed for the activation of conceptual structures. The enactment process then gives rise to inferences, allowing a deeper understanding of the message conveyed by the language user. As a result, we can say that analysis is the process which “does the linguistic job”, representing the interaction between linguistic and conceptual knowledge, while enactment is a more intrasubjective process. As such, enactment cannot be totally specified by language (Feldman and Narayanan 2004). In a sense, as underlined by Feldman (2010), the process of analysis captures what is the 108Compositionality is also supposed by other constructionist approaches to grammar (e.g. Lakoff 1987: case study 3; Croft 2001: ch. 5), but in ECG it is awarded special emphasis, as highlighted by Ettlinger (2005) and Feldman et al. (2009). 109The terms enactment and simulation are often used interchangeably (see e.g. Feldman 2002; Bergen et al. 2004; Bergen and Chang 2005; Feldman et al. 2009). 110These inferences are not necessarily direct. As a matter of fact, they are often metaphorical (see Narayanan 1997; Bergen 2005a). 103 Grounding Meaning in Everyday Experience in the World meaning which is shared by the members of a language community on the basis of similar genetics and experience, i.e. a selective and conventionally schematic map between language and ideas. From this perspective, the process of analysis shades some light on the intersubjective dimension of language, whereas the enactment process will (unavoidably) be dependent on the personal experience of the specific language-user. 2.3.2.3. Schemas and constructions As underlined in various papers, the ECG formalism needs both to provide a coherent interface between constructional analysis and embodied simulation and to be defined with sufficient precision to be implemented in a computational model (in particular, this issue is addressed briefly in Chang et al. 2004; Bergen and Chang 2005; and more incisively in Chang 2008: ch. 3; Feldman et al. 2009). In order to satisfy these conditions, ECG makes use of the two basic primitives mentioned above: schemas and constructions. As we shall see clearly later, these primitives play a crucial role in the process of analysis. ECG schemas define meaning constraints: relations between schemas are defined in terms of roles, subcases, evoked structures and constraints111, in order to create a coherent lattice of interrelated schemas. Roles are structured relationships between a set of participants and can be instantiated by particular values, called “fillers”. Subcases are specific cases of more general schemas. The subcase of keyword connects the schema to its type lattice, allowing for structure sharing through inheritance. Evoked structures are schemas against which the schema under consideration can be defined. Evoking a structure makes it locally available without imposing a part-of or subtype relation between the evoking and the evoked structure112. Finally, ECG posits some kinds of constraints on roles. Type constraints (indicated with a colon, as a:b) restrict a role to be filled by a specific kind of filler. Identification constraints (indicated with a double-headed arrow, as a ↔ b) cause fillers to be shared between a and b. Slot-chain notation is exploited to refer to a role b of a structure a (as a.b). Additionally, the ECG formalism makes use of filler constraints (expressed using a single-headed arrow, as a ← b), to specify that the role a is filled by the constant value b. Finally, the keyword self is used to refer to the structure being defined. The selfreference faculty allows constraints to be established at the level of the entire structure. Even though in our brief explanation this may seem a rigid mechanism, as Bergen and Chang (2005: 153) emphasize, “Overall, the ECG schema formalism provides precise but flexible means of expressing schematic meanings, ranging from individual schemas to structured scenarios in which multiple schemas interact” (italics added). Schemas are used to represent a variety of conceptual structures, including some conceptual configurations described in the Cognitive Linguistics literature113, and they 111This statement holds for schemas with a certain degree of complexity. In fact, the simplest schemas, such as the RD schema shown below, can be drawn simply as lists of roles, as Bergen and Chang (2005: 151) underline. 112There is a sort of trend among scholars working on ECG to make use of Langacker's example of the relation between the concept “hypotenuse” and the concept “right triangle” to explain the evokes keyword (see Feldman et al. 2009; Dodge and Bryant forthcoming): while the former is not a kind of right triangle, the latter is not a role of the hypotenuse. The evoking structure is meant to represent relations along the lines of that which lies between the hypotenuse and its right triangle conceptual base. As Bergen and Chang (2005: 153) point out, perhaps it may be construed as a formalization of the “profiling” notion used in CG (and Frame Semantics). 113A rather central role is played by image-schemas and, to a certain extent, also by frames (for an 104 2. From Cognition to Grammar are defined as a sort of gestalt figures with a narrow number of internal parts, represented as roles. Schemas are also defined as parts of a larger schema-lattice, with each of these schemas having various types of specified connections to other schemas in the lattice. As Dodge and Bryant (forthcoming) highlight, “This reflects the complexity and interconnectivity of the conceptual network these schemas are being used to (partially) represent”. Dodge and Bryant posit that there are several primitive schemas forming a crucial part of this lattice, and that such supposed primitives reflect recurrent schematic shared features in basic experiences. It may be relevant to note that this issue indirectly “makes a case” for the analysis of non-English data we are going to undertake in ch. 3. Indeed, Dodge and Bryant go on with the following chunk: Such experiences are presumably shared by people, all of whom process them using some of the same basic functional networks in the brain. Therefore, these schemas are likely to be universally-available to speakers of all languages, though they may of course be utilized in different ways by different languages. A fully defined grammar will also include schemas that represent commonalities in more culturally-specific experiences. These schemas, akin to frames, will also specify relations to other schemas in the lattice. (Dodge and Bryant forthcoming) The lines quoted above underline the importance of working on different languages in order to find out the way in which the supposedly universal schema-lattices are organized in each of them. At present, some ECG analysis on various languages have been carried out (see for instance Chang and Gurevich's 2004 paper on Russian morphology), but the bulk of the ECG work has dealt with English114. With reference to constructions, as mentioned above this label applies to pairings of form and meaning of various sizes and levels of concreteness. Their potential forms are not as open-ended as potential meanings, but they can include different kinds of information, ranging from phonological schemas to prosodic patterns (past orthographic forms, gestures, and so on). For the sake of clarity (and often also space), studies generally refer to the phonological or orthographic aspects, only115. The construction formalism derives from the schema formalism, so it comprises all the elements outlined for schemas116 (i.e. roles, subcases, evoked structures, and constraints), and it also adds two devices standing for their linked poles, form and meaning. These poles provide a list of the elements and constraints (if any) within each domain, but (as highlighted in Bergen and Chang 2005) they should also be viewed as special components of the construction that can be referred to and constrained, similarly to schema roles. Like schemas, constructions are organized in a construction-lattice. In the remainder of this subsection, we are going to outline a recent study of Dodge and Bryant's and to provide a simple example of the ECG formalism at work using the grammar built by these two authors. overview of these and many more cognitive structures, see e.g. Croft and Cruse 2004: part I) 114Moreover, as Francisco Gonzálvez García (personal communication) underlined, very little has been done on Italian from any constructionist perspective. Nevertheless, some work has been carried out (from Goldberg's perspective) in recent years (see e.g. Quochi 2007; Masini and Pietrandrea 2010). 115In our analysis we shall focus on phonological information only. 116Actually, making use of a metaphor sketched by Lakoff and Johnson (1999: 134), we may say that a certain embodied schema represents the skeleton of the concept “fleshed out” by the respective construction. 105 Grounding Meaning in Everyday Experience in the World 2.3.2.4. Dodge and Bryant's grammar and an illustrative example In order to provide an illustration of how the ECG formalism works, in this subsection we will illustrate the analysis of the simple Italian sentence you can find in (37) below: (37) Sara tagliò la pagnotta Sara cut:PRT.3SG the.F.SG loaf.SG “Sara cut the loaf” Sara tagliò la pagnotta is an Italian active transitive construction (a sentence which is analogous to those used for English by Feldman et al. 2009; Dodge and Bryant forthcoming), which includes a verb of force-exertion. Now, we are going to introduce Dodge and Bryant's (forthcoming) methodology and their grammar, successfully put to use in a study on English transitive constructions. This grammar employs, together with other, the schemas and constructions outlined below, developed by ECG researchers during the last years. It is basically written for English but, as Ellen K. Dodge (personal communication) pointed us out, it has been worked out in order to be also suitable for other languages. We shall exploit it to analyze the simple Italian transitive sentence in (37) above, but before we begin there are some issues to be addressed. As typologists have often pointed out (see e.g. Næss 2007), transitivity is not as simple a concept as one may tend to believe. It is generally seen as an action involving two participants: a volitional agent who performs an action on a patient, affecting them someway. In Italian, as in English, these two participants are expressed by a subject NP and a direct object NP, respectively. The analysis of transitivity has to face some problems both in English and Italian (even though it seems that English has to deal with more). We shall mention just one problem for each of the two languages. As Dodge and Bryant (forthcoming) underline, in English “many verbs that can be used transitively can also appear in intransitive sentences, in which 'patient' is expressed as subject” (e.g. He slid the glass across the table vs the glass slid down the ramp)”. Similarly, we can observe the fact that in Italian some verbs can be used both transitively and intransitively, with some weird syntactic consequences. For instance, the verb scendere (“to go down”) can be used transitively or intransitively, as shown in the two clauses in (38) below117: (38) (a) Scesi le scale. go.down:PRT.1SG the.F.PL stair.PL “I went down the stairs” Scesi dall' auto. go.down:PRT.1SG from-the.F.SG car “I got off the car” (b) Dodge and Bryant highlight the fact that if we wanted to handle argument realization patterns at verb-level, we would need to posit different constructions for each different pattern, while using A-S constructions enables us to capture broader generalizations about the argument realization patterns for groups of semantically similar verbs, also reducing the number of verb constructions to be posited118. At least one transitive A-S 117It should also be noted that, depending on its transitive or intransitive use, this verb in the present and past perfect takes two different auxiliary verbs. 118As a matter of fact, setting argument structures free from verbs is one of the major results obtained by 106 2. From Cognition to Grammar construction, identified with a prototypical119 transitive event, is posited to unify with the different types of various verbs. It also specifies links between these scene roles and their fillers (agent ↔ subject, patient ↔ direct object). The unification of the A-S construction with a verb causes the blending of its roles with those of the verb. This methodology potentially permits to posit a single verb construction for verbs which appear in different argument realization patterns. Such a verb construction will unify with the transitive A-S construction but also potentially with more. Therefore, the same verb construction can describe different kinds of events. Dodge and Bryant define several different transitive A-S constructions, each of which aims at revealing relatively local generalizations over constituency, form, and meaning, also recognizing relations between and generalizations over these constructions. A crucial role is played by the representation of A-S constructions as a hierarchy. In Dodge and Bryant's grammar, A-S constructions have three important types of meaning specification relevant to the present analysis. The first concerns the type of scene, since not all transitive sentences describe the prototypical transitive scene. The solution proposed is to use a different schema to represent each scene. Compare, for instance, the following examples: (39) (40) (41) He cut the bread. The falling shard of glass cut the bread. He saw the bread. The simple clause in (39), on which model our target sentence was built, depicts the prototypical transitive scene, and is represented by the CauseEffectAction schema, whose properties will be illustrated in Fig. 2.11 below. This kind of construction will be described in detail during the discussion of our Italian example. The kind of scene captured by the example in (40) differs from that depicted by the previous sentence because it lacks agency. This sentence represent a two-participant event in which the non-agentive action of one participant (the falling shard of glass) affects the other (the bread) in some way (being cut). In this kind of “causal action”, the causer may not be an animate entity. This kind of scene can be captured using a different schema, which Dodge and Bryant label CauseEffectProcess, which is similar to CauseEffectAction but has looser constraints on the type of causal process involved. In particular, its describes a transfer of force to another entity which does not necessarily involve a motor-control action. This is consistent with scenarios in which force is transferred from the non-agentive moving causer to the other entity upon impact120. The example in (41) describes a scene involving a subject associated with a process, but this process does not causes any effect on the other participant. A sentence of this kind can be analyzed as describing an event involving one entity who undergoes some perceptual experience (he), and another entity that provides the content of this experience (the bread). Such an event can be represented with a schema containing (at Goldberg (1995), as underlined in Du Bois (2003). 119Dodge and Bryant conceive A-S constructions as a radial category, with a central case and several extensions. On the nature of radial categories and their application to grammar, see Lakoff (1987). 120For a detailed characterization of the functioning of the CauseEffectProcess schema, the reader is referred to Dodge and Bryant (forthcoming). 107 Grounding Meaning in Everyday Experience in the World least) these two roles121. This kind of scene is related to the prototypical transitive scene depicted in (39) in the fact that the experiencer, as well as the agent in that kind of scene, in order to successfully perform the action needs to “attend to” various properties of the entity he is acting on. There is an asymmetrical experiential / perceptual relation between the two participants. The second type of meaning specification we are concerned with regards the relation between verb and A-S construction meaning. As Dodge and Bryant point out, transitive sentences that describe prototypical transitive scenes face the problem that the same verb can often be used to describe different types of scenes, which are often accompanied by difference in argument realization patterns. Instead of positing different verb constructions for each type of scene that a verb can be used to portray, Dodge and Bryant posit a single verb construction which can potentially unify with different A-S constructions, each of which may describe a different type of scene. The third type concerns perspective: A-S constructions also specify from which participant's perspective this scene is conceptualized and described. Perspective especially plays a crucial role in the distinction between active and passive sentences. Taking all these aspects into consideration, now we can move on to analyze the simple Italian sentence in (37) above, applying several insights from Dodge and Bryant's methodology (without, of course, failing to consider other contributions from other papers on ECG). In order to carry out an analysis of the sentence, we have to identify and represent the relevant meaning elements of the verb and A-S constructions involved, by means of schemas and bindings122. We shall be dealing with the referring expressions (i.e. those describing the two ontological entities which take part of the action) later, focusing at present on just those capturing processes. The sentence we aim to analyze describes a prototypical transitive scene: a forceful motor-control action. As such, it captures two processes tied together by a causal relation: the agent's act of force-exertion affects the patient, and such effect provides the agent with some feedback 123. To represent the scene described by the sentence in (37) above, it will be useful to introduce some basic schemas and constructions, beginning with the schemas representing embodied processes (i.e. Process and ComplexProcess), shown in Fig. 2.1 and Fig. 2.2, respectively124. Fig. 2.1: the Process schema schema Process roles x-net: X-net protagonist 121To our knowledge, such a schema still has to be defined in its details. 122Dodge and Bryant (forthcoming) suggest that most of the apparent schematic complexity is “part of the rich, complex structure we, as agents, need to successfully interact with the world”. This is one of the ultimate reasons why the present model is called Embodied Construction Grammar. 123Here, we can clearly see a gestalt conceptualization with a complex internal structure. 124Consistent with a convention adopted in studies on ECG, in our study inherited roles may be omitted in the representation of a schema or construction. 108 2. From Cognition to Grammar Fig. 2.2: the ComplexProcess schema. schema ComplexProcess subcase of Process roles protagonist // inherited protagonist2 x-net: @complexx-net process1: Process process2: Process constraints protagonist ↔ process1.protagonist protagonist2 ↔ process2.protagonist The Process schema has two roles: protagonist for its only participant, and x-net for the specific embodied process associated to such a participant. ComplexProcess is a structure-building schema specifying how two simple processes can form a single complex process compositionally. It has two roles for these “subprocesses”: process1 and process2. Its x-net role is a complexx-net which integrates the x-net of these subprocesses. The ComplexProcess schema shows the bindings between roles which are requested to share the same filler. The protagonist of the first process (inherited from the Process schema) is co-indexed with the protagonist role of process1, while protagonist2 is bound to the protagonist of process2. Schemas from the same schema-lattice of the Process schema can be picked and used to represent the meanings of both verb and A-S constructions. As Dodge and Bryant (forthcoming) point out, “using schemas from the same lattice to represent the meaning of both verb and AS constructions makes it a fairly straightforward matter to identify and specify meaning relations between them”. Still, verb differs from A-S constructions since the former's meaning is analyzed as having a core schematic component but can also have other, more lexically-specific, while the latter's is taken to be fairly general. Therefore, these outlined semantic relations illustrate why and how the meanings of these constructions will compose when the constructions unify. Obviously, these relations bind both processes and their associated roles, which sometimes result in more complex roles than specified by either construction. It is important to underline that these specifications will apply to semantically close verbs, which identify their meaning with a given schema. Therefore, there is no need to postulate lexically-specific relations. An instance of simple lexical construction is the verb used in our example sentence, tagliò, shown in Fig. 2.3 below. The TagliareSimplePast3sg construction is a subcase of a more general SimplePastTense construction (Fig. 2.4), subcase of a more general FiniteVerb construction (Fig. 2.5), which in turn is subcase of an even more general HasVerbFeatures construction (Fig. 2.6)125. 125We will not be showing all the more abstract schemas and constructions, unless they provide relevant information to our analysis. For a detailed illustration of ECG hierarchies of schemas and constructions, the reader is advised to refer to Starter2 grammar, freely available for download together with Luca Gilardi's ECG Workbench at: http://www.icsi.berkeley.edu/~lucag/. 109 Grounding Meaning in Everyday Experience in the World Fig. 2.3: the TagliareSimplePast3sg construction. construction TagliareSimplePast3sg subcase of SimplePastTense, Singular, 3rdPerson constructional constraints self.features.verbform ← simplepast // inherited self.features.number ← singular // inherited self.features.person ← 3 // inherited form: WordForm // inherited constraints self.f.phon ← /ta'ʎo/ meaning: CauseEffectAction constraints self.m.x-net ← @cut Fig. 2.4: the SimplePastTense construction construction SimplePastTense subcase of Finite constructional: VerbFeatureSet // inh. constraints self.features.verbform ← simplepast Fig. 2.5: the Finite construction construction Finite subcase of HasVerbFeatures constructional: VerbFeatureSet // inh. constraints self.features.verbform ← finite SimplePastTense inherits the constructional block from the Finite construction. This is bound to the VerbFeatureSet schema (Fig. 2.7 below), which represents a subcase of the AgreementFeatureSet schema (not shown). The constructional block is used to perform a double duty (see e.g. Chang et al. 2004; Bergen and Chang 2005): to list constructional constituents in complex constructions, but also to specify the elements or constraints applicable to a (simple or complex) construction as a whole, i.e. the information which cannot be properly ascribed to either the form or the meaning pole126. Fig. 2.6: the HasVerbFeatures construction construction HasVerbFeatures subcase of HasAgreementFeatures constructional: VerbFeatureSet 126When a constructional block is inherited from by a construction from a more general one, it (or some of its roles) may be omitted in the representation of the less general construction. 110 2. From Cognition to Grammar Fig. 2.7: the VerbFeatureSet schema schema VerbFeatureSet subcase of AgreementFeatureSet roles verbform number // inherited person // inherited TagliareSimplePast3sg is also a subcase of the Singular and the 3rdPerson constructions, shown in Fig. 2.8 and Fig. 2.9 below, respectively. Fig. 2.8: the Singular construction construction Singular subcase of HasAgreementFeatures constructional constraints self.features.number ← singular Fig. 2.9: the 3rdPerson construction construction 3rdPerson subcase of HasAgreementFeatures constructional constraints self.features.person ← 3 Observing Fig. 2.3, you will have noticed that the form pole of the construction is constrained to be a word using the WordForm schema, shown in Fig. 2.10 below127. Fig. 2.10: the WordForm schema. schema WordForm roles phon The phonological structure of tagliò is specified by setting the right value for the phon role. The self keyword is used to refer to the construction itself, and the f keyword to refer to its form pole. The meaning pole of the TagliareSimplePast3sg construction is specified to be a CauseEffectAction, a semantic schema shown in Fig. 2.11 below, which represents a subcase of the ComplexProcess schemas outlined above. Moreover, the inherited x-net role is given the appropriate value. 127The reader should remember that phon is not the only role of the WordForm schema. For the sake of both space and simplification, scholars working on ECG just specify the phonological (see Bergen and Chang 2005) or orthographical properties of a word (see Feldman et al. 2009; Dodge and Bryant forthcoming). As already specified, we shall select the former option. 111 Grounding Meaning in Everyday Experience in the World Fig. 2.11: the CauseEffectAction schema. schema CauseEffectAction subcase of ComplexProcess roles process1: ForceApplication process2: Process // inherited causer affected constraints protagonist ↔ causer protagonist2 ↔ affected process1.ActedUpon ↔ affected The CauseEffectAction schema composes two (or more) schemas to represent a more complex situation. CauseEffectAction is a subcase of ComplexProcess in which process1 (the cause) has a ForceApplication type constraint and process2 (the effect) is more loosely constrained to be some Process. It also adds roles for a causer and an affected participant, and uses co-indexation to bind the causer and affected roles to the appropriate roles in process1 and process2. A further constraint specifies that the affected participant is bound to the actedUpon role of ForceApplication. Consequently, causer is forceApplication.actor, and affected is protagonist of process2 as well as forceApplication.actedUpon. Fig. 2.12 below shows the ForceApplication schema. Fig. 2.12: the ForceApplication schema. schema ForceApplication subcase of MotorControl evokes ForceTransfer as ft roles actedUpon effort // inherited constraints actor ↔ ft.supplier actedUpon ↔ ft.recipient effort ↔ ft.amount The ForceApplication schema, a subcase of the MotorControl schema, describes motor-control actions which imply force-exertion. As a consequence, it evokes another schema, ForceTransfer, which describes a transfer of force between a supplier and a recipient. ForceApplication inherits the roles of MotorControl, adding also a role for effort and one for an actedUpon entity. The constraints block of the schema specifies, 112 2. From Cognition to Grammar through its bindings, that the entity that is acted upon receives the force supplied by the actor, and that the actor's effort is correlated with the amount of force transferred. The MotorControl schema and the ForceTransfer schema are shown in Fig. 2.13 and Fig. 2.14 below. Fig. 2.13: the MotorControl schema schema MotorControl subcase of Process roles actor: animate effector effort routine constraints actor ↔ protagonist routine ↔ x-net Fig. 2.14: the ForceTransfer schema schema ForceTransfer roles supplier recipient amount As shown above, the MotorControl schema is a subcase of Process. It adds roles for an actor, effector, and effort. The actor is the protagonist, the effector is the controlled body part, and the effort is the energy expenditure. Dodge and Bryant's grammar also contains several types of phrasal constructions, a crucial type of which is A-S constructions, whose meaning can be defined as that of basic scenes128. Each A-S construction clarifies the pattern in which the participants are expressed and indicates how they are related to the meaning of the verb. ECG A-S constructions should also provide guidance about how a scene should be simulated. For instance, as Feldman et al. (2009) point out, active and passive voice supply different perspectives on the same scene, and this shift of perspective must be indicated in the SemSpec to support the appropriate simulation. In order to carry out this duty, Dodge and Bryant exploit the EventDescriptor schema. The roles of this schema will, in any given utterance, be bound to the schemas and/or roles specified by the argument structure and other instantiated constructions. The eventType role is bound to the Process that represents the scene which is being described. The filler of this role is provided by an A-S construction. The profiledProcess role is bound to the process (or subprocesses) being profiled in the scene. Its filler is supplied by the verb. The profiledParticipant role is bound to the participant in the scene that is being profiled. It can be conceived as the semantic correlate of subject and it is bound to different roles in a scene depending on whether the utterance is active or passive. The topic role is used when the topic of the discourse segment does not correspond to the profiled participant of the event. Consequently, we can observe that the role of the EventDescriptor schema catches some key parameters in the description of any event129. 128On the semantic relation between basic scenes and A-S constructions, see Goldberg (1995, 2006). 129Feldman et al. (2009) suggest that the EventDescriptor schema extends the central function of predication in grammar as proposed in Croft (2001), proposing that an utterance has two primary construals: the scene provided by the A-S construction and a particular process provided by the verb (which may or may not be the same). 113 Grounding Meaning in Everyday Experience in the World EventDescriptor is crucial, together with the VerbPlusArguments construction, in the description of phrasal constructions. Phrasal construction differ from lexical construction in the presence of constituents in the constructional block. VerbPlusArguments represents the root of the A-S hierarchy. It is marked with general keyword, for it represents a generalization over all its subcases (namely, all A-S constructions in this grammar have a verb constituent labeled v). It has no form constraints, so the form block is omitted. In its meaning block, VerbPlusArguments inherits an evoked EventDescriptor from a more general construction VP (not shown) and the constraint that its meaning be bound to the EventDescriptor's eventType role. It specifies that the meaning of the verb is bound to the profiledProcess role of the EventDescriptor schema. Both EventDescriptor and VerbPlusArguments are shown in Fig. 2.15 and Fig. 2.16, respectively. Fig. 2.15: the EventDescriptor schema schema EventDescriptor roles eventType: Process profiledProcess: Process profiledParticipant topic Fig. 2.16: the VerbPlusArguments construction. construction VerbPlusArguments subcase of VP constructional: VerbFeatureSet constituents v: Verb constraints self.features ↔ v.features meaning: Process evokes EventDescriptor as ed constraints self.m. ↔ ed.eventType v.m. ↔ ed.profiledProcess One subcase of VerbPlusArguments is TransitiveCEA, which is the one involved in our Italian sentence above. It represents transitive VPs that have clausal verbs with a force-application component, such as Italian tagliare. It defines an NP constituent to represent its object, and adds the form constraint that the verb's form (v.f) comes before the np's form (np.f). The meaning of the TransitiveCEA construction can be easily identified with the CauseEffectAction schema. Its semantic roles are referred to using the slot chain self.m. Therefore, self.m.causer is related to its semantic causer role, which is co-indexed with the profiledParticipant role of EventDescriptor. Similarly, the affected participant is co-indexed with the meaning of the np constituent. Moreover, the meaning of the construction itself (self.m) is co-indexed with the meaning of the verb. This co-indexation has two consequences: it reduces the kinds of verbs which can co-occur with the TransitiveCEA construction to those whose poles can unify with CauseEffectAction, and it unifies the meaning of the verb with the meaning pole of the construction. TransitiveCEA is shown in Fig. 2.17 below. Dodge and Bryant's grammar also includes clause-level constructions, such as Declarative, which identifies its meaning with EventDescriptor schema, stating that 114 2. From Cognition to Grammar this type of construction describes an event. Declarative is a subcase of S-With-Subj construction, from which it derives a subj constituent, together with some properties related to the subject. It also inherits the constraint that the meaning of subj is bound to the EventDescriptor's profiledParticipant role. Declarative in this way signals that the event should be simulated from the perspective of the entity referred to by the subject constituent. Fig. 2.17: the TransitiveCEA construction construction TransitiveCEA subcase of VerbPlusArguments constructional constituents np: NP form constraints v.f before np.f meaning: CauseEffectAction constraints self.m.causer ↔ ed.profiledParticipant self.m ↔ v.m self.m.affected ↔ np.m At the same time, Declarative does not precise either what type of event is being described or which event-related semantic role the profiledParticipant is associated with. This information is provided by the analysis of the second constituent of the Declarative construction, a finite verb phrase (fin). The unification of an A-S construction with Declarative causes the semantic role bound to profiledParticipant to be co-indexed with the meaning of Declarative's subj constituent. The constructional block of Declarative also provides us with the information that the clause's mood feature is filled by the declarative value. The Declarative construction is shown in Fig. 2.18 below. 115 Grounding Meaning in Everyday Experience in the World Fig. 2.18: the Declarative construction construction Declarative subcase of S-With-Subj constructional constituents subj: NP // inherited fin: VP constraints self.features.mood ← declarative form constraints subj.f before fin.f meaning: EventDescriptor constraints self.m. ↔ fin.ed subj.m ↔ fin.ed.profiledParticipant // inherited The meanings of nominal and pronominal constructions in the grammar are represented using a referent descriptor (RD) schema. This schema contains several roles which can be used to define various constraints related to a referent. The RD schema is shown in Fig. 2.19 below. Fig. 2.19: the RD schema schema RD roles category natgender number mass accessibility referent The sentence Sara tagliò la pagnotta instantiates some of the constructions described above: TransitiveCEA (Fig. 2.17), TagliareSimplePast3sg (Fig. 2.3), and Declarative (Fig. 2.18). In order to specify the nature of the event described (i.e. a twoparticipant event in which the actor exerts force on the patient, affecting it), the A-S construction TransitiveCEA identifies its meaning with the CauseEffectAction schema. The meaning of TransitiveCEA is bound to the eventType of an evoked EventDescriptor schema (Fig. 2.15). Therefore, the eventType has the same causal structure of the CauseEffectAction schema. The profiledParticipant of TransitiveCEA is clarified to be the causer role of 116 2. From Cognition to Grammar CauseEffectAction. TransitiveCEA inherits a verb constituent and the constraint which binds its meaning to the profiledProcess of the EventDescriptor. Moreover, TransitiveCEA binds the meaning of the verb constituent to that of the A-S construction (so, to the CauseEffectAction schema). The TagliareSimplePast3sg construction also identifies its meaning with the CauseEffectAction schema and, as a result, it meets the constraints specified for the verb constituent of TransitiveCEA. Furthermore, TagliareSimplePast3sg specifies that the cause-effect action involves a particular kind of causal action (a “cut” x-net) and may also specify details about the effector and the type of effect. It follows that the verb constructions and the A-S construction share the same general schematic meaning, but the verb construction also supplies more specific meaning specification. TransitiveCEA has also an np constituent. The form of this constituent is contrained to follow that of the verb130. Meaning constraints specify that the meaning of this np constituent is bound to the affected role of the CauseEffectAction schema. Therefore, this construction specifies that the entity expressed by the “direct object” np constituent is acted upon and potentially affected by the causer's actions. Declarative's meaning is identified with an EventDescriptor schema. It has an inherited subj constituent, whose meaning is bound to the profiledParticipant role of EventDescriptor. In addition, it has a second constituent, fin, which unifies with TransitiveCEA. Furthermore, Declarative specifies that the EventDescriptor evoked by the A-S construction is to be identified with that of Declarative, signaling that both constructions describe the same event. Also, the profiledParticipant roles referred to by each of these constructions will be identified with one another. As a consequence, the meaning of Declarative's subj constituent will be identified with the causer role of CauseEffectAction. The constructions listed above supply quite general constraints on the fillers of various participant roles. More specific information on the fillers of these roles for a particular event is provided when the NP constructions instantiated in the utterance unify with the other instantiated constructions. In our example sentence, there are two NP constructions: one for Sara and one for la pagnotta131. The Sara construction is shown in Fig. 2.20 below. The constructional block is bound to the NominalFeatureSet schema, shown in Fig. 2.21 below. 130These constraint only holds for canonical Italian transitive sentences with unmarked SVO word order. 131Dodge and Bryant (forthcoming) do not provide any details on referring expressions, but for clarity's sake we will represent them in Figg. 2.20-2.31. 117 Grounding Meaning in Everyday Experience in the World Fig. 2.20: the Sara construction construction Sara subcase of ProperNounNP constructional: NominalFeatureSet constraints self.features.number ← singular self.features.grammgender ← female form: WordForm // inherited constraints self.f.phon ← /'sara/ meaning: RD // inherited constraints self.m.category ← Human self.m.natgender ← female self.m.number ← one self.m.accessibility ← inactive self.m.referent ↔ Sara Fig. 2.21: the NominalFeatureSet schema schema NominalFeatureSet subcase of AgreementFeatureSet roles case grammgender // inherited number // inherited person // inherited The roles of the NominalFeatureSet schema, to which the constructional block of nouns is bound, allows us to specify the grammatical features of the referents of the constructions keeping them separate from their properly semantic features, which are bound to the roles of the RD schema shown in the meaning block, instead. Declarative's subj constituent will unify with Sara, providing the information that the entity that fills the causer and profiledParticipant role is one, female, Human, referent called Sara. Obviously, the level of accessibility of a proper noun like Sara will depend on the communicative context. In Fig. 2.20, the referent is considered inactive because proper nouns can be used in the absence of any previous mention132; as a matter of fact, unlike pronouns, proper nouns refer to specific ontological entities. TransitiveCEA's np constituent will unify with the NP construction whose noun constituent is Pagnotta, thus providing more specific information about the filler of the affected role. The Pagnotta construction instantiates the FemaleSingularNoun construction. The FemaleSingularNoun and Pagnotta constructionsare shown in Fig. 2.22 and Fig. 2.23, respectively. 132Following Chang (2008), we use a modified version of Lambrecht's (1994) scale of accessibility. Nevertheless, we shall restrict the levels of accessibility to the following possible values: “active”, “uniquely-identifiable”, “type-identifiable”, “inactive”, “unidentifiable”. 118 2. From Cognition to Grammar Fig. 2.22: The FemaleSingularNoun construction construction FemaleSingularNoun subcase of SingularNoun, Female constructional: NominalFeatureSet // inherited constraints self.features.grammgender ← female // inherited self.features.number ← singular // inherited form: WordForm // inherited meaning: RD // inherited constraints self.m.number ← one // inherited Fig. 2.23: the Pagnotta construction construction Pagnotta subcase of FemaleSingularNoun form: WordForm // inherited constraints self.f.phon ← /pa'ɲot:a/ meaning: RD // inherited constraints self.m.category ← Loaf self.m.natgender ← neuter self.m.number ← one // inherited The FemaleSingularNoun construction is a subcase of the constructions SingularNoun (Fig. 2.24) and Female (Fig. 2.25). SingularNoun is a subcase of the constructions Noun (Fig. 2.26 below) and Singular (Fig. 2.8 above). Noun is a subcase of the more general construction Nominal (Fig. 2.27 below), which is a subcase of the HasNominalFeatures (Fig. 2.28), which in turn is a subcase of HasAgreementFeatures (not shown). Fig. 2.24: the SingularNoun construction construction SingularNoun subcase of Noun, Singular constructional: NominalFeatureSet // inherited constraints self.features.number ← singular // inherited form: WordForm // inherited meaning: RD // inherited constraints self.m.number ← one Fig. 2.25: the Female construction construction Female subcase of HasAgreementFeatures constructional constraints self.features.grammgender ← female We also have to provide a representation of the article la, which is a subcase of the FemSgDefiniteArticle construction. Both FemSgDefiniteArticle and La are shown in Fig. 2.29 and Fig. 2.30, respectively. 119 Grounding Meaning in Everyday Experience in the World Fig. 2.26: the Noun construction construction Noun subcase of Nominal constructional: NominalFeatureSet // inherited form: WordForm meaning: RD // inherited Fig. 2.27: the Nominal construction construction Nominal subcase of HasNominalFeatures constructional: NominalFeatureSet // inherited meaning: RD Fig. 2.28: the HasNominalFeatures construction construction HasNominalFeatures subcase of HasAgreementFeatures constructional: NominalFeatureSet Fig. 2.29: the FemSgDefiniteArticle construction construction FemSgDefiniteArticle subcase of SgDefiniteArticle, Female constructional: NominalFeatureSet // inherited constraints self.features.grammgender ← female // inherited self.features.number ← singular // inherited form: WordForm // inherited meaning evokes RD as rd // inherited constraints rd.accessibility ← uniquely-identifiable // inherited rd.number ← one // inherited 120 2. From Cognition to Grammar Fig. 2.30: the La construction construction La subcase of FemSgDefiniteArticle form: WordForm // inherited constraints self.f.phon ← /la/ meaning evokes RD as rd // inherited constraints rd.accessibility ← uniquely-identifiable // inherited rd.number ← one // inherited The La construction and the Pagnotta construction are then combined in the DeterminerPlusNP construction133, shown in Fig. 2.31 below. Fig. 2.31: the DeterminerPlusNP construction construction DeterminerPlusNP subcase of NPSpecifierPlusNP constructional: NominalFeatureSet // inherited constituents s: Determiner np: Np // inherited constraints self.features ↔ np.features // inherited self.features ↔ s.features // inherited form constraints s.f before np.f // inherited meaning: RD constraints self.m ↔ np.m // inherited self.m ↔ s.m // inherited The grammgender role of the NominalFeatureSet schema in the constructional block of FemSgDefiniteArticle is used to specify the fact that the grammatical gender of this noun is female because Italian, unlike languages like English, has no neuter gender; consequently, any noun in Italian must be classified as either masculine or 133Indeed, FemSgDefiniteArticle is a subcase of SgDefiniteArticle and Female. SgDefiniteArticle is a subcase of DefiniteArticle and Singular. Definite Article, in turn, is a subcase of the Determiner construction. 121 Grounding Meaning in Everyday Experience in the World feminine, on the basis of mainly historical, formal, or semantic features 134. Nevertheless, in the meaning constraint, the natgender role is filled by a neuter value, since a loaf, quite trivially, can be neither male or female. The definite article La adds (in the meaning block) the information that the referent of the NP is uniquely-identifiable. Then, the DeterminerPlusNP construction combines the two constructions just outlined, establishing a formal constraint on the order of the constituents (i.e. the determiner must precede the noun) and a semantic constraint which binds the meaning of the complex construction to the meaning of both constituent constructions. This example can be analyzed as instantiating various constructions, including a transitive argument structure construction. When these instantiated constructions unify, they produce a SemSpec, which supports the simulation of the event described by the sentence. Such a SemSpec consists of a set of schemas, value constraints and bindings. The SemSpec includes an EventDescriptor schema, which provides many key simulation parameters. Its profiledProcess role is bound to the meaning of TagliareSimplePast3sg. Such a profiledProcess is one of CauseEffectAction, and the x-net in this process is one of cutting. EventDescriptor's eventType role is bound to the meaning of TransitiveCEA. In this quite prototypical transitive construction, the meaning of the argument structure is closely bound to that of its verb constituent. Consequently, the eventType in this example is also bound to CauseEffectAction. Therefore, the A-S construction provides information about the general type of event being described, and the verb supplies additional information about the specific processes involved in this event. Furthermore, the unification of TransitiveCEA, TagliareSimplePast3sg, Declarative and instantiated NP constructions results in several bindings associated with each of the participants of this event. Precisely, the causer role of CauseEffectAction is bound to: – the protagonist of CauseEffectAction; – the actor and protagonist of ForceApplication; – the referent called Sara; – the profiledParticipant. At the same time, the affected role of CauseEffectAction is bound to: – the protagonist of CauseEffectAction's process2; – the actedUpon of ForceApplication; – the entity referred to as la pagnotta. This SemSpec will support an enactment of an event in which a female human causal actor performs some kind of forceful cutting action on a loaf of bread, affecting it in a particular way. This event is described from the perspective of this actor/causer, and should be simulated from the same perspective. The process of enactment of this event can produce some more inferences regarding, for instance, the amount of energy the actor spent, the kind of instrument she used, the purpose on which she performed this action, and so on. Of course, the production of inferences by the language understander will depend on variable factors, such as the communicative context, the language user's experiences with the action defined by the verb, etc. However, this grammar (like most ECG grammars currently available) is meant to provide just the core conceptual structures which are likely to be used by a range of users in several different contexts. Concluding the present section, it must be highlighted that, in order to provide a brief 134On the different gender systems of the world's languages, see Corbett (1991). Useful information on the organization of gender in Italian can be found in Doleschal (2006). 122 2. From Cognition to Grammar account of the ECG formalism at work, a very simple instance of an Italian active transitive force-exertion construction was chosen. There are two main reasons for such a choice: simplicity's sake and our scope to “shadow” Dodge and Bryant's analysis, outlining the functioning of (a slightly modified version of) their grammar in the most consistent way possible. In the following chapter, Dodge and Bryant's grammar will be used for the analyses of Italian caused-motion constructions. This kind of work originates from the assumption that such a grammar should be largely adequate for the description of Italian data. Nevertheless, since there are various substantial differences between English and Italian, we can be quite sure that the analysis of different Italian constructions using Dodge and Bryant's grammar will be more difficult and exacting than it was in the case of the sentence examined above. As a result, it must be kept in mind that such a grammar may be modified according to our necessities, also considering insights from previous work on ECG (with particular reference to Gilardi's Starter2 grammar). Indeed, it should not be forgotten that, while Dodge and Bryant (forthcoming) is taken as our main reference point, the present study has not the relatively narrow scope to evaluate the descriptive power of their grammar, but the more general goal to assess the explanatory adequacy of the ECG model on the basis of our Italian data. 123 3. An ECG analysis of Italian caused-motion constructions 3. An ECG analysis of Italian caused-motion constructions 3.1. Introduction: data and methodology The present chapter will be dedicated to an analysis of Italian data, in order to test the descriptive and explanatory adequacy of the ECG model to account for grammatical phenomena of this language. More precisely, we shall be dealing with some instances of Italian caused-motion constructions with verbs of force-exertion and volitional agents, making use of an adapted version of Dodge and Bryant's grammar briefly outlined in the previous chapter. In ECG, the label “caused-motion construction” applies to any sentence-level construction which expresses an action whereby an entity causes another entity's movement through space by means of an act of force transmission. Exploiting a slightly modified version of Goldberg's (2006) notation, we can assert that a causedmotion construction will show the form represented in (1) (in the active voice, only), the meaning reported in (2), and the argument structure illustrated in (3) below1: (1) (2) (3) Subj V Obj Oblpath X causes Y to move Zpath CAUSE-MOVE (causer patient path)2, 3 The choice to work on Italian caused-motion constructions in our case-study chapter drew inspiration from Goldberg's work. In her 1995 influential publication, Goldberg claims for the existence of some particular sentence-level constructions, whose interest lies in their being inherently meaningful, without reference to the words fulfilling them. In order to corroborate and fortify her statement, in the second part of her study she offers a detailed investigation of four particular grammatical constructions which are ubiquitous in the English language: the distransitive construction, the caused-motion construction, the resultative construction, and the “way” construction. In the following years, several scholars have adopted Goldberg's perspective and a lot of work have been developed on these kinds of constructions (especially in English); moreover, a few variants of Goldberg's model have arisen. Nevertheless, very little has been done on Italian from a constructionist approach. The fact that a very small number of constructionist studies on Italian are currently available drove us to decide to direct our efforts to study a phenomenon of the grammar of this language, which represents the author's mother-tongue language. In particular, the scarceness of constructional work on Italian sentence-level constructions led us to choose one of the constructions analyzed by Goldberg (1995) for English and try to provide an account of the corresponding Italian construction. The choice to investigate the caused-motion construction was 1 However, as specified in § 2.3.1.2, it should not be forgotten that Goldberg adopts the label construction in a bit stricter sense than it is used in ECG. 2 The two arguments in bold type are obligatory, while the third is optional. 3 Since we focus our attention on constructions including verbs of force-exertion which require volitional, animate actors, in this chapter we may sometimes use the terms “causer” and “agent” interchangebly. 125 Grounding Meaning in Everyday Experience in the World quite straightforward, for in Italian the “way” construction does not exist, resultative constructions are less frequent and their status is unclear (see Broccias 2003), and ditransitive constructions only occur with dative pronouns4. On the contrary, causedmotion constructions are not so rare in this language, representing a suitable and interesting phenomenon to observe, despite their frequency being much lower in Italian than in English. Italian caused-motion constructions are quite different from their English counterparts for at least one main reason, i.e. they occur with a much more restricted range of verbs. First of all, this is probably due to the fact that Romance languages are verb-framed languages, while Germanic languages are satellite-framed languages5; therefore, an English sentence such as (4) (drawn from Goldberg 1995: 161) can be translated into Italian as (5), with a more generic motion verb, or as in (6), using a circumlocution with the verb fare (“to make”): (4) (5) Frank sneezed the tissue off the nightstand. Frank spinse la salvietta dal tavolino con uno starnuto Frank push:PRT.3SG the.FSG tissue.SG from-the.FSG nightstand.SG with a.MSG sneeze.SG “Frank pushed the tissue off the table by sneezing” Frank fece cadere la salvietta dal tavolino con uno Frank make:PRT.3SG fall:INF the.FSG tissue.SG from-the.MSG nightstand.SG with a.MSG starnuto sneeze.SG “Frank made the tissue fall off the table with a sneeze” (6) Moreover, while in English it is not uncommon for the subject of an intransitive-motion construction to appear also as the object of a caused-motion construction which involves the same verb (in the same form), this phenomenon does not seem to occur in Italian. For instance, while in English the sentences in (7) and (8) are both fine, the Italian counterpart of the intransitive-motion construction in (7) takes itself the form of an intransitive-motion construction, as you can see in (9), but the caused-motion construction in (8) would be translated differently, for instance using a circumlocution with the verb fare (“to make”) + infinitive, as in (10) below: (7) (8) (9) The bread slid off the table. Linda slid the bread off the table. Il pane scivolò dal tavolo. The.MSG bread.SG slide:PRT.3SG from-the.MSG table.SG “The bread slid off the table” Linda fece scivolare il pane dal tavolo. Linda make:PRT.3SG slide:INF the.MSG bread.SG from-the.MSG table.SG “Linda made the bread slide off the table” (10) 4 Not all scholars recognize such constructions as ditransitives (see Goldberg 2006: 76, footnote 2). 5 In Talmy's (2000) typological distinction, satellite-framed languages (mostly) encode manner in the verb stem and path in verbal satellites, while verb-framed languages (prevalently) encode path in the verb stem and manner in adverbial phrases or gerund forms. The reader is advised to see also Ochsenbauer and Hickmann (2010). 126 3. An ECG analysis of Italian caused-motion constructions In both English and Italian, caused-motion constructions are very often exploited convey figures of speech, especially metaphors. Indeed, this kind of construction most frequently found in metaphorical predication, when the situation as a whole mapped from a source domain (an actual caused-motion action) to a target domain more abstract concept), as in the example in (11): (11) Il divorzio gettò Vito nella disperazione. The.MSG divorce throw:PRT.3SG Vito into-the.FSG despair.SG “The divorce threw Vito into despair” to is is (a However, in the present study we shall concentrate on (quite typical) caused-motion constructions which occur in literal predication, leaving aside their possible extensions to figurative language. In other words, we shall deal with constructions which express actual caused-motion events. As an example, consider the sentence reported in (12) below: (12) Il rapinatore buttò il portafogli a terra. The.MSG robber.SG toss:PRT.3SG the.MSG wallet.SG to ground.SG “The robber tossed the wallet to the ground” The sentence-level constructions analyzed in the next section were built on analogy with instances of real language in use drawn from ItWaC (Italian Web as Corpus, see Baroni and Kilgariff 2006), a very large electronic corpus containing almost two billions Italian words crawled from the web, created by Marco Baroni and then updated by Francesca Masini and other scholars affiliated to Roma Tre University6. The process of chosing the data to analyze went through a few stages. Our first step was the preparation of a list of twelve Italian force-exertion verbs7, selected primarily on the basis of our own intuitions about their likelihood to occur in caused-motion constructions, which was later tested through the investigation of ItWaC, exploiting the online corpus query system Sketch Engine (http://www.sketchengine.co.uk). Then, we gathered a sample of about onehandred sentences expressing caused-motion (evenly divided between active and passive), selected in a somewhat random fashion, but being careful to pick only those involved in literal predication, thus avoiding the ones used in a figurative sense. Then, we further restricted the number of our sentences selecting just a few of them (some active and some passive) to serve as a model for the sentences to be illustrated in §3.2. These sentences, which showed roughly a common form, argument structure, and core meaning, were chosen on the basis of their level of simplicity and, in a sense, “normality” of the message they convey8. For instance, a sentence like the one shown in (13) below looked much more natural than that illustrated in (14), whose meaning would probably result a bit odd (at least out of context): (13) Michela lanciava sassi nell' acqua. Michela throw:IMPF.3SG stone.PL into-the.FSG water.SG “Michela was throwing stones into the water” 6 Università degli Studi Roma Tre is an Italian state-supported university based in Rome. 7 The list included the following verbs: lanciare, spingere, tirare, schiacciare, sollevare, posare, premere, gettare, scagliare, trascinare, scaraventare, trainare. 8 Of course, judgments of “normality” are highly dependent on subjective experience. As a result, different analysts might evaluate the degree of “normality” of the same sentence differently. 127 Grounding Meaning in Everyday Experience in the World (14) Il giovane lanciava bottiglie contro la porta del cinema. The.MSG youth.SG throw:IMPF.3SG bottles.PL against the.FSG door.SG of-the.MSG theater.SG “The youth was throwing bottles against the theater doors” Therefore, the former was considered more suitable than the latter for our purposes. We then decided to make use of eight declarative sentences with a rather simple structure (each involving a different verb, drawn from the list mentioned above), half in the active voice and half in the passive voice. The number of sentences to be actually analyzed, which is fairly small, was determined to cope with the fact that carrying out an ECG analysis of even a very short clause is rather costly, in terms of both time and space9. There was no specific criterion relating the nature of a particular verb with the voice of the example-sentence it appears in, from this aspect our choice was rather arbitrary. We also decided that each verb should be in a different tense: simple future, simple past, present perfect, imperfect, future perfect, passive simple past, passive present perfect, and passive recent pluperfect. While we recognize that, from a usage-based perspective, these tenses are not equivalent10, this choice was made in order to cover a wider range of cases, in spite of the fact that their levels of frequency differ greatly. Since in Italian verbs agree with subjects in number and person, we also wanted for the subjects of our sentence-level constructions to cover all the six persons11. Furthermore, as already specified above, we wanted for our the agents of the actions expressed by our sentences to be volitional. In addition, we required that the path argument in all sentences and the agent argument in passive constructions be always expressed. Finally, we established that the path arguments should be expressed by prepositional phrases and that at least five different types of literal motion should be illustrated in our data. As you can see, these are very strict requirements, but they were necessary in order to allow us to provide an account of homogeneous data, which embody a very circumscribed and clearly defined phenomenon, i.e. the linguistic expression of quite prototypical volitional forceful caused-motion action in Italian. As a consequence of the imposition of these restrictions, the choice of our target sentences cannot be seen as either completely based on introspection or entirely based on the observation of empirical data: the instances of real language found in ItWaC were exploited as a model in order to avoid to build implausible examples, but we had to modify (more or less heavily, depending on the particular case) the instances drawn from the corpus, since the specificity of the requirements we posited for our target data were too strict to be simultaneously satisfied by the examples extracted from the corpus. Once we had collected our eight target sentences, we analyzed them (beginning from those with the simplest structure) adopting an ECG approach similar to that exploited by Ellen Dodge in some studies of hers (Dodge 2010a, 2010b; Dodge and Bryant forthcoming). In particular, we made use of a grammar similar to the one used in Dodge and Bryant (forthcoming), opportunately modified to handle the properties of the Italian language and adjusted in order to be more consistent with the network model of polysemy (see below). Building such a grammar was made significantly easier by the possibility to refer to the taxonomies included in Luca Gilardi's Starter2 grammar. The 9 As Feldman (2006: 293-294) points out, ECG offers long explanations for short examples, due to the fact that ECG scholars conceive language as inherently complex. 10 Indeed, the respective constructions will remarkably vary with regard to their degree of entrenchment and conventionalization (see §2.3.1 above). Indeed, the pluperfect is not as common as the other tenses. 11 Since we have to deal with eight sentences, two persons will be represented twice in our study. 128 3. An ECG analysis of Italian caused-motion constructions starting point of our analysis lies in the formulation of the two hypotheses expressed below: • a single A-S construction will be able to capture generalizations about prototypical Italian caused-motion constructions, i.e. those whose verb's meaning perfectly coincides with the construction's meaning. At the same time, possible radial extensions of the relevant A-S construction may be posited to handle less prototypical caused-motion constructions (those whose verb's meaning correspond to a part of the A-S construction only); • a limited set of schemas can account for the core meaning of Italian causedmotion constructions. At first glance, these hypotheses might be somewhat misleading for the reader; therefore, a sort of “disclaimer” may be necessary. The proposed need for a small number of schemas and constructions should not be interpreted as a claim for minimalism. Indeed, it is just being suggested here that a single A-S construction (with its radial extensions) and a limited number of schemas can be sufficient to handle the phenomenon under consideration, which is a very restricted target. Actually, as pointed out by Dodge and Bryant and hopefully will become clear in the next sections, ECG can be seen as closer to the maximalist (rather than the minimalist) pole of the ideal minimalism-maximalism continuum. It has to be kept in mind that the present study should not be strictly considered an analysis of Italian caused-motion constructions. Indeed, this piece of work has the more general scope to assess the analytical potential of the ECG model on the basis of Italian data; from this perspective, the circumscribed topic of caused-motion constructions can be intended as a Trojan horse to evaluate the (descriptive and explanatory) adequacy of this approach to deal with some grammatical phenomena of the Italian language. The main issues of the present study have to do with compositionality. Indeed, compositionality is a foundational notion in ECG: as highlighted in the previous chapter, the meaning of a construction is seen as a gestalt with a complex internal structure, and the meaning of the whole is considered to be partially, but critically, provided by the meaning of its constituents. However, since compositionality is a node which has long been a matter of debate among (both cognitive and non-cognitive) linguists, it seems worth spending some time on this topic. According to the traditional point of view, the meaning of a complex unit of language, e.g. a sentence, is a function of the meaning of its constituent items: the sum of the meaning of the parts is seen as sufficient to give rise to the meaning of the larger unit. In this perspective, a complex unit is built from scratch combining the meaning of the lexical items present in a sort of mental dictionary according to the rules included in a mental grammar (the so-called “building-block metaphor”). This point of view is normally adopted by linguists working in the generative tradition. Cognitive linguists generally reject this totally compositional perspective. As mentioned in §2.3.1.2, Goldberg (1995) addressed this issue in detail, arguing that A-S constructions represent symbolic units in their own right, that they are inherently meaningful and play a crucial role in the determination of the meaning of a sentence. Therefore, while she recognizes that single words contribute, to a large extent, to the meaning of a sentence, Goldberg asserts that a purely “words and rules” approach is not sufficient to provide the meaning of a complex unit, i.e. word-level units contribute meaning to a sentence, but not all the meaning. In order to “get the whole picture”, Goldberg states that the full meaning of a sentence is provided by the integration of the 129 Grounding Meaning in Everyday Experience in the World semantics of the A-S construction with the semantics of the verb (and its arguments). This weakened version of compositionality is widely spread among cognitive linguists. Nevertheless, there is a growing body of empirical evidence which allows the claim for more radical constructionist approaches to grammar (one of them being RCG, for instance), which further reduce or even deny the role of compositionality in determining the meaning of a complex unit. In the next lines, we shall briefly outline Cristofaro's (2008) proposal for a fully noncompositional model of grammar. Relying on her analysis of complement sentences12 in a corpus of Ancient Greek and a series of studies previously carried out by herself and other functional and cognitive linguists, Cristofaro proposes the adoption of an entirely noncompositional constructionist approach to grammar. In her model, larger constructional units are taken as the only basic primitives having a specific meaning in their own right, while it is argued that its individual components may (but this is not a strict requirement) be assigned meaning derivatively via process of abstraction over the larger constructions in which they occur. Though such a radical approach may probably sound “iconoclast”, Cristofaro provides strong arguments to justify its adoption, showing that empirical data do not support the assumption that the meaning of the components of a larger construction are independently stored in the language users' mind; instead, it is the meaning of the complex construction as a whole which is stored. For instance, she argues that the combination of perception predicates and participial complements in Ancient Greek expresses sensory perception of an ongoing situation (see (15)); at the same time, this meaning cannot be found when either participial complements are combined with other complement-taking predicates or perception predicates are combined with different complementation patterns (see (16)), which shows the combination of a perception predicate and an indicative complement, and the complement sentence as a whole just expresses the acquisition of a process): (15) Kaí ē gunè eporaî min exiónta and the woman saw:3SG 3SG.ACC go.out:PRES.PART-ACC.MSG “and the woman saw him go out” (Herodotus, 1.10.6; Cristofaro 2008: 575) Horaís ō Melete hōti sigaîs kaí ouk écheis eipeîn? see VOC Meletus that be.silent:PRES.IND.2SG and NEG have:PRES.IND.2SG say:AOR.INF “Do you see, Meletus, that you are silent and cannot tell?” (Plato, Apology, 24d; Cristofaro 2008: 575) (16) This fact favors Cristofaro's proposal according to which meaning should just be seen as the meaning of the complex structure as whole. Her position can be summarized by the following excerpt: One may assume that particular aspects of meaning are part of the mental representation of the larger combinations in which individual syntactic patterns occur, rather than the mental representation of individual syntactic patterns as such (...) This analysis is supported by distributional evidence, in that particular meanings are only found with particular combinations of complement-taking predicates and complementation patterns. At the same time, by assuming that meaning is associated with particular combinations involving a given complementation pattern, this analysis accounts for the meaning differences found when other complementation patterns are involved. (Cristofaro 2008: 589) 12 “By a complex sentence is meant (...) a complex sentence where a clause (the complement clause) functions as an argument of a main predicate (the complement-taking predicate)” (Cristofaro 2008: 571). 130 3. An ECG analysis of Italian caused-motion constructions Cristofaro also specifies that in her approach speakers are supposed to abstract particular aspects of the meaning of complex structures as a whole and associate these aspects of meaning to individual components of complex structures, rather than deriving the meaning of the whole by combining the meaning of the parts. Therefore, only the larger construction has inherent meaning, while the meaning of its components is merely derivative. This approach allows the analyst to account for the difference of meaning encountered in distinct patterns, and this view is consistent with the usagebased model of grammar advocated by Langacker and Croft (among others) and recent developments in (cognitive) lexical semantics (see e.g. Taylor 2003). ECG studies generally assign an important role to compositionality, but this is not necessarily at odds with radical perspectives such as Cristofaro's: indeed, the kind of compositionality adopted by ECG scholars13 is rather far from the strict compositionality principle advocated in the formal linguistics tradition (see Feldman 2010). In our opinion, the key difference between totally noncompositional constructionist models and the ECG approach, which allows compositionality a role in the process of “meaning construction”, lies in the fact that these approaches tackle grammatical issues taking different starting points. From an ECG perspective, the issue of compositionality is discussed in detail by Ettlinger (2005). In his paper, Ettlinger proposes a general theory of “constructional compositionality” which posits the existence of two clines: one concerning composition, the other regarding productivity. These continua are related to the level of abstractness of the different constructions. Ettlinger, following the proponents of the usage-based model of language, assumes that a high token frequency leads to the entrenchment of a construction, and a high type frequency accounts for its productivity. In Ettlinger's perspective, abstract constructions are productive since there is a wide range of options for fulfilling its constituents with other constructions14, while less abstract constructions are less productive since there is a lesser degree of variation from one instance of the construction to another. Consider the two constructions below (see Ettlinger 2005: 26): (17) (18) V NP kick the bucket The reader will probably have no doubt that the A-S construction in (17) is more productive than the idiomatic construction in (18). Ettlinger also proposes the existence of a cline of composition from dynamically composed to statically stored constructions which correlates to their abstractness. For instance, since the meaning of idiomatic/idiosyncratic constructions is not predictable from their component parts, these expressions must be explicitly part of the grammar of the speakers15 and it will be little productive and closer to the “static” pole of the continuum. On the other hand, constructions like the combination of a verb and an A-S construction will be quite productive. Ettlinger argues that it is unlikely that every verb + A-S construction exists in the speaker's grammar prior to the time of the utterance; therefore, the composition of 13 With the possible exception of Bretones Calleja (2004). 14 These constructions are those traditionally labeled as “categories”. Nevertheless, converging with Croft (2001), Ettlinger and other ECG proponents advocate that such syntactic categories do not exist outside the larger constructions they appear in, i.e. they are construction-specific. 15 Ettlinger also clarifies that identifying an independent construction does not mean to neglect its relationship to other constructions. 131 Grounding Meaning in Everyday Experience in the World verb and A-S construction is productive and should be considered as “dynamic”. The composition of two given constructions will lie somewhere along these two clines: idiosyncratic constructions will have static representations in the grammar and will be little productive, while more regular and predictable combinations are likely to be dynamically composed by rules of unification16. The main issues addressed in Ettlinger's study are summarized as follows: Because grammars are a highly structured, intricately organized, system of constructions, all constructions involve inheritance. However, not all inheritances are created equal. Some inheritances are dynamic and involve ad-hoc unification constraints by the speaker. This type of inheritance and composition, dynamic composition, provides for the creative, productive properties of language. The other type of inheritance involves learned constructions with partial inheritance for pieces of the grammar that are highly conventionalized. This type of inheritance accounts for the idiomatic, exemplar-based properties of language. (Ettlinger 2005: 35) In other words, a sentences like Mary ate an apple does not need to be stored whole in the speaker's mental grammar, since it can be simply regarded as an instantiation of well-entrenched abstract constructions such as the Declarative construction, the TransitiveCEA construction, the verb to eat etc. and, as a result, it can be composed effortlessly. At the same time, this process of dynamic composition also accounts for the speakers' creativity in the use of language17. On the other hand, a highly idiosyncratic construction like, for instance, by and large cannot be composed in this way, and it must be statically stored in the grammar. The two approaches just outlined above are not mutually exclusive. Cristofaro's (2008) top-down model highlights the importance of rejecting the traditional view according to which language is made up of a mental dictionary of words which are then combined together according to the strict rules specified in a mental grammar. Moreover, it also clarifies that the meaning of a larger construction should not be seen as derived from the meaning of its parts; on the contrary, the larger construction is the semantic primitive and the meaning of the parts is abstracted from their occurrence in the whole. On the other hand, Ettlinger (2005) and other ECG scholars claim for the importance of compositionality, but compositionality is understood in a way which seems compatible with Cristofaro's approach. As mentioned above, consistent with the usage-based model of grammar, a construction is considered to be entrenched when it occurs with a high level of token frequency, and it is considered productive when it occurs with a sufficient level of type frequency. This means that smaller constructions are stored via abstraction from the larger constructions they appear in. Nevertheless, once they are stored, they can be used in novel and creative ways. For instance, once that the caused-motion construction is part of the grammar of a speaker, they can use it filling its “constituent slots” with the stored lexical construction they prefer. An analogous condition holds with regard to lexical constructions: once the speaker stores the meaning and the context of use of a word, they can use these words to create novel sentences. In short, we posit a constant interaction between the “whole” and its “parts”. It is in this sense that here we speak of 16 This perspective is quite close to the notion of “sanctioning” used by scholars who adopt Langacker's approach: “schematic patterns sanction both established and novel structures, and a novel structure is automatically acceptable to the degree that it directly elaborates a well-established, elaboratively close schema or set of schemas” (Tuggy 2007: 99-100). 17 On the pervasiveness of creativity in language use, see the contributions on English collected in Maybin and Swann (2006). 132 3. An ECG analysis of Italian caused-motion constructions compositionality in grammar18. Needless to say, this does not mean that any stored construction can be combined with any other stored construction, which would lead to a completely free inventory of units. On the contrary, this is a point which underlines the need for top-down constraints, which allows the larger construction to constrain (at a higher, schematic level) the range of its possible constituents. Therefore, in our study we shall assume that “analysis should proceed in both bottom-up and top-down fashion, with surface features of the utterance providing bottom-up cues to the constructions involved, and cued constructions potentially supplying top-down constraints on their constituents”, as pointed out by Bergen and Chang (2005: 174). The need for bottom-up analysis in ECG work is also due to the fact that this approach, as already stated in the previous chapter, sets apart from other cognitive and constructionist approaches to grammar in being prominently interested in modeling online language processing, rather than dealing with stored language knowledge. This is a crucial factor, since language processing is a dynamic activity, while language knowledge represents the (relatively) static repository of constructions which informs and constrains the process of language comprehension (and, barely differently, production). Dealing with language processing, a holistic and top-down model seems necessary, but hardly sufficient. Indeed, in both spoken and written language, the language understander is not provided with the larger construction (say, a sentence) altogether; rather, its component constructions are presented sequentially. This fact led several scholars (see e.g. Bicknell and Levy 2009) to posit that language is processed incrementally, through the interpretation of partial structure and expectations for future input, i.e. people begin to process linguistic input as soon as they receive it, trying to understand partial information and making predictions on what next input is to come (and this often leads them to misunderstandings or longer processing times, see e.g. Narayanan and Jurafsky 2001). Then, when the whole larger construction is processed, the provisional interpretation can be reviewed, refined, confermed, or modified; furthermore, it can finally be more confidently segmented into phrases. We have already highlighted the importance of the combination of a verb and an A-S construction in capturing generalizations over a group of verbs which share a core meaning, but the role of compositionality is also crucial both at a higher and at a lower level. On the one hand, as seen in the previous chapter, the A-S construction represents the VP of the sentence, while the instance as a whole is handled by a clause-level construction. This situation does not raise any problems until we have to deal with a sentence involving a compound verb. Then, we will need to bridge the gap between the clause-level and the A-S construction. As will be shown in the following section, Gilardi's provides us with a further schematic construction which permits us to solve the problem quite comfortably. Another possible problem at this level may be represented by the fact that Italian is an instance of what linguists (especially in the generative tradition, see e.g. Haegeman 1991) usually label pro-drop languages, i.e. a language which allows the use of implicit subjects (in fact, the Italian language makes extensive use of this strategy). It will be argued that this situation can be handled introducing a specific subcase construction for each person's null subject pronoun, providing an inventory-based solution to the problem, simple but functional and 18 This approach is also adopted by Schneider (2010) in his detailed account of Hebrew verbal morphology, and it is also very similar to Chen's (2008) “Gestaltist” approach to Chinese verb-verb compounds. Furthermore, a compatible view is at the core of Gurevich's (2006) extensive study on Georgian morphology.. 133 Grounding Meaning in Everyday Experience in the World consistent with the approach adopted here19. At the same time, the meaning of a particular phrase which appears in a sentence cannot prescind from the meaning of the lexical items it includes. It is necessary to specify how these lexical constructs interacts with both the other components of the phrase and the schematic structure of the phrase itself, e.g., which piece(s) of information each of them contributes to the meaning of the NP as a whole. This point inevitably intersects the problem of the differences between Italian, the language under consideration in this study, and English, the language analyzed in the bulk of ECG literature which represents the main background for our work. Indeed, given the different structures of the two languages, a compositional model has to deal with different issues. The main potential sources of problems for this approach are represented by the richness of Italian morphology on the one hand, and the lack of a neuter gender on the other, which unavoidably leads to an inconsistency between the grammatical gender of nouns denoting inanimate entities and the natural gender of these entities themselves. Nevertheless, given the high level of flexibility of the ECG formalism, there is no reason, in principle, to believe that these problems cannot be answered satisfactorily. Consider the following NP: (19) Il divano The.MSG couch.SG “The couch” This NP is made up of two constituents: a definite article and an NP constituted by a simple noun. The definite article il shows the following grammatical features: male gender and singular number. Therefore, it suggests that the following noun will share the same features. Moreover, the article has the form of a word with a particular phonological structure. The meaning of the article evokes a referent, providing some information on its semantic features: it refers to just one entity, which is uniquely identifiable. The noun divano shares the same grammatical features of the preceding lexical item, but obviously it has a different phonological form. With regard to its meaning, this noun denotes a referent, providing information on its ontological category (a couch), gender (neuter), and number (singular). There are a couple of semantic issues to be carefully noticed here. First, there is a direct relation between the two pieces of semantic information provided to the NP by the noun alone, i.e. the category and the gender of the referent. Indeed, knowing that the role category of the RD schema is filled by the value “couch” (which is a subcase of the category “inanimate”) lets us know that the referent is an inanimate being. Since divano is a noun which is used only to denote inanimate beings, we can conclude that the natural gender of the referent is neuter. The second point regards the fact that while the information about the number of the referent is shared between the article and the noun, the information about its accessibility level is provided by the article20 alone. As can be observed in the analogous example sketched in the previous chapter, an ECG grammar along the lines of Dodge and Bryant's is able to capture these semantic issues and include them in the analysis of the phrase. Of course, according to the specific phrase 19 As Evans and Green (2006: 502) underline, “Implicit elements have no phonetic realization but represent speaker knowledge of grammatical categories like noun and verb, subcategories (for example, count and mass noun), and grammatical functions (also known as “grammatical relations”) like subject and object.” 20 With regard to accessibility, Italian and English NPs are analogous. 134 3. An ECG analysis of Italian caused-motion constructions one is dealing with, an ECG analysis may be more or less difficult and exacting, as we shall see in §3.2 below. However, all in all, ECG seems to be on the track to provide a satisfactory account of this kind of syntactic unit. With regard to studies on language processing, during the last decades this issue was addressed not only by scholars working within the NTL research program, but also by a number of other academics working in the fields of psycholinguistics, cognitive psychology, and artificial intelligence21, some of whom developed computational models of human language processing (critical reviews can be found in e.g. Christiansen and Chater 2001; Chater and Manning 2006; Onnis et al. 2006) based on probabilistic factors emerging from the analysis of actual use of language (usually exploiting corpus data, but sometimes also relying on the results of previous psycholinguistic experiments). Since many of these models, though not directly related to the ECG model, are largely consistent with it and can also offer useful insights for our scope, in §3.3 some of them will be taken as additional reference point for the illustration of the interaction between bottom-up cues and top-down constraints in the comprehension of the sentence-level constructions analyzed in §3.2. As a result, taking the computational work developed by NTL researchers and colleagues as our basis (see e.g. Narayanan and Jurafsky 1998, 2001; Mok and Bryant 2006; De Beule and Bergen 2006), we shall also pay attention to the insights offered by computational models of language processing and learning built within the connectionist paradigm (see e.g. McClelland and Elman 1986; Elman 1993; Christiansen et al. 1998; Allen and Seidenberg 1999; Rohde and Plaut 2003; McClelland and Cleeremans 2009) and possibly other probabilistic modeling work. The discussion of the problem of compositionality also intersects the issue of (both lexical and constructional) polysemy, another topic which has long been a matter of debate among linguists. Here, we shall adopt a point of view close to Taylor's (2003: ch. 8). In such a piece of work, Taylor points out that the compositionality principle, which captures the fact that language users can create an indefinite number of novel expressions by combining smaller units in certain ways, is complicated by the existence of polysemy. Indeed, in order to compute the meaning of a complex expression, we need to select just one of the senses of each polysemous word in the expression. Taylor shows that, since the verb open can assume fifteen different senses, and the noun window can assume eight, the short clause open the window can assume 15*8 = 120 distinct senses, which is rather unreasonable. The problem also arises of distinguishing polysemy (a linguistic unit shows more than one sense) from contextual modulation (different construals of a single sense emerge due to pragmatic factors, see Croft and Cruse 2004: ch. 5). Several approaches have tried to solve this problem. One of them favors the proliferation of different word senses, but it does not specify the relevant parameters to distinguish between a new sense and a contextual modulation of a single sense. The opposite approach minimizes polysemy in favor of contextual interpretation, yet it fails to define the sense of the word with sufficient generality to cover the full range of different uses. Within the Cognitive Linguistics framework, the problem was remarkably simplified in the late 80s, when Langacker (1987) elaborated the network model, according to which the senses of a unit constitute the nodes of a network, linked horizontally by relations of similarity, and vertically by a sort of “subcase” relation 22. 21 See Gaskell (2007a: sections I-III) for a substantial body of work on the phenomenon. 22 See also Tuggy (2007) for an illustration of the practical use of the network model in the analysis of a number of grammatical issues. 135 Grounding Meaning in Everyday Experience in the World The lowest nodes of the network can be seen as the most specific senses of a unit, while the highest node can be seen as the most abstract representation of the meaning of that unit. Consistent with the usage-based model, specific senses of the item arise first, coming to light via entrenchment due to high token frequency; subsequently, the more schematic representation arises as a result of an operation of abstraction performed by the language user. It is not the case that the most abstract representation is necessarily invoked in the understanding of contextual variants; on the contrary, using a computer-related metaphor, we may say that each specific sense of a unit, in context, can be accessed randomly rather than sequentially by the language user, without having to examine all the possible senses of the unit23. For instance, the meaning of open and window in an expression such as open the window will be immediately understood by a native (or proficient) English speaker, without any need to take all the possibilities into consideration. As Taylor points out: One consequence of adopting the network model is that the question of whether a word is polysemous or not turns out to be incapable of receiving a definite answer. The answer will depend on the level of abstraction at which the word is accessed (…) To consider only the particular to neglect the schematic – and vice versa – impoverishes our understanding of word meaning. (Taylor 2003: 167) An adapted version of the network model will be exploited in the present study in order to deal with polysemous units, an issue often either neglected or taken for granted in ECG work, but central to language understanding (and production). We shall address the problem of prepositional polysemy, since the role of prepositional phrases in caused-motion constructions is strikingly important, providing information about the direction of the motion of the patient participant. The Italian grammar include a limited set of prepositions, most of which show a number of different, but related senses. If we confine ourselves to considering spatial prepositions, which represent the majority of the constructions we shall be dealing with, a very general point can be made: we can observe that a bundle of Italian prepositions (such as a, da, su, in, tra) can take either a directional or a stative meaning, depending on the kind of construction they are used in. Moreover, passive caused-motion constructions show the presence of the causative PP da, used to denote the causer of the action described by the sentence. The solution to the problem of prepositional polysemy adopted here is to consider the various senses of the preposition under consideration as different constructions, partially inheriting their meaning from a common construction (a subcase of the general Preposition construction) and partially from distinct constructions (subcases of the Preposition construction as well). This solution allows us to maintain the distinction between the two senses without neglecting the connection holding between each other. Moreover, this criterion does not force us to contradict the view of prepositions as radial categories (see e.g. Lakoff 1987, Lewandowska-Tomaszczyk 2007), with one sense constituting the prototypical center of a category, and the other(s) representing extensions 24. As a matter of fact, it 23 This argument is supported by experimental evidence (see. e.g. Klein and Murphy 2001). 24 However, here we are not interested in finding the prototypical center of each preposition. On this topic, the reader is referred to Tyler and Evans's (2003) proposal for an objective distinction of the different senses of the prepositions stored in semantic memory and an objective selection of a central sense. Moreover, the account of the Italian preposition da offered by Luraghi (2009) on the basis of 136 3. An ECG analysis of Italian caused-motion constructions should be noted that the radial category structure is not incompatible with a categorization based on the network model. On the contrary, as pointed out by Tuggy (2007: 90), structuring prepositions as prototypically or schematically organized can be considered equivalent strategies. A similar solution will then be adopted with regard to nominal, adjectival, and verbal polysemy, even though there is a notable difference: nouns, adjectives, and verbs do not inherit their meaning from distinct constructions. While the meaning of prepositions can be seen as deriving from the combination of two higher constructions (e.g. In and PathPreposition), nouns, adjectives, and verbs are just subcases of the common construction and the meanings of nominal, adjectival, and verbal constructions is not inherited but intrinsic to them. From this point of view, the status of prepositions is a bit different from that of nouns, adjectives, and verbs. This is probably related to the fact that prepositions represent a small, closed class of function words, each of which can take several meanings, but it is unlikely that they develop new senses (unless we consider a very long timespan). As a result, it is relatively easy (and also convenient) to classify them into stable, precise categories (spatial prepositions, causative prepositions, etc.)25. On the other hand, nouns, adjectives, and verbs are large, open classes of content words and their acquisition of new meanings is rather probable, if not even likely. As a result, their several senses cannot be seen as combinations of the meaning inherited from the common construction with distinct constructions. Finally, in the (rare) case of monosemous items, a single construction can be posited26. We have also to consider that polysemy does not concern us at the lexical level only, but at the constructional level as well. As already specified, the status of caused-motion constructions in ECG is awarded only to those sentences which express an actual caused-motion action accomplished via force-exertion (i.e. those active and passive constructions whose meaning pole is bound to the CauseMotionAction schema) which correspond to Goldberg's (1995) prototypical caused-motion constructions. As a result, Goldberg's radial extensions to caused-motion constructions are not included in the category. Among these extensions, for instance, we can find sentences expressing prevented motion, i.e. the expression conveys the “imposition of a barrier, causing the patient to stay in a location despite its tendency to move” (Goldberg 1995: 162). Consider the following example: (20) Harry locked Joe into the bathroom. In ECG, sentence-level constructions like these are not considered, strictly speaking, caused-motion constructions. However, this does not mean that Goldberg's prototypical constructions and their extensions are taken to be unrelated, but simply that the generalizations between these types of constructions are to be captured at a higher, more schematic level. This is not an easy task to carry out and its accomplishment is still under way, indeed. Nevertheless, as will be shown in §3.3, some little steps in this direction are being taken (see Dodge and Bryant forthcoming). diachronic evidence also provides relevant insights. Since during the last decades the semantic status of prepositions has been a subject of interest and debate among cognitive linguists, the reader is also referred to the constributions collected in Zelinsky-Wibbelt (1993). 25 This situation seems plausible from the perspective of the language user, not merely that of the analyst. 26 Not having the possibility to rely on empirical data, with regard to the distinction between monosemous and polysemous words, we shall make reference to the online version of the Dizionario Garzanti della Lingua Italiana (http://www.garzantilinguistica.it/it/dizionario/it). 137 Grounding Meaning in Everyday Experience in the World Polysemy, together with compositionality, represents an issue which, with regard to language processing, can realistically be dealt with by the adoption of a simultaneously top-down and bottom-up approach. Keeping in mind that we are adopting the network model briefly outlined above, we can posit that the incremental processing of a sentence allows the understander to have a direct rather than sequential access to a particular sense of a lexical unit without invoking its more abstract representation. This is, naturally, especially relevant with regard to the units coming later in the sentence, since the units previously introduced influence the processing of the subsequent. At the same time, new information will then influence the interpretation of the previous units. In this way, incremental processing drives the understander to an interpretation of the sentence, which will be refined according to the constraints posited by the sentencelevel construction itself. Nevertheless, advancing in processing can also reveal that the first selected sense is not the correct one, so the understander will have to search for another sense, requiring that it be more appropriate in the context of the sentence. For instance, in the sentence reported in (21) below, the verb shoot will probably be first understood with the meaning of “throw the ball at the goal”, while after mentioning the NP the robber it will be corrected to “make a weapon fire” and the entire sentence will be understood against a different frame (from a “soccer match” frame to a “soccer player being robbed” frame), but there will be a price to pay in terms of processing time27. (21) The soccer player shot the robber. While compositionality and polysemy are the two most prominent issues that will be more extensively examined and discussed in the remainder of the present chapter, they are not the only. In the following section, we shall be illustrating what emerges from an ECG analysis of our eight Italian caused-motion constructions. Then, in §3.3, we shall discuss the theoretical and operational issues raised by our analysis. Along with the relationship between compositionality and polysemy, we shall be spending some time on the discussion of the ontological nature of a couple of verbs involved in our sentence-level constructions, the status of Italian possessives, the treatment of scalar modifiers in ECG, and the problem of Italian past participle forms. The discussion of these topics will often be cross-cut by the reference to two parameters: the differences between Italian and English and the need for an ECG analysis to flow in both top-down and bottom-up manner. Finally, in §3.4 we shall provide the reader with a brief summary of the chapter, followed by an evaluation of the ECG model, a brief discussion of its weak points, and some suggestions for future work. 3.2. Analysis In this section, which is divided into three subsections, we are going to provide the reader with a detailed analysis of our selected caused-motion constructions. In the first section, we shall be making use of the ECG formalism to analyze four active sentencelevel constructions, which show a simpler syntactic structure, focusing on the lexical, 27 This problem of misinterpretations will not be directly treated in the present study, but a brief point on this topic will be made in §3.3.2. 138 3. An ECG analysis of Italian caused-motion constructions phrasal, and clausal constructions they instantiate. The second section will be dedicated to the application of the formalism to four passive sentences, whose structure is a bit more complex. The third subsection, finally, is meant to be a sort of “appendix” to the previous subsections, where we shall briefly introduce a couple of less simple (though frequently observed) grammatical phenomena, proposing the the formalism used in §3.2.1 and §3.2.2 can be efficiently extended to deal with these syntactic constructions. Before starting, it is important for us to specify that since some of the very important schemas and constructions which will be exploited during the analysis of these sentences were already introduced in §2.3.2.3, when we analyzed the transitive construction Sara tagliò la pagnotta (e.g. Process and ComplexProcess, shown in Fig. 2.1 and Fig. 2.2), we are not going to show them again, unless their repetition is particularly convenient for the reader to understand the analysis. 3.2.1. Active caused-motion constructions The first construction to be illustrated is that reported in (22) below: (22) I bambini lanceranno i sassi al bersaglio. The.MPL child.PL throw:FUT.3PL the.MPL stone.PL to-the.MSG target.SG “The children will throw the stones to the target” We begin our analysis illustrating the lexical construction corresponding to the verb used in the sentence in (22), lanceranno, shown in Fig. 3.1 below. Since lanceranno is the third-person plural form of the verb lanciare in the simple future tense, this construction is labeled Lanciare1SimpleFuture3Pl. Lanciare1, shown in Fig. 3.2, is the construction referring to the verb lanciare when it takes the specific meaning implied by the sentence, and it is a subcase of the more schematic construction Lanciare (Fig 3.3 below). Fig. 3.1: the Lanciare1SimpleFuture3Pl construction. construction Lanciare1SimpleFuture3Pl subcase of Lanciare1, SimpleFutureTense, Plural, 3rdperson constructional: VerbFeatureSet // inh. constraints self.features.verbform ← simplefuture // inh. self.features.number ← plural // inh. self.features.person ← 3 // inh. form: WordForm // inh. constraints self.f.phon ← /lanʧe'ran:o/ meaning: CauseMotionAction // inh. constraints self.m.x-net ← @throw // inh. 139 Grounding Meaning in Everyday Experience in the World Fig. 3.2: the Lanciare1 construction construction Lanciare1 subcase of Lanciare constructional: VerbFeatureSet // inh. constraints self.features.verbform ← base // inh. form: WordForm // inh. constraints self.f.phon ← /lan'ʧare/ // inh. meaning: CauseMotionAction constraints self.f.x-net ← @throw Fig. 3.3: the Lanciare construction construction Lanciare subcase of LexVerb constructional: VerbFeatureSet // inh. constraints self.features.verbform ← base form: WordForm // inh. constraints self.f.phon ← /lan'ʧare/ meaning: Process // inh. The Lanciare1SimpleFuture3Pl construction is a subcase of the Lanciare1, the SimpleFuture, the Plural, and the 3rdperson constructions. Its form pole is constrained to be a word, being bound to the WordForm schema (see Fig. 2.10), with its phon role providing the specific phonological structure of the construction. The meaning pole of LanciareSimpleFuture3Pl is constrained to be CauseMotionAction, a schema shown in Fig. 3.4 below. CauseMotionAction is a subcase of the CauseEffectAction schema used in the analysis of the transitive construction carried out in §2.3.2.3 (see Fig. 2.11). In addition, the x-net role of Lanciare1SimpleFuture3Pl is assigned the proper x-net value (i.e. @throw). Being a subcase of CauseEffectAction (and, consequently, of ComplexProcess), CauseMotionAction is composed by two different processes: the first one is constrained to be a process of ForceApplication (see Fig. 2.12), since we have a causer participant performing a forceful action on an affected participant, while the second process has a MotionAlongAPath type constraint, for the affected participant becomes the protagonist of a motion event along a path. The MotionAlongAPath schemas is shown in Fig. 3.5 below, where we can also observe the addition of an x-net role, constrained to be a motionalongapath action. Fig. 3.4: the CauseMotionAction schema schema CauseMotionAction subcase of CauseEffectAction roles process1: ForceApplication // inh. process2: MotionAlongAPath causer // inh. affected // inh. x-net: @causemotionaction constraints protagonist ↔ causer // inh. protagonist2 ↔ affected // inh. process1.actedUpon ↔ affected // inh. Fig. 3.5: the MotionAlongAPath schema schema MotionAlongAPath subcase of Motion evokes SPG as spg roles x-net: @motionalongapath constraints mover ↔ spg.trajector 140 3. An ECG analysis of Italian caused-motion constructions MotionAlongAPath is a subcase of the Motion schema and it evokes the SPG (SourcePathGoal) schema, shown in Fig. 3.6 and Fig. 3.7 below, respectively. The MotionAlongAPath schema inherits the roles of the Motion schema, with the mover role being particularly important, since it is bound to be the trajector of the SPG schema, a subcase of the TL schema, shown in Fig. 3.8. The TL schema is a subcase of the Relation schema (not shown). The Motion schema binds the mover role to the protagonist of the second process of CauseMotionAction, i.e. MotionAlongAPath. Fig. 3.6: the Motion schema schema Motion subcase of Process roles mover speed heading constraints mover ↔ protagonist Fig. 3.7: the TL schema schema TL subcase of Relation roles trajector landmark profiledArea Fig. 3.8: the SPG schema schema SPG subcase of TL roles source path goal The evocation of the SPG schema is crucial, since it specifies that the mover of MotionAlongAPath is a trajector which moves from a source along a path to a goal. The SPG schema, introduced by Johnson (1987), structures our comprehension of directed motion. It is a subcase of Langacker's (1987) TL (TrajectorLandmark) schema, which portrays a spatial relationship involving a trajector, whose orientation, location, or motion is defined with reference to a landmark28. The third role of the TL schema, profiledArea, serves to define the attentionally-profiled region of space. The importance of the SPG schema will become clearer later, at present the reader can just bear in mind that the mover of MotionAlongAPath is the trajector of SPG. We can move on to consider the ActiveCauseMotion1 construction (Fig. 3.9 below), the A-S construction involved in the sentence currently under consideration, which is a subcase of the ActiveCauseMotion construction (Fig. 3.10), which in turn is a subcase of VerbPlusArguments (Fig. 2.16). Since it is a phrasal construction, ActiveCauseMotion presents a constructional block, where its constituents are defined. It defines an np constituent, which represents its object, and a pp constituent, which represents the movement of such an object along a path. The meaning of ActiveCauseMotion is identified with the CauseMotionAction schema. As an A-S construction, ActiveCauseMotion inherits from VerbPlusArguments an evoked EventDescriptor schema (Fig. 2.15) in order to clarify how the scene described by the sentence should be simulated. Also, ActiveCauseMotion inherits the fact that its meaning as a whole is bound to the eventType role of the EventDescriptor schema, while the meaning of its verb is bound to the profiledProcess of the same schema. This meaning is then elaborated by the construction which, in addition to these inherited meaning constraints, constrains its causer participant to be co-indexed with the profiledParticipant role of EventDescriptor, and its affected participant to be coindexed with the meaning of the np constituent. Then, In ActiveCauseMotion1, 28 Bergen and Chang (2005) offer a very brief but clear characterization of TL and SPG. 141 Grounding Meaning in Everyday Experience in the World ProfiledParticipant is also identified with the Topic role of the EventDescriptor schema. Moreover, the meaning of the construction itself is co-indexed with the meaning of the verb, while its second process is co-indexed with the SPG schema, which is the meaning of the pp constituent (as will be shown below). ActiveCauseMotion1 has also a form block, which adds two form constraints: the verb's form (v.f) is constrained to come before the NP's form (np.f), which in turn is contrained to come before the PP's form (pp.f). Fig. 3.9: the ActiveCauseMotion1 construction construction ActiveCauseMotion1 subcase of ActiveCauseMotion constructional constituents v: Verb // inh. np: NP // inh. pp: PathPP // inh. form constraints v.f before np.f np.f before pp.f meaning: CauseMotionAction // inh. evokes EventDescriptor as ed // inh. constraints self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m ↔ v.m self.m.affected ↔ np.m // inh. self.m.causer ↔ ed.profiledParticipant // inh. ed.profiledParticipant ↔ ed.topic self.m.process2 ↔ pp.m.spg Fig. 3.10: the ActiveCauseMotion construction construction ActiveCauseMotion subcase of VerbPlusArguments constructional constituents v: Verb // inh. np: NP . pp: PathPP meaning: CauseMotionAction evokes EventDescriptor as ed // inh. constraints self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m.affected ↔ np.m self.m.causer ↔ ed.profiledParticipant As regards clause-level constructions, we can exploit Declarative (Fig. 2.18), which identifies its meaning with the EventDescriptor schema, asserting that this type of construction portrays an event. The Declarative construction and its relationship with EventDescriptor were illustrated and described in §2.3.2.3. Before directing our efforts at the description of the other constructions instantiated by the sentence in (22), we can now provide the reader with a brief summary of what observed so far. In order to specify the nature of the event described (a two-participant event in which the agent exerts force on the patient, causing their movement from a source along a path to a goal), the A-S construction ActiveCauseMotion1 inherites from ActiveCauseMotion the identification of its meaning with the CauseMotionAction schema. The meaning of ActiveCauseMotion1 is bound to the eventType of an evoked EventDescriptor schema. Therefore, the eventType role of 142 3. An ECG analysis of Italian caused-motion constructions EventDescriptor and the CauseMotionAction schema share the same causal structure. The profiledParticipant of ActiveCauseMotion1 is specified to be the causer role of CauseMotionAction. ActiveCauseMotion1 inherits a verb constituent and the constraint which binds its meaning to the profiledProcess of the EventDescriptor schema. Moreover, ActiveCauseMotion1 binds the meaning of the verb constituent to that of the A-S construction itself. The Lanciare1SimpleFuture3Pl construction also identifies its meaning with the CauseMotionAction schema; therefore, it meets the constraints specified for the verb constituent of ActiveCauseMotion1. Moreover, Lanciare1SimpleFuture3Pl specifies that the caused-motion action involves a particular kind of action (a throwing x-net). Consequently, the verb construction and the A-S construction share the same general schematic meaning, with the verb construction also providing a more specific meaning, as in the case of our example sentence analyzed in §2.3.2.3. ActiveCauseMotion1 has also inherited an np constituent and a pp constituent. The form of the np constituent is constrained to follow the form of the verb. Meaning constraints precise that the meaning of np is bound to the affected role of the CauseMotionAction schema. As a result, this construction specifies that the entity expressed by the “direct object” np is acted upon and affected by the action of the causer. The pp constituent is constrained to be filled by a PP expressing path, and its form is constrained to follow that of the np constituent. Meaning constraints specify that the second process of CauseMotionAction (i.e. MotionAlongAPath) is bound to the SPG schema which is the meaning of pp. Declarative's meaning is identified with an EventDescriptor schema. It shows an inherited subj constituent, whose meaning is bound to the profiledParticipant role of EventDescriptor. Moreover, it has a second constituent, fin, which unifies with ActiveCauseMotion1. In addition, Declarative specifies that the EventDescriptor evoked by the A-S construction is to be identified with that of Declarative, highlighting the fact that both constructions describe the same event. Furthermore, the profiledParticipant roles referred to by each of these constructions will be identified with each other. As a result, the meaning of the referent of Declarative's subj constituent will be identified with the causer of CauseMotionAction. As in the case of our example sentence analyzed in §2.3.2.3, the constructions (and the relative schemas) illustrated above provide fairly general constraints on the fillers of various participant roles. More specific information on the fillers of these roles for a particular event will be supplied when the NP and PP constructions instantiated in the utterance unify with the other instantiated constructions. In the sentence presently under consideration, there are two NP constructions (one for the subject i bambini and one for the object i sassi) and one PP construction (for the spatial phrase al bersaglio). The expression bambini in Italian can instantiate two different constructions: one which refers to two or more children of different or unknown sex, the other to two or more male children. It is not possible to know which of the two constructions is actually instantiated without information on the discourse and situational context. In our analysis, we shall consider the first hypothesis, but this is just an arbitrary choice29. The Bambini1 construction, which inherits all the features of a more general Bambini 29 Note that there is only one difference between the two constructions: in the first case (the one represented in Fig. 3.11 below), the meaning block of the construction does not offer any information on the referent's natural gender; in the second case (not shown) the natgender role is filled with the value male. 143 Grounding Meaning in Everyday Experience in the World construction (not shown)30 adding the crucial information related to the value of the category of the referent described, is represented in Fig. 3.11 below, while in Fig. 3.12 you can find the illustration of the the I construction (a subcase of the MalePlDefiniteArticle construction, not shown), respectively. Fig. 3.11: the Bambini1 construction construction Bambini1 subcase of Bambini constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /bam'bini/ // inh. meaning: RD // inh. constraints self.m.category ← Child self.m.number ← morethanone // inh. Fig. 3.12: the I construction construction I subcase of MalePlDefiniteArticle constructional: NominalFeatureSet // inherited constraints self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /i/ meaning evokes RD as rd // inh. constraints rd.number ← morethanone // inh. rd.accessibility ← uniquely-identifiable // inh. The two constructions illustrated above will be combined by the DeterminerPlusNP construction (Fig. 2.30), which establishes a formal constraint on the order of the constituents and a semantic constraint which binds the meaning of the complex construction to the meaning of both constituents. The DeterminerPlusNP construction will unify with Declarative's subj constituent, supplying the following information: the entity that fills the causer and profiledParticipant roles is morethanone and it is represented by members of the category Child. The level of accessibility of i bambini is uniquely-identifiable, since the use of a definite determiner denotes the possibility to uniquely identify the referent. ActiveCauseMotion1's np constituent will unify with the NP whose noun constituent is Sassi, providing more specific information on the filler of the affected role. The Sassi construction is shown in Fig. 3.13 below. Sassi and I will be combined together by the DeterminerPlusNP construction. 30 The same is true for all the constructions relative to ambiguous or polysemous nouns, though it will not be repeated everytime. 144 3. An ECG analysis of Italian caused-motion constructions Fig. 3.13: the Sassi construction construction Sassi1 subcase of Sassi constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /'sas:i/ // inh. meaning: RD // inh. constraints self.m.category ← Stone self.m.natgender ← neuter // inh. self.m.number ← morethanone // inh. Both the Bambini and the Sassi constructions instantiate the MalePluralNoun construction, shown in Fig. 3.14 below. Fig. 3.14: the MalePluralNoun construction construction MalePluralNoun subcase of PluralNoun, Male constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. meaning: RD // inh. constraints self.m.number ← morethanone // inh. Now, we come to the PP al bersaglio, which is perhaps the most interesting component of the sentence-level construction under consideration. First of all, a representation of the Bersaglio1 construction is provided in Fig. 3.15 below. 145 Grounding Meaning in Everyday Experience in the World Fig. 3.15: the Bersaglio1 construction construction Bersaglio1 subcase of Bersaglio constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /ber'saʎo/ // inh. meaning: RD // inh. constraints self.m.category ← Target self.m.number ← one // inh. Al is a complex preposition, composed of the simple preposition a and the determiner il. In this kind of sentence-level construction, a has a dynamic meaning represented in a construction labeled A2 (illustrated in Fig. 3.20 below), subcase of the PathPreposition construction (Fig. 3.18) and of a generic A construction (Fig. 3.19). PathPreposition is a subcase of the SpatialPreposition construction (Fig. 3.17). Both SpatialPreposition and A are subcases of the Preposition construction (Fig. 3.16). The meaning of the PathPreposition construction is constrained to be SPG (already shown in Fig. 3.7 above), The meaning of the A construction, instead, is more generically constrained to denote a Relation. The A2 construction inherits from both PathPreposition and A, but its meaning also evokes a Proximity schema (Fig. 3.21). Proximity has two roles which help to describe the relation expressed by the preposition: center, which is connected to the landmark role of SPG, and proximalArea, bound to the profiledArea role of SPG. Also, A2 adds the meaning constraint that the goal role of the SPG schema has to be bound to its landmark role (while it was already specified above that the trajector of this motion is the mover role of MotionAlongAPath. i.e. the affected role of CauseMotionAction). Identifying the goal of the motion with its landmark we define the type of event being described: a motion toward a certain entity (cfr. Zlatev's 2007b point on cognitive approaches to path and direction). Fig. 3.16: the Preposition construction construction Preposition subcase of Word form: WordForm // inh. meaning: Relation Fig. 3.17: the SpatialPreposition construction construction SpatialPreposition subcase of Preposition form: WordForm // inh. meaning: TL 146 3. An ECG analysis of Italian caused-motion constructions Fig. 3.18: the PathPreposition construction construction PathPreposition subcase of SpatialPreposition form: WordForm // inh. meaning: SPG Fig. 3.19 the A construction construction A subcase of Preposition form: WordForm // inh. constraints self.f.phon ← /a/ meaning: Relation // inh. Fig. 3.20: the A2 construction construction A2 subcase of PathPreposition, A form: WordForm // inh. constraints self.f.phon ← /a/ // inh. meaning: SPG // inh. evokes Proximity as p constraints self.m.landmark ↔ self.m.goal self.m.landmark ↔ p.center self.m.profiledArea ↔ p.proximalArea Fig. 3.21: the Proximity schema schema Proximity subcase of SpatialRelation roles center proximalArea The construction corresponding to the definite article Il is shown in Fig. 3.22 below. Fig. 3.22: the Il construction construction Il subcase of MaleSgDefiniteArticle constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /il/ meaning evokes RD as rd // inh. constraints rd.number ← one // inh. rd.accessibility ← uniquely-identifiable // inh. 147 Grounding Meaning in Everyday Experience in the World A2 and Il are then combined by the ComplexPathPreposition construction, shown in Fig. 3.23 below. Fig. 3.23: the ComplexPathPreposition construction construction ComplexPathPreposition subcase of PathPreposition constructional constituents prep: PathPreposition s: Determiner form: WordForm // inh. constraints prep.f fuses with s.f meaning: SPG // inh. constraints self.m ↔ prep.m self.m ↔ det.m The ComplexPathPreposition construction specifies that the form of A2 fuses with the form of Il, while the meaning of the construction sums that of the simple preposition with that of the determiner. ComplexPathPreposition is sufficient to illustrate the combination of the preposition a and the determiner il but, since the form of the complex preposition al is not a simple agglutination of a + il (which would presumibly be *ail), we shall represent the construction Al2 explicitly in Fig. 3.24 below (complex prepositions found in the next sentences will not be shown). Fig. 3.24: the Al2 construction construction Al2 subcase of ComplexPathPreposition constructional constituents prep: A2 s: Il form: WordForm // inh. constraints self.f.phon ← /al/ meaning: SPG // ihn. constraints self.m ↔ prep.m // inh. self.m ↔ s.m // inh. 148 3. An ECG analysis of Italian caused-motion constructions The ComplexPathPreposition construction and the Bersaglio construction are finally combined together in the PathPP construction, shown in Fig. 3.25 below. Fig. 3.25: the PathPP construction construction PathPP subcase of SpatialPP constructional constituents prep: PathPreposition np: NP // inh. form constraints prep.f before np.f // inh. Meaning: SPG constraints self.m ↔ prep.m // inh. self.m.landmark ↔ np.m // inh. PathPP has two constituents: prep, constrained to be a PathPreposition, and np, more loosely constrained to be an NP. The form of np is constrained to follow that of prep. Moreover, meaning constraints specify that the meaning of PathPP is bound to the meaning of prep, and that its landmark role is bound to np. The unification of the constructions instantiated in our example produce a SemSpec, consisting in schemas, constraints, and bindings, which supports an enactment of the event described by the sentence. The EventDescriptor schema provides several crucial simulation parameters. Its profiledProcess role is bound to the meaning of Lanciare1SimpleFuture3Pl. It is a process of CauseMotionAction and has an x-net of throwing. The eventType role is bound to the meaning of ActiveCauseMotion1, in which the meaning of the argument structure is bound to that of its verb constituent. Therefore, eventType is also bound to CauseEffectAction. As in the case of our prototypical transitive sentence exemplified in §2.3.2.3, the A-S construction provides information about the general type of event being described, and the verb adds more information about the specific processes involved in such an event, including information on their structure and setting time31. Moreover, the unification of ActiveCauseMotion1, Lanciare1SimpleFuture3Pl, Declarative, and the instantiated NP and PP constructions results in various bindings associated with each of the participants of this event. The causer role of CauseMotionAction is bound to: – the protagonist of CauseMotionAction – the actor and protagonist of ForceApplication; – the profiledParticipant of EventDescriptor; 31 In Italian (as in other Romance languages), the simple future is normally used to describe actions and events which still have to happen. 149 Grounding Meaning in Everyday Experience in the World – the subj constituent of Declarative; – the referent of i bambini. The affected role of CauseMotionAction is bound to: – the protagonist2 of CauseMotionAction; – the actedUpon of ForceApplication; – the mover and protagonist of MotionAlongAPath; – the np of ActiveCauseMotion1; – the trajector of SPG; – the entity referred to as i sassi. Furthermore, PathPP specifies the direction of the motion undergone by the trajector, i.e. toward the entity referred to as il bersaglio. This SemSpec, capturing the conceptual core of the sentence-level construction under consideration, will drive the enactment of an event in which two or more uniquely identifiable children perform a forceful throwing action on two or more uniquely identifiable stones, obtaining their movement toward a uniquely-identifiable target. A contribution to the enactment process may be offered by some components of the sentence which are endowed with a high semantic/pragmatic strength and are most likely to activate a certain frame: as an example, the noun i bambini and the phrase al bersaglio suggests that the children are playing some kind of game. Since the event is described from the perspective of the agent, it should be enacted from the same perspective. The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agents, the purpose on which they performe the action, and the effector they exploit (at the very least). Then, on the basis of their world knowledge and the (discourse and situational) context, the language understander will arrive at a deeper understanding of the utterance, also drawing further inferences32. For instance, the sentence as a whole invites us to think that the agent participants are supposed to have a good time, since we know that children generally like playing games. Most of the schemas and constructions exploited to carry out our analysis of the sentence in (22) may also be used to analyze a number of active caused-motion constructions in Italian (but probably this situations holds for other languages as well). Indeed, following Dodge and Bryant's (forthcoming) proposal for transitive constructions illustrated in §2.3, we suggest that ActiveCauseMotion1, the A-S construction which played a crucial role in the analysis of the sentence-level construction just examined above, can be identified with a prototypical caused-motion event conceptualized and described from the perspective of the agent participant. As such, it is posited to unify with a group of verbs which share some semantic basic characteristics. Consider the sentence in (23) below: (23) Posasti il portacenere sul comodino. Lay:PRT.2SG the.MSG ashtray on-the.MSG bedside-table.SG “You layed the ashtray on the bedside table” This sentence is analogous to the previous one. The only points of divergence concern 32 Some scholars who work on ECG (see for instance Bailey 1997; Bergen and Chang 2005; Bergen 2005) also provide a dynamic representation of the enactment process using the x-schema formalism. For a brief overview of this formalism, see e.g. Narayanan (1999). 150 3. An ECG analysis of Italian caused-motion constructions the lexical verb construction, the NPs and the PP involved, but the basic meaning of the verb and the A-S construction involved remains the same. Indeed, this sentence can be analyzed exploiting the same schemas and constructions used to investigate the previous one. Therefore, we shall provide the reader with just the remaining constructions: the lexical verb construction, the constructions related to the NPs and those regarding the PP. Since posasti is is the second-person singular form of the verb posare in the simple past tense, the corresponding lexical construction (shown in Fig. 3.26) is labeled Posare1SimplePast2sg. Fig. 3.26: the Posare1SimplePast2Sg construction construction Posare1SimplePast2Sg subcase of Posare1, SimplePastTense, Singular, 2person constructional: VerbFeatureSet // inh. constraints self.features.verbform ← simplepast // inh. self.features.number ← singular // inh. self.features.person ← 2 // inh. form: WordForm // inh. constraints self.f.phon ← /po'zasti/ meaning: CauseMotionAction // inh. self.m.x-net ← @lay // inh. Posare1SimplePast2Sg is a subcase of Posare1, SimplePastTense, Singular, and 2ndperson. Its form is obviously constrained to be a word and is assigned the correct phonological structure. Its meaning is constrained to be CauseMotionAction, and its xnet role is filled with the lay value. Given the structure of the sentence, which is analogous to that analyzed previously, we can exploit again the ActiveCauseMotion1 construction at the phrasal level and the Declarative construction at the clausal level. Therefore, having established that the general constraints on the fillers of the participant roles are the same already observed in the analysis of the previous sentence, we can directly turn to the NPs and PPs, which provide more specific information on the fillers of these roles for the particular event described by the sentence in (23). We can observe a null subject NP, the object NP il portacenere and the spatial phrase sul comodino. Since the verb is conjugated in the second singular person, the null subject construction, illustrated in Fig. 3.27, is labeled DropTu, and it is a subcase of the Tu construction, shown in Fig. 3.28. The Tu construction, in turn, is a subcase of the SingularSubjPersPron construction, reported in Fig. 3.29 below and the 2ndperson construction (not shown). It is important to underline that the Tu construction's case is constrained to be nominative, since the Italian language maintains two distinct forms for subject and object personal pronouns, with the only exceptions being the first and second-person plural person. It is also worth specifying that the use of the ignore command in the form block of DropTu allows us 151 Grounding Meaning in Everyday Experience in the World to clarify that the phonological form of the pronoun is suppressed. DropTu will unify with the referent of Declarative's subj constituent, providing the information that the entity that fills the causer and profiledParticipant role is one animate referent represented by the current addressee. The level of accessibility of a first-person plural pronoun is active, since pronouns are generally used to denote active referents. This is just a rule of thumb, which can be cancelled by pragmatic devices (e.g. gestures), but in this case it is not possible, for its referent is one of the current interlocutors. The ontological category of the referent is assigned the value “animate”, since the pronouns just lets us know that the addressee is animate, while it does not specify if it is a human being or an animal33. Fig. 3.27: the DropTu construction construction DropTu subcase of Tu constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh. self.features.number ← singular // inh. self.features.person ← 2 // inh. form: WordForm // inh. constraints ignore: self.f.phon ← /tu/ meaning: RD // inh. self.m.category ← Animate // inh. self.m.number ← one // inh. self.m.accessibility ← active // inh. self.m.referent ← currentaddressee // inh. Fig. 3.28: the Tu construction construction Tu subcase of SingularSubjPersPron, 2ndperson constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh self.features.number ← singular // inh. self.features.person ← 2 // inh form: WordForm // inh. constraints self.f.phon ← /tu/ meaning: RD // inh. self.m.category ← Animate self.m.number ← one // inh. self.m.accessibility ← active // inh. self.m.referent ← currentaddressee 33 The situation will be disambiguated when all the constructions instantiated in the sentence unify. 152 3. An ECG analysis of Italian caused-motion constructions Fig. 3.29: the SingularSubjPersPron construction construction SingularSubjPersPron subcase of SubjPersPron, Singular constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh. self.features.number ← singular // inh. form: WordForm // inh. meaning: RD // inh. Constraints self.m.accessibility ← active // inh. self.m.number ← one ActiveCauseMotion1's np constituent will unify with the NP whose noun constituent is portacenere, providing more specific information on the filler of the affected role. Here, we can posit the existence of just one construction, since portacenere is a monosemous word. Nevertheless, since the singular and the plural forms of the noun portacenere coincide, here we have to make a decision: we have to consider portacenere as instantiating two different constructions (one for the singular form and one for the plural), but we can distinguish them at two different levels. On the one hand, we can posit a comprehensive Portacenere construction (not shown), which does not provide any information about number, as a subcase of a MaleNoun construction (subcase of Male and Noun) and then a PortacenereSg construction which is a subcase of Portacenere and Singular. On the other hand, we may also choose to posit the PortacenereSg construction as a subcase of the MaleSingularNoun construction, subcase of SingularNoun (in turn, a subcase of Noun and Singular) and Male. We shall select the former option, for establishing the distinction at a lower level captures the fact that the number of portacenere can only be known by looking at the rest of the NP (in this case, the determiner), leaving room for an ambiguous interpretation out of context (even though the singular reading would probably be the default option). The PortacenereSg construction is shown in Fig. 3.30 below. Fig. 3.30: the PortacenereSg construction construction PortacenereSg subcase of Portacenere, Singular constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /porta'ʧenere/ // inh. meaning: RD // inh. 153 self.m.category ← Ashtray // inh. self.m.natgender ← neuter // inh. self.m.number ← one Grounding Meaning in Everyday Experience in the World Now, we come to the PP sul comodino. This phrase is particularly interesting and its prepositional constituent deserves to be awarded special attention. Indeed, while Italian speakers can easily explain the different conceptualizations which underlie the path prepositions su and in, the description of this difference in formal terms may seem less straightforward. A convenient solution to this problem is offered by the adoption of the BoundedObject schema (Fig. 3.31 below), subcase of the BoundedRegion schema (Fig. 3.32), illustrated in Gilardi's Starter2 grammar. Fig. 3.31: the BoundedObject schema schema BoundedObject subcase of BoundedRegion roles boundedobject: entity boundary: closedcurve // inh. interior: region // inh. exterior: region // inh. Fig. 3.32: the BoundedRegion schema schema BoundedRegion roles boundary: closecurve interior: region exterior: region The BoundedObject schema is evoked in the meaning block of the Su2 construction (Fig. 3.33 below), a subcase of the Su construction (not shown). The Comodino construction is shown in Fig. 3.34 below (comodino seems to have a single meaning). Fig. 3.33: the Su2 construction construction Su2 subcase of PathPreposition, Su form: WordForm // inh. constraints self.f.phon ← /su/ // inh. meaning: SPG // inh. evokes BoundedObject as bo constraints self.m.landmark ↔ bo.boundedobject self.m.profiledArea ↔ bo.boundary self.m.goal ↔ self.m.profiledArea self.m.source ↔ bo.exterior Fig. 3.34: the Comodino construction construction Comodino subcase of MaleSingularNoun construction: NominalFeatureSet constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /komo'dino/ meaning: RD // inh. constraints self.m.category ← bedside-table self.m.natgender ← neuter self.m.number ← one // inherited Binding the profieldArea of the SPG schema to both its goal role and to the boundary role of the BoundedObject schema allows us to capture (and properly formalize) the specificity of the type of motion event described by the preposition su in a dynamic construction: the Su2 construction denotes the motion of the trajector role of the SPG 154 3. An ECG analysis of Italian caused-motion constructions schema from an outer area to the uppermost layer of an object34. Su2 and Il are then combined by the ComplexPathPreposition construction. The ComplexPathPreposition construction specifies that the form of Su2 fuses with the form of Il, while the meaning of the construction is obtained summing that of the simple preposition with that of the determiner. Again, the combination of the phonological form of the complex preposition, is not the simple fusion of su + il (i.e. */suil/), but /sul/. The ComplexPathPreposition construction is finally combined with the Comodino construction in the PathPP construction. Again, PathPP has two constituents: prep, constrained to be a PathPreposition, and np, more loosely constrained to be an NP. The form of np is constrained to follow that of prep. Moreover, meaning constraints specify that the SPG meaning of PathPP is bound to the meaning of prep, and that its landmark role is bound to the referent of np. The landmark role of SPG unifies with the entity represented by the noun Comodino, and the pp constituent of ActiveCauseMotion1 unifies with the PathPP construction, specifying the path followed by the mover of MotionAlongAPath (i.e. the affected of CauseMotionAction): from the exterior to the surface of the object defined by the noun Comodino. The SemSpec produced by the unification of the constructions instantiated by this sentence will be similar to the one of the sentence analyzed above. The main difference will be represented by the profiledProcess role of the EventDescriptor schema, filled by the meaning of Posare1SimplePast2Sg, which maintains a basic meaning of CauseMotionAction, but its x-net is one of laying. Also, the tense of the verb specifies that the event described in this sentence represents an action that occurred during a period of time and has no continuing effect on reference to the present (see Moneti and Lazzarino 2007: 140-142). Again, this SemSpec will capture the conceptual core of our sentence, driving the enactment of an event in which the current addressee performs a laying action on a uniquely identifiable ashtray, obtaining its movement on a uniquely identifiable bedside table. Once more, the event will be enacted from the perspective of the agent. The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agent (probably very low in this case), the purpose on which he performed the action etc. The presence of words such as portacenere and comodino suggests that the action took place in a bedroom (and perhaps somebody had just smoked a cigarette). Then, on the basis of their encyclopedic knowledge and the (discourse and situational) context, the language understander will be able to mentally reconstruct the event in more detail. In this particular case, the degree of identification of the language-understander will probably be very strong, since he has to simulate an action performed by themself in the past. The two sentences considered so far represent fairly simple instances of causedmotion constructions. In the next lines, we will be analyzing a similar clause, which share the same A-S construction and the same core meaning with those previously illustrated, but it presents a bit more complex grammatical structure, showing the presence of an auxiliary verb and an adjective. Consider the sentence-level construction in (24): (24) Il giovane netturbino ha sollevato un bidone dal The.MSG young.SG dustman.SG have:PRES.3SG lift:PSTPART a.MSG garbage.can.SG from-the.MSG marciapiede. sidewalk.SG. 34 Cfr. the In2 construction in Fig. 3.57 below. 155 Grounding Meaning in Everyday Experience in the World “The young dustman has lifted a garbage can from the sidewalk” This sentence is in the present perfect tense; therefore, instead of a simple finite verb, we have to deal with the present form of a finite auxiliary and the past participle form of a lexical verb. We will first provide a representation of the former. Since ha is the thirdperson singular form of the verb avere, the corresponding construction, shown in Fig. 3.35 below, is labeled AvereAuxSimplePresent3Sg. Fig. 3.35: the AvereAuxSimplePresent3Sg construction construction AvereAuxSimplePresent3Sg subcase of AvereAux, SimplePresentTense, Singular, 3rdPerson constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← simplepresent // inh. self.features.number ← singular // inh. self.features.person ← 3 // inh. form: WordForm // inh. constraints self.f.phon ← /a/ meaning: Function // inh. constraints self.m.function ← grounding // inh. Since AvereAuxSimplePresent3Sg is a subcase of the AvereAux construction, such a construction is shown in Fig. 3.36 below35. The constructional block of the AvereAux construction is bound to the AuxFeatureSet schema, shown in Fig. 3.37. It was inherited from the Aux construction, shown in Fig. 3.38 below. 35 The more general Avere construction (not shown) is similar to the Lanciare construction shown in Fig. 3.1 above, with the difference that its meaning is unspecified, since avere can be either a lexical or an auxiliary verb, and therefore it is to be considered as a subcase of HasVerbFeatures, like Verb (the general construction for lexical verbs) or Aux (the general construction for auxiliaries). 156 3. An ECG analysis of Italian caused-motion constructions Fig. 3.36: the AvereAux construction construction AvereAux subcase of Avere, Aux constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← base // inh. form: WordForm // inh. constraints self.f.phon ← /a'vere/ // inh. meaning: Function // inh. constraints self.m.function ← grounding // inh. Fig. 3.37: the AuxFeatureSet schema schema AuxFeatureSet subcase of VerbFeatureSet roles expectedverbform verbform // inh. number // inh. Fig. 3.38: the Aux construction construction Aux subcase of Verb constructional: AuxFeatureSet constraints self.features.expectedverbform ← nonfinite form: WordForm // inh. meaning: Function constraints self.m.function ← grounding person // inh. Being an auxiliary verb, the meaning of the AvereAuxSimplePresent3Sg construction does not specify an x-net schema (which will be provided by the lexical verb), but clarifies that the AvereAuxSimplePresent3Sg has the function of grounding. Indeed, auxiliary verbs are functional words which indicate morphological markers (mode, tense, person, and to some extent also aspect) rather than denote processes. Nevertheless, as often pointed out in CG, this does not mean that they are meaningless; rather, they can be considered as grounding predications, i.e. elements which link linguistic expressions to the time and place of speaking, the participants in the conversation and their shared knowledge36. More specific information about the process described by the sentence is provided by the lexical verb, sollevato. It represents the unmarked part participle form of the verb sollevare; consequently, the corresponding contructions is labeled Sollevare1UnmarkedPastParticiple37. This construction is illustrated in Fig. 3.39. Sollevare1UnmarkedPastParticiple inherits the binding of 36 Langacker (2008: 126) underlines that “When have or be combines with a participle, the former imposes its temporality on the latter's more specific content” . 37 The reason why we use the term “unmarked” to label this construction will be made clear in §3.2.2. 157 Grounding Meaning in Everyday Experience in the World its meaning to the CauseMotionAction schema, also filling the x-net role with the correct value, which is one of lifting. Notwithstanding the fact that the sentence can still be analyzed making use of the ActiveCauseMotion1 construction at the phrasal level and the Declarative construction at the clausal level, we need to posit an intermediate construction in order to unify the auxiliary verb with the A-S construction. In order to accomplish this duty, we can make use of the FiniteAuxPlusVP construction, shown in 3.40 below. Fig. 3.39: the Sollevare1UnmarkedPastParticiple construction construction Sollevare1UnmarkedPastParticiple subcase of Sollevare1, PastParticiple constructional: VerbFeatureSet // inh. constraints self.features.verbform ← pastparticiple // inh. form: WordForm // inh. constraints self.f.phon ← /sol:e'vato/ meaning: CauseMotionAction // inh. constraints self.m.x-net ← @lift // inh. Fig. 3.40: the FiniteAuxPlusVP construction construction FiniteAuxPlusVP subcase of VP constructional constituents aux: Aux vp: VP constraints self.features.verbform ← finite self.features ↔ aux.features aux.features.expectedverbform ↔ vp.features.verbform form constraints aux.f before vp.f meaning evokes EventDescriptor as ed // inh. constraints self.m ↔ vp.m ed ↔ vp.ed self.m ↔ ed.eventType // inh. 158 3. An ECG analysis of Italian caused-motion constructions In the constructional block of Fig. 3.40, the verbform of the FiniteAuxPlusVP construction as a whole is bound to be finite, while the features of the construction are identified with the features of the auxiliary verb. Then, the expectedverbform role of Aux is bound to the verbform of vp. As a result, we have a finite auxiliary verb which will unify with the lexical verb of vp. Furthermore, the meaning block of FiniteAuxPlusVP evokes the same EventDescriptor schema evoked by ActiveCauseMotion1 and binds its meaning to that of the ActiveCauseMotion1 construction and to the eventType role of EventDescriptor. As a consequence, FiniteAuxPlusVP will also unify with the fin constituent the of Declarative construction. Having added the relevant constraints necessary to combine the auxiliary verb with the A-S construction, we can now turn to the NPs and the PP involved in the sentence under consideration. We have a subject NP (il giovane netturbino), a direct object NP (un bidone), and a path PP (dal marciapiede). This time, the subject NP also contains an adjective. With regard to the adjective giovane, it is important to underline that it is one of those Italian adjectives whose male and female forms are identical (both in the singular and in the plural numbers), so we are again at a crossroads: we can posit two different constructions, one for each gender, or only one construction which does not specify anything about gender. Consistent with the solution choosen above for the number of the noun portacenere, we opt for the former solution. The constructions corresponding to the noun Netturbino (a monosemous noun) and to the adjective Giovane are shown in Fig. 3.41 and Fig. 3.42 respectively. The Giovane1m construction is a subcase of the Giovane1 construction (not shown), which specifies the sense of the polysemous adjective, and the Male construction, deriving all their features. Giovane1 is a subcase of the more general Giovane construction (not shown). The Giovane construction is a subcase of the SingularAdjective construction and its meaning is bound to the ScalarModifier schema (then inherited by the Giovane1m construction), respectively shown in Fig. 3.43 and Fig. 3.44 below. Fig. 3.41: the Netturbino construction construction Netturbino subcase of MaleSingularNoun constructional: NominalFeatureSet constraints grammgender ← male // inh. number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /net:ur'bino/ meaning: RD // inh. constraints self.m.category ← Dustman self.m.gender ← male // inh. self.m.number ← one // inh. Fig. 3.42: the Giovane1m construction construction Giovane1m subcase of Giovane1, Male constructional: NominalFeatureSet // inh. constraints self.features.gender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'ʤovane/ // inh. meaning: ScalarModifier // inh. evokes RD as rd // inh. constraints self.m.scalarvalue ← young // inh. self.m.scale ← age // inh. self.m.modifiedEntity ↔ rd // inh. rd. category ← animate // inh. 159 rd.natgender ← male rd.number ← one // inh. Grounding Meaning in Everyday Experience in the World Fig. 3.43: the SingularAdjective construction construction SingularAdjective subcase of Adjective, Singular constructional: NominalFeatureSet // inh. constraints self.features.number ← singular // inh. form: WordForm // inh. meaning: Modifier // inh. evokes RD as rd // inh. constraints modifiedEntity ↔ rd // inh. rd.number ← one Fig. 3.44: the ScalarModifier schema schema ScalarModifier subcase of Modifier roles scale scalarValue modifiedEntity // inh. While the Netturbino construction does not show any particularly interesting features, the Giovane1m constructions deserves to be payed attention, especially for its meaning. As a matter of fact, the role of the ScalarModifier schema is crucial: there are several concepts, such as age, which cannot be seen as having clear-cut boundaries, but rather must be considered as matters of degree (see e.g. Taylor 2003; Langacker 2008). Accordingly, the ScalarModifier schema makes it clear that adjectives such as “young” should not be approached as absolute values, but should be considered as portions of a continuum (see §3.3.3). Netturbino and Giovane1m are combined in the ModifiedNP1 construction38 shown in Fig. 3.45, which establishes a formal constraint on the order of the constituents (i.e. the modifier precedes the noun) and a semantic constraint which binds the meaning of the complex construction to the meaning of both constituent constructions. Then, the resulting phrase is combined with the Il construction by the DeterminerPlusNP construction. 38 This construction is labeled ModifiedNP1 and posited to be a subcase of ModifiedNP due to formal reasons: in Italian the order of modifier and NP is not rigid, but can vary on the basis of syntactic, semantic, pragmatic, or stylistic factors, as will be examplified later. 160 3. An ECG analysis of Italian caused-motion constructions Fig. 3.45: the ModifiedNP1 construction construction ModifiedNP1 subcase of ModifiedNP constructional: NominalFeatureSet constituents mod: Modifier // inh. np: NP // inh. form constraints mod.f before np.f meaning constraints self.m ↔ np.m // inh. self.m ↔ mod.m // inh. The DeterminerPlusNP construction will then unify with Declarative's subj constituent, providing the information that the entity that fills the causer and profiledParticipant role is one, male, young dustman. The level of accessibility of il giovane netturbino is uniquely identifiable, for the use of a definite determiner denotes the possibility to uniquely identify the referent, as already mentioned above when discussing the NP i bambini. ActiveCauseMotion1's np constituent will unify with the NP un bidone, providing more specific information on the filler of the affected role. The Bidone1 construction and the Un construction are shown in Fig. 3.46 and Fig. 3.47 below. They will be combined together in the DeterminerPlusNP construction. Un is the first indefinite article we encounter in the present study39. It is relevant to note that the level of accessibility of the referent defined by the indefinite article is type-identifiable, since it provides information on the kind of entity represented by noun component of the NP, but does not denote a uniquely identifiable entity. 39 The hierarchy of indefinite articles is specular to that of definite articles. Consequently, the MaleSgIndefiniteArticle construction is a subcase of SgIndefiniteArticle, which in turn is a subcase of IndefiniteArticle, itself a subcase of the Determiner construction. 161 Grounding Meaning in Everyday Experience in the World Fig. 3.46: the Bidone construction construction Bidone1 subcase of Bidone constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /bi'done/ // inh. meaning: RD // inh. constraints self.m.category ← Garbage can self.m.natgender ← neuter self.m.number ← one // inh. Fig. 3.47: the Un construction construction Un subcase of MaleSgIndefiniteArticle constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /un/ meaning evokes RD as rd // inh. constraints rd.number ← one // inh. rd.accessibility ← type-identifiable // inh. Now, we have to deal with the PP dal marciapiede. In Fig. 3.48 below you can find the Marciapiede1 construction. Dal is a complex preposition, composed by the simple preposition da and the definite article il. The preposition da can be used both to express the process opposite to that denoted by in (see below) and to that identified by su40 (see above). In this case, the latter condition is the relevant one, as shown by the Da2a construction represented in Fig. 3.49, subcase of Da (not shown). Fig. 3.48: the Marciapiede1 construction construction Marciapiede1 subcase of Marciapiede constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /marʧa'pjede/ // inh. meaning: RD // inh. constraints self.m.category ← Sidewalk self.m.number ← one // inh. 40 The preposition da is also used to express causation in passive constructions, as will be shown in §3.2.2. 162 3. An ECG analysis of Italian caused-motion constructions Fig. 3.49: the Da2a construction construction Da2a subcase of PathPreposition, Da form: WordForm // inh. constraints self.f.phon ← /da/ // inh. meaning: SPG // inh. evokes BoundedObject as bo constraints self.m.landmark ↔ bo.boundedobject self.m.profiledArea ↔ bo.boundary self.m.source ↔ self.m.profiledArea self.m.goal ↔ bo.exterior The meaning of Da2a is constrained to be SPG and it evokes the BoundedObject schema already exploited in the representation of Su2 above. Nevertheless, this time the trajector of SPG had to go from the profiledArea, i.e. the surface of an object (in this case, a specific sidewalk), represented by the boundary role, to the region outside the object, represented by the exterior role. As a result, this time the profiledArea role of TL is bound to the source role of SPG. Then, the Da2a construction is combined with the Il construction in the ComplexPathPreposition construction. Once more, we have to precise that the phonological form of the complex preposition is not the simple fusion of da + il (i.e. */dail/), but rather the phonological string /dal/. The ComplexPathPreposition construction is finally combined with the Marciapiede construction in the PathPP construction. As usual, PathPP has two constituents: prep, constrained to be a PathPreposition, and np, constrained to be an NP. The form of np is constrained to follow that of prep. Moreover, meaning constraints specify that the meaning of PathPP is bound to the meaning of prep, and that its landmark role is bound to the referent of np. The landmark role of SPG unifies with the entity represented by the noun Marciapiede, and the pp constituent of ActiveCauseMotion1 unifies with the PathPP construction, specifying the path followed by the mover of MotionAlongAPath (i.e. the affected of CauseMotionAction): from the surface of the object defined by the noun Marciapiede to an unprecised surrounding region. The SemSpec produced by the unification of the constructions instantiated by this sentence will be similar to the one of the sentences analyzed above. The main difference will be represented by the profiledProcess role of the EventDescriptor schema, filled by the meaning of SollevarePastParticiple, which maintains a basic meaning of CauseMotionAction, but its x-net is one of lifting. Furthermore, being conjugated in the present perfect tense, the verb specifies that the event described in this sentence is a completed action set in the past (see Proudfoot and Cardo 2005: 155-158; Moneti and Lazzarini 2007: 82-84). Again, this SemSpec, will capture the conceptual core of our sentence, driving the enactment of an event in which a uniquely identifiable young male dustman performs a lifting action on an unspecified garbage can, obtaining its upward 163 Grounding Meaning in Everyday Experience in the World movement from a uniquely identifiable sidewalk. Once more, the event will be enacted from the perspective of the agent participant. The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agent (in this case, it seems probable that the force amount spent by the agent is rather high), the purpose on which he performed the action (probably connected to his job), the subsequent actions (e.g., getting the garbage bag out of the can and loading it an on a truck), etc. The words bidone and marciapiede hint that the action took place in a street, while the noun netturbino suggests that operations of routine maintainance of such a street were going on. Then, on the basis of their world knowledge and the (discourse and situational) context, the language understander will be able to mentally build the scene more precisely. So far, we have dealt with three active caused-motion constructions which can be considered as instantiating the same A-S construction. In the next pages, we shall conclude this subsection on active constructions illustrating a simple clause which shows a slight less prototypical structure than the three sentences outlined above. Consider the sentence in (25) below: (25) Spingevamo le ceste nel Push:IMPF.1PL the.FPL basket.PL into-the.MSG “We used to push the baskets into the lumber-room” ripostiglio. lumber-room.SG The structure of this sentence is highly similar to that of the ones already observed. Nevertheless, its meaning is a bit different, since the meaning of the verb spingere does not exactly match the meaning of the A-S construction, as highlighted by Dodge (2010b) for the English corresponding verb push. Rather, it unifies with just a part of the A-S construction, precisely, the first process which is part of the CauseMotionAction (namely ForceApplication) schema. Indeed, if we consider the verb spingere alone (i.e. outside the context of caused-motion constructions), we can notice that it denotes an action in which a causer participant exerts force on a patient participant in order to move it, but it does not entail that the causer reaches their goal. That is, the verb spingere can also be used to denote an action in which the causer exerts force on the patient, but fails to move it. The same condition cannot be observed with any of the other three verbs involved in the three sentences taken into consideration above. Therefore, we shall follow Dodge (2010b) in adopting a different subcase of the ActiveCauseMotion construction, in which this slight difference between the sentence in (25) and those previously illustrated is captured. This extension will be labeled ActiveCauseMotion2 and shown in Fig. 3.50 below. 164 3. An ECG analysis of Italian caused-motion constructions Fig. 3.50: the ActiveCauseMotion2 construction construction ActiveCauseMotion2 subcase of ActiveCauseMotion constructional constituents v: Verb // inh. np: NP // inh. pp: PathPP // inh. form constraints v.f before np.f // inh. np.f before pp.f // inh. meaning: CauseMotionAction // inh. evokes EventDescriptor as ed // inh. constraints v.m ↔ self.m.process1 self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m.affected ↔ np.m // inh. self.m.causer ↔ ed.profiledParticipant ed.profiledParticipant ↔ ed.topic self.m.process2 ↔ pp.m.spg // inh. The most important consequence of the unification between the verb and the ForceApplication schema is the unification between the profiledProcess role of the EventDescriptor schema with such a process rather than with the CauseMotionAction schema. Apart from the fact that the first process is awarded a “privileged” status (particularly important for the enactment process), the sentence under consideration does not remarkably diverge from the other three analyzed in this section. The meaning of the construction is still bound to the CauseMotionAction schema and, at the clausal level, we can still exploit the Declarative construction to describe it. The other points which distinguish the sentence in (25) concern the lexical verb construction and the NPs and the PP involved. A representation of the lexical verb construction can be found in Fig. 3.51 below. Because spingevamo is the first-person plural form of the verb spingere in the imperfect tense, this construction is labeled Spingere1Imperfect1Plural. The Spingere1Imperfect1Pl construction is a subcase of the Spingere1, Imperfect, the Plural, and the 1stperson constructions. Its form pole is constrained to be a word by the WordForm schema (see Fig. 2.10), with the phon role which provides its specific phonological structure. The meaning pole of SpingereImperfect1Pl is constrained to be ForceApplication (Fig. 2.12). The x-net role of SpingereImperfect1Pl is also assigned the proper value, namely one of pushing. 165 Grounding Meaning in Everyday Experience in the World Fig. 3.51: the Spingere1Imperfect1Pl construction construction Spingere1Imperfect1Pl subcase of Spingere1, ImperfectTense, Plural, 1stperson constructional: VerbFeatureSet // inh. constraints self.features.verbform ← imperfect // inh. self.features.number ← plural self.features.person ← 1 // inh. form: WordForm // inh. constraints self.f.phon ← /spinʤe'vamo/ meaning: ForceApplication // inh. constraints self.m.x-net ← @push // inh. // inh. Having established the general constraints on the fillers of the participant roles which depart from those observed in the analysis of the previous sentence, we can directly turn to the NPs and PPs, which provide more specific information on the fillers of these roles for the particular event described by the sentence-level construction under consideration. In this sentence, we can observe two NP constructions (one for the null subject and one for the noun phrase le ceste) and one PP construction (for the spatial phrase nel ripostiglio). The null subject construction, labeled DropNoi, is shown in Fig. 3.52 below, while in Fig. 3.53 you can find the Noi construction, of which DropNoi is a subtype. Fig. 3.52: the DropNoi construction construction DropNoi subcase of Noi constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh. self.features.number ← plural // inh. self.features.person ← 1 // inh. form: WordForm // inh. constraints ignore: self.f.phon ← /noi/ meaning: RD // inh. constraints self.m.category ← Animate // inh. self.m.number ← morethanone // inh. self.m.accessibility ← active // inh. self.m.referent ← currentspeakerandotheranimate // inh. 166 3. An ECG analysis of Italian caused-motion constructions The referent of Declarative's subj constituent will unify with DropNoi, providing the information that the entity that fills the causer and profiledParticipant role is a morethanone, animate, referent represented by the current speaker together with at least one more animate entity. The level of accessibility of a first-person plural pronoun is active, since pronouns are generally used to denote active referents, as already specified above. The ontological category of the referent is assigned the value animate, since the pronouns just lets us know that one of them is human, while we do not know anything about the other(s), who may be a human being or an animal41. ActiveCauseMotion2's np constituent will unify with the NP whose noun constituent is Ceste1, providing more specific information on the filler of the affected role. The Ceste1 construction is represented in Fig. 3.54 below, while the Le construction is shown in Fig. 3.55. Fig. 3.53: the Noi construction construction Noi subcase of PluralSubjPersPron, 1stPerson constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh. self.features.number ← plural self.features.person ← 1 // inh. form: WordForm // inh. constraints self.f.phon ← /noi/ meaning: RD // inh. constraints self.m.category ← Animate self.m.number ← morethanone // inh. self.m.accessibility ← active // inh. self.m.referent ← currentspeakerandotheranimate // inh. 41 Again, the situation will be disambiguated when all the constructions instantiated in the sentence unify. 167 Grounding Meaning in Everyday Experience in the World Fig. 3.54: the Ceste1 construction construction Ceste1 subcase of Ceste constructional: NominalFeatureSet // inh. constraints grammgender ← female // inh. number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /'ʧeste/ // inh. meaning: RD // inh. constraints self.m.category ← Basket self.m.natgender ← neuter // inh. self.m.number ← morethanone // inh. Fig. 3.55: the Le construction construction Le subcase of FemPlDefiniteArticle constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /le/ meaning evokes RD as rd // inh. constraints rd.number ← morethanone // inh. rd.accessibility ← uniquely-identifiable // inh. Finally, we come to the PP nel ripostiglio. First of all, the reader can find a representation of the Ripostiglio (a monosemous noun) construction in Fig. 3.56 below. Fig. 3.56: the Ripostiglio construction construction Ripostiglio subcase of MaleSingularNoun constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /ripo'stiʎo/ meaning: RD // inh. constraints self.m.category ← Lumber-room self.m.natgender ← neuter self.m.number ← one // inh. Nel is a complex preposition, composed of the simple preposition in, and the determiner il. Fig. 3.57 illustrates the In2 construction, subcase of In and PathPreposition. 168 3. An ECG analysis of Italian caused-motion constructions Fig. 3.57: the In2 construction construction In2 subcase of PathPreposition, In form: WordForm // inh. constraints self.f.phon ← /in/ // inh. meaning: SPG // inh. evokes BoundedObject as bo constraints self.m.landmark ↔ bo.boundedobject self.m.profiledArea ↔ bo.interior self.m.goal ↔ self.m.profiledArea self.m.source ↔ bo.exterior As already mentioned, the In2 construction is quite similar to the Su2 construction. Binding the proximalArea role of the SPG schema to the interior role – instead of the boundary – role of the BoundedObject schema allows us to recognize the similarities but also the very important differences between these two constructions. Indeed, while the Su2 construction describes the motion of the trajector role of SPG from a regioe outside an object to its surface, the In2 construction denotes a motion from an outer area to the internal region of an object. The ComplexPathPreposition construction specifies that the form of In2 fuses with the form of Il, while the meaning of the construction sums that of the simple preposition with that of the determiner. Again, the form of the complex preposition nel is not a simple agglutination of in + il (which would presumibly be */inil/), but /nel/. The ComplexPathPreposition construction and the Ripostiglio construction will be finally combined together in the PathPP construction. The landmark role of SPG unifies with the entity represented by the noun Ripostiglio, and the pp constituent of ActiveCauseMotion2 unifies with the PathPP construction, specifying the path followed by the mover of MotionAlongAPath (i.e. the affected of CauseMotionAction): from the exterior to the interior of the object defined by the noun Ripostiglio. The SemSpec produced by the unification of the constructions instantiated by this sentence will be a bit different from the one of the sentences analyzed above. The main difference will be represented by the profiledProcess role of the EventDescriptor schema, filled by the meaning of Spingere1Imperfect1Sg, which unifies with ForceApplication, and its x-net is one of pushing. This is a very important point, for it signals that the enactment process will be prominently focus on ForceApplication. Also, the verb specifies the setting time of the event: in Italian (as in other Romance languages), the imperfect is normally used to express repetition and continuity in the past, though other uses are possible (for a brief discussion, see Polesini Karumanchiri and Uslenghi Maiguashca 1988: 166-175). Again, this SemSpec will capture the conceptual core of our sentence, driving the enactment of an event in which the current speaker and at least one more animate being perform a pushing action on more than one 169 Grounding Meaning in Everyday Experience in the World uniquely identifiable baskets, obtaining their movement into a uniquely identifiable lumber-room. Once more, the event will be enacted from the agent's perspective. The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agent, the purpose on which he performed the action, the kind of baskets which undergo the action etc. The words ceste and ripostiglio will hint that probably the action took place in a domestic environment. Then, on the basis of their encyclopedic knowledge and the (discourse and situational) context, the language understander will be able to mentally reconstruct the scene more in detail. So far, we have analyzed four simple active caused-motion constructions in Italian, showing that the CauseMotionAction schema and two subcases of the ActiveCauseMotion A-S construction allows us to define a certain degree of generalization over several verbs of caused-motion. Indeed, we have analyzed sentences involving the verbs lanciare (“to throw”), posare (“to lay”), sollevare (“to lift”), and spingere (“to push”),: these four verbs describe quite different processes, but they share some common characteristics. In particular, the CauseMotionAction schema captures the fact that all the sentence-level constructions analyzed above denote events in which an agent participant makes a patient participant move in a certain direction by intentionally exerting force on it, while the semantic peculiarity of each verb is specified by their x-net role. At the same time, the ActiveCauseMotion1 construction captures the fact that all these events are described from the agent's perspective. In the case of the sentence involving the verb spingere, we noticed that such a verb shows a slight semantic peculiarity, namely the meaning of the verb fuses with the ForceApplication process rather than the complex process as a whole; therefore, we have preferred to analyze such a sentence using another subcase of the ActiveCauseMotion, labeled ActiveCauseMotion2, able to capture this (rather minimal) difference. Moreover, the PathPP construction used in the analysis of all the examples above, each of which involves a different preposition, proved to be able to let us capture different kinds of movement through space. In addition, in one of our sentences we also dealt with the issues of compound verbs and modification, succeeding to handle such phenomena quite “painlessly”. In §3.2.2, we will exploit the ECG formalism to carry out an analysis of some passive caused-motion constructions which are, by their very nature, a bit more complex than active constructions (at least in Italian). 3.2.2. Passive caused-motion constructions In the present subsection, we are going to analyze some passive sentence-level constructions, whose meaning does not diverge from the meaning of the active constructions observed in §3.2.1, but whose syntactic structure is quite different. First of all, consider the sentence in (26) below, whose verb is in the future perfect tense. (26) Una pietra sarà scagliata contro la folla da un infiltrato. A.FSG stone.SG be:FUT.3SG hurl:PSTPART.FSG against the.FSG crowd.SG by an.MSG infiltrator.SG “A stone will be hurled against the crowd by an infiltrator” Since the general meaning of this sentence is analogous to those of the active sentences 170 3. An ECG analysis of Italian caused-motion constructions outlined above, the analysis of this sentence-level construction will involve the same schemas used for their investigation in the previous subsection. However, it will be necessary to make use of some different constructions to represent its different structure. First of all, we need to posit an adequate A-S construction. This duty can be performed by the PassiveCauseMotion1 construction shown in Fig. 3.58 below, which exactly represent the passive counterpart of the ActiveCauseMotion1 construction illustrated in Fig. 3.9 above. Fig. 3.58: the PassiveCauseMotion1 construction construction PassiveCauseMotion1 subcase of PassiveCauseMotion constructional constituents v: PastParticiple // inh. pp1: PathPP // inh. optional pp2: CausativePP // inh. form constraints v.f before pp1.f pp1.f before pp2.f meaning: CauseMotionAction // inh. evokes EventDescriptor as ed // inh. constraints self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m ↔ v.m self.m.affected ↔ ed.profiledParticipant // inh. self.m.causer ↔ pp2.m.np // inh. ed.profiledParticipant ↔ ed.topic self.m.process2 ↔ pp1.m.spg The constructional block of this phrasal construction defines the following constituents: a verb constituent, labeled v, which is bound to the past participle tense, a pp1 constituent, bound to be a path PP, which represents the motion of the patient participant along a path, and an optional pp2 constituent constrained to be a causative PP, which captures the agent participant. The form block adds two form constraints: the verb's form (v.f) is constrained to precede the path PP's form (pp1.f), which in turn is contrained to come before the causative PP's form (pp2.f). The meaning block of PassiveCauseMotion1, quite obviously, identifies itself with the CauseMotionAction schema, like its active counterpart, and it inherits an evoked EventDescriptor schema, to specify how the scene described by the sentence should be simulated. Therefore, PassiveCauseMotion1 also inherits the fact that its meaning as a whole is bound to the eventType role of the EventDescriptor schema, while the 171 Grounding Meaning in Everyday Experience in the World meaning of its verb is bound to the profiledProcess of the same schema. Moreover, the meaning of the construction itself is co-indexed with the meaning of the verb, while its second process is co-indexed with the SPG schema which is the meaning of the pp. In addition to these constraints, unlike its corresponding active A-S construction, PassiveCauseMotion1 constrains its affected participant to be co-indexed with the profiledParticipant role of EventDescriptor, and its causer participant to be coindexed with the np which is part of the pp2 constituent. These two constraints, inherited from the more general PassiveCauseMotion construction, play a key role in capturing the difference between the active and the passive voice. Moving to the clause-level, we can exploit Declarative, as usual, but since we have to deal with a passive compound verb, we need to posit the FiniteAuxPlusVP construction in order to unify the auxiliary and the A-S construction. The FiniteAuxPlusVP construction was already used in the analysis of the sentence (24) above and shown in Fig. 3.38. As already underlined, such a construction will unify with the fin constituent of Declarative. Now, we can focus on the auxiliary verb sarà, the third-person singular form of the verb essere in the simple future tense, whose corresponding construction is labeled EssereAuxSimpleFuture3sg and shown in Fig. 3.59 below. Fig. 3.59: the EssereAuxSimpleFuture3sg construction construction EssereAuxSimpleFuture3sg subcase of EssereAux, SimpleFutureTense, Singular, 3rdPerson constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← simplefuture // inh. self.features.number ← singular // inh. self.features.person ← 3 // inh. form: WordForm // inh. constraints self.f.phon ← /sa'ra/ meaning: Function // inh. constraints self.m.function ← grounding // inh. Since EssereAuxSimplePresent3Sg is a subcase of the EssereAux construction, such a construction is shown in Fig. 3.60 below. As in the case of the AvereAux construction (Fig. 3.36 above), the constructional block of the EssereAux is bound to the AuxFeatureSet schema (Fig. 3.37). 172 3. An ECG analysis of Italian caused-motion constructions Fig. 3.60: the EssereAux construction construction EssereAux subcase of Essere, Aux constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← base // inh. form: WordForm // inh. constraints self.f.phon ← /'es:ere/ // inh. meaning: Function // inh. constraints self.m.function ← grounding // inh. The reader will have noticed that, unlike the example in (24) above, the auxiliary verb involved in the sentence-level construction we are considering now is essere (literally “to be”), rather than avere (“to have”). This is due to the fact that Italian transitive verbs take the auxiliary verb avere in active clauses, but essere in passive sentences42. As already pointed out, the meaning of the auxiliary verb does not specify an x-net schema (which will be provided by the lexical verb), but performs a grounding function. Therefore, information about the process described by the sentence is provided by the lexical verb, scagliata. It is worth remarking that the past participle of verbs which in compound forms take the auxiliary essere agree with the subject in gender and number43. Consequently, scagliata represents the female singular past participle form of the verb scagliare; the corresponding contructions is therefore labeled Scagliare1FemSgPastParticiple. This construction is illustrated in Fig. 3.61 below. Scagliare1FemSgPastParticiple binds its meaning to the CauseMotionAction schema, adding information on the correct x-net, which is one of hurling, while its form is constrained to be a word and is assigned the correct phonological form. Now that we have established the general constraints on the fillers of the participant roles, we can focus our attention to the NP and the PPs present in the clause, which provide more specific information on the fillers of these roles for the particular event described by the sentence in (26). We can find a subject NP (una pietra) and two PPs: a spatial phrase (contro la folla) and a causative one (da un infiltrato). The subject NP sums an indefinite article and a female singular noun, shown in Fig. 3.62 and Fig. 3.63 below, respectively. These two constructions will be unified by the DeterminerPlusNP construction. 42 With regard to intransitive verbs, the situation is less clear, but this point goes beyond the scope of the present analysis. 43 The status of past participle forms found in Italian passive sentences can arguably be seen as a sort of intermediate level intersection point between verbs and adjectives, sharing features with more prototypical members of both classes: in a sense, they look similar to predicate adjectives (see Croft 2001: 88). This phenomenon is also present in other Romance languages: cfr. Gómez Torrego (2002: 140-141) on Spanish participles. 173 Grounding Meaning in Everyday Experience in the World Fig. 3.61: the Scagliare1FemSgPastParticiple construction construction Scagliare1FemSgPastParticiple subcase of Scagliare1, PastParticiple, Singular, Female constructional: VerbFeatureSet // inh. constraints self.features.verbform ← pastparticiple // inh. self.features.grammgender ← female // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /ska'ʎata/ meaning: CauseMotionAction // inh. evokes RD as rd constraints self.m.x-net ← @hurl rd.number ← one Fig. 3.62: the Una construction construction Una subcase of FemSgIndefiniteArticle constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'una/ meaning evokes RD as rd // inh. constraints rd.number ← one // inh. rd.accessibility ← type-identifiable // inh. Fig. 3.63: the Pietra1 construction construction Pietra1 subcase of Pietra constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'pjetra/ // inh. meaning: RD // inh. constraints self.m.category ← Stone self.m.natgender ← neuter // inh. self.m.number ← one // inh. The referent of the subj constituent of the Declarative construction will then unify with the DeterminerPlusNP construction, supplying the information that the entity that fills the affected and profiledParticipant role is one and it is represented by a member of the category Stone. The level of accessibility of una pietra is type-identifiable, since the use of an indefinite determiner denotes the impossibility to identify the accessibility of the referent, but permits to identify its category. 174 3. An ECG analysis of Italian caused-motion constructions We can now turn to the path PP contro la folla, which is made up of a path preposition and an NP similar to those just observed. A representation of the somewhat controversial noun folla is provided in Fig. 3.64 below. The referent of a word like folla is not exactly a mass (cfr., e.g., sangue “blood” or sabbia “sand”) but rather a collector of people. On the one hand, since it is generally perceived as a sort of mass, the mass role of the RD schema is filled by the value yes. On the other hand, since the word folla functions in a way more similar to a classifier44 than to a proper mass noun, its number role is also filled by the value one (though it is uncountable). The Folla1 construction will be unified with the La construction by the DeterminerPlusNPNoun construction. The La construction will add the information that the accessibility of the referent is uniquely-identifiable. Now we have to deal with the preposition contro, whose corresponding construction is labeled ControPrep2 (Fig. 3.65), and is a subcase of ControPrep45 (not shown) and PathPreposition. Fig. 3.64: the Folla1 construction construction Folla1 subcase of Folla constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'fol:a/ // inh. meaning: RD // inh. constraints self.m.category ← Crowd self.m.mass ← yes // inh. self.m.number ← one // inh. Fig. 3.65: the ControPrep2 construction construction ControPrep2 subcase of PathPreposition, ControPrep form: WordForm // inh. constraints self.f.phon ← /'kontro/ // inh. meaning: SPG // inh. evokes Proximity as p constraints self.m.landmark ↔ self.m.goal self.m.landmark ↔ p.center self.m.profiledArea ↔ p.proximalArea ControPrep2's meaning is analogous to that of the A2 construction seen in the previous section. The meaning of the A2 construction is constrained to be SPG and its landmark role is bound to the goal role, while the trajector of this motion is the mover role of MotionAlongAPath. i.e. the affected role of CauseMotionAction. Identifying the goal of the motion with the landmark we define the type of event being described: a motion toward an entity. The ControPrep2 and the DeterminerPlusNP construction will then be combined together in the PathPP construction. PassiveCauseMotion1's 44 As a matter of fact, the crowd can be seen as a collector of people. When Italians say folla, they imply folla di persone (literally “crowd of people”). Therefore, folla assumes a meaning similar to that of gregge (“herd”) or sciame (“flock”). Though a detailed discussion of mass nouns is beyond the scope of our study, the solution adopted here is to consider folla as having both a singular meaning and a mass meaning. For a brief, but deep and insightful discussion of this kind of nouns, the reader is advised to see Dodge and Wright's (2002) paper on this phenomenon in English, for the findings of their study seem (in large measure) revealing for Italian (along with other languages) as well. 45 The choice of this label is due to the fact that contro can also be an adverb. 175 Grounding Meaning in Everyday Experience in the World pp1 constituent will unify with this path PP. Finally, we come to the causative PP da un infiltrato, composed by a causative preposition and an NP. We begin providing a representation of the noun, illustrated in Fig. 3.66 below. The Infiltrato construction will then unify with the Un construction via the DeterminerPlusNP. The Un construction lets us know that the level of accessibility of the referent is type-identifiable. The construction corresponding to the causative preposition da, labeled Da3, is shown in Fig. 3.67. Fig. 3.66: the Infiltrato1 construction construction Infiltrato1 subcase of Infiltrato constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /infil'trato/ // inh. meaning: RD // inh. constraints self.m.category ← Infiltrator self.m.number ← one // inh. Fig. 3.67: the Da3 construction construction Da3 subcase of CausativePreposition, Da form: WordForm // inh. constraints self.f.phon ← /da/ // inh. meaning: Causation // inh. evokes CauseEffectAction as cea constraints self.m.causer ↔ cea.causer The Da3 construction is a subcase of the Da construction and of the CausativePreposition construction, shown in Fig. 3.68 below, which binds its meaning to the Causation schema (Fig. 3.69). Fig. 3.68: the CausativePreposition construction construction CausativePreposition subcase of Preposition form: WordForm // inh. meaning: Causation Fig. 3.69: the Causation schema schema Causation subcase of Relation roles event causer The Da3 and the DeterminerPlusNP construction are then combined in the CausativePP construction, shown in Fig. 3.70 below. The constructional block of the CausativePP construction specifies that its constituents are a causative preposition (prep) and a noun phrase (np). The form of np is constrained to follow that of prep. Moreover, meaning constraints specify that the meaning of PathPP is bound to the meaning of prep, and that its protagonist role is bound to np. The CausativePP construction will unify with the pp2 constituent of the PassiveCauseMotion1 176 3. An ECG analysis of Italian caused-motion constructions construction, providing the information that its np constituent is the filler of the causer role. Fig. 3.70: the CausativePP construction construction CausativePP subcase of PP constructional constituents prep: CausativePreposition np: NP // inh. form constraints prep.f before np.f // inh. meaning: Causation constraints self.m ↔ prep.m // inh. self.m.causer ↔ np.m The unification of the constructions instantiated in our example produces a SemSpec which supports an enactment of the event described by the sentence. The EventDescriptor schema provides several crucial simulation parameters. Its profiledProcess role is bound to the meaning of Scagliare1FemSgPastParticiple. It is a process of CauseMotionAction and has an x-net of hurling. The eventType role is bound to the meaning of PassiveCauseMotion1, in which the meaning of the argument structure is bound to that of its verb constituent. Therefore, profiledProcess is also bound to CauseMotionAction. As in the case of our prototypical transitive sentence exemplified in §2.3.2.3, the A-S construction provides information about the general type of event being described, and the verb adds more information about the specific processes involved in such an event, including information on their structure and setting time. Moreover, the unification of PassiveCauseMotion1, Scagliare1FemSgPastParticiple, Declarative, and the instantiated NP and PP constructions results in various bindings associated with each of the participants of this event. The affected role of CauseMotionAction is bound to: – protagonist2 of CauseMotionAction; – actedUpon of ForceApplication; – mover and protagonist of MotionAlongAPath; – profiledParticipant of EventDescriptor; – trajector of SPG; – the subj constituent of Declarative; – the referent of una pietra. Instead, the causer role of CauseMotionAction is bound to: – protagonist of CauseMotionAction; – actor and protagonist of ForceApplication; 177 Grounding Meaning in Everyday Experience in the World – the np constituent of the pp2 constituent of PassiveCauseMotion1; – the entity referred to as un infiltrato. Furthermore, PathPP specifies the direction of the motion undergone by the trajector, i.e. toward the entity referred to as la folla, which functions as the landmark and goal of the motion event. This SemSpec will drive the enactment of an event in which an infiltrator performs a forceful hurling action on an unspecified stone, obtaining its movement toward a uniquely identifiable crowd. The lexical elements included in the sentence will probably help in the mental reconstruction of the event adding some details. The presence of a noun such as infiltrato may evoke a scenario based on the rivalry between different organizations or groups of people, while a word like folla suggests the presence of a lot of people gathered in a place, and a noun like pietra hints that the event probably took place in the open air. Since the event is described from the perspective of the patient participant, it should be enacted from the same perspective, though it might not be exactly easy to simulate an event from the perspective of an inanimate entity. The process of enactment of this event will produce more inferences, about the identity of the agent, the purpose on which the agent performed the action, the amount of energy they spent and the amount of speed of the stone (probably, both very high in this case), the effector they exploited, and so on. Then, on the basis of their world knowledge and the (discourse and situational) context, the language understander will arrive at a deeper understanding of the utterance and may also generate further inferences. The next sentence-level construction we are going to analyze is similar to the one in (26), with which share the same A-S construction and the same core meaning. Nevertheless, it presents a slightly more complex grammatical structure, showing a different auxiliary and the presence of a cluster of two path PPs. Consider the sentence in (27) below: (27) Gli arbitri vennero scaraventati dal ring sul tavolo da The.MPL referee.PL come:PRT.3PL fling:PSTPART.MPL from-the.MSG ring on-the.MSG table.SG by Undertaker. Undertaker. “The referees were flung from the ring to the table by The Undertaker” These sentence is analogous to the previous one, but its verb is in the passive simple past tense. The points of divergence concern the auxiliary verb used and its morphological markers, the lexical verb construction, the NP and the PPs involved, while the basic meaning of the verb and the A-S construction remain the same. Furthermore, this sentence again exploits the same device used above to link the auxiliary verb and the A-S construction, namely the FiniteAuxPlusVP construction, which also overlaps with the fin constituent of our clause-level construction, Declarative. As a consequence, we shall provide the reader with just the remaining constructions: the auxiliary and the lexical verb constructions, the constructions related to the NP and those regarding the PPs. Vennero is the third-person plural form of the auxiliary verb venire (“to come”) in the simple past tense, the corresponding construction (shown in Fig. 3.71) is labeled VenireAuxSimplePast3Pl. To be accurate, the Italian language recognizes the status of auxiliary verbs to essere and avere only. Nevertheless, the passive conjugation of verbs in the simple tenses (simple present, simple past, imperfect, simple future), which 178 3. An ECG analysis of Italian caused-motion constructions would normally take the auxiliary verb essere, often replace it with the corresponding form of the verb venire, both in spoken and written Italian. The VenireAux construction is shown in Fig. 3.72 below. Fig. 3.71: the VenireAuxSimpleFuture3Pl construction construction VenireAuxSimplePast3Pl subcase of Venire Aux, SimplePastTense, Plural, 3rdPerson constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← simplepast // inh. self.features.number ← plural // inh. self.features.person ← 3 // inh. form: WordForm // inh. constraints self.f.phon ← /'ven:ero/ meaning: Function // inh. constraints self.m.function ← grounding // inh. Fig. 3.72: the VenireAux construction construction VenireAux subcase of Venire, Aux constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite inh. self.features.verbform ← base // inh. form: WordForm // inh. constraints self.f.phon ← /ve'nire/ // inh. meaning: Function // inh. constraints self.m.function ← grounding // inh. // The construction relative to the lexical verb scaraventati, which provides information about the precise type of event described by the sentence is illustrated in Fig. 3.73. Since scaraventati is the male plural form of the past participle of the verb scaraventare, the construction is labeled Scaraventare1MalePlPastParticiple. 179 Grounding Meaning in Everyday Experience in the World Fig. 3.73: the Scaraventare1MalePlPastParticiple construction construction Scaraventare1MaleSgPastParticiple subcase of Scaraventare1, PastParticiple, Plural, Male constructional: VerbFeatureSet // inh. constituents self.features.verbform ← pastparticiple // inh. self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /skaraven'tati/ meaning: CauseMotionAction // inh. evokes RD as rd constraints self.m.x-net ← @fling // inh. rd.number ← morethanone // inh. Scaraventare1MaleSgPastParticiple binds its meaning to the CauseMotionAction schema, adding information on the correct x-net, which is one of flinging, while its form is constrained to be a word and is assigned the correct phonological form. We can now turn to the NP and the PPs present in the sentence, which provide more specific information on the fillers of the roles for the particular event described by the utterance in (27). We can find a subject NP (gli arbitri) and two PPs: a cluster of two spatial phrases (dal ring sul tavolo) and a causative phrase (da Undertaker). The subject NP includes a definite article and a male plural noun, shown in Fig. 3.74 and Fig. 3.75 below, respectively. These two constructions will be unified by the DeterminerPlusNP construction. Fig. 3.74: the Gli construction construction Gli subcase of MalePlDefiniteArticle constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /ʎi/ meaning evokes RD as rd // constraints rd.number ← morethanone // inh. rd.accessibility ← uniquely-identifiable // inh. Fig. 3.75: the Arbitri construction construction Arbitri1 subcase of Arbitri constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /'arbitri/ // inh. meaning: RD // inh. constraints self.m.category ← Referee self.m.number ← morethanone // inh. 180 3. An ECG analysis of Italian caused-motion constructions The DeterminerPlusNP construction will then unify with Declarative's subj constituent, providing the information that the entity that fills the affected and profiledParticipant roles is morethanone members of the category Referee. The level of accessibility of gli arbitri is uniquely identifiable, for the use of a definite determiner denotes the possibility to uniquely identify the referent, as already mentioned. We can observe that the male plural definite article used in this case is not i, as in the previous examples, but gli. Gli is the plural form of the male singular definite article lo, which replaces il when the subsequent word begins with a vowel or with a particular consonant cluster (/s/ + consonant, /p/ + consonant, /gn/, etc.). ActiveCauseMotion's pp1 constituent will unify with the complex phrase dal ring sul tavolo, providing more information on the path routed by the patient participant, which corresponds to the syntactic subject of the sentence. We can analyze this complex construction splitting it up into two simple path phrases. The first of these phrases combines the complex path preposition dal and the noun ring. As usual, we begin providing an illustration of the construction corresponding to the nominal element. The word ring is a borrowing from English. It was assigned the male grammatical gender and, as usual for loanwords in Italian, it does not inflect for number. As in the case of portacenere, we distinguish two different constructions for each number and exploit the relevant one. The Ring1Sg construction is shown in Fig. 3.76 below, while the Ring1 construction, which denotes the relevant meaning of ring, can be observed in Fig. 3.77. We have already seen that the complex preposition dal unifies the path preposition da and the male singular definite article il, but in this case, unlike the sentence in (24) above, the preposition expresses the process opposite to that denoted by in rather than su, as shown by the Da2b construction observable in Fig. 3.78. Fig. 3.76: the Ring1Sg construction construction Ring1Sg subcase of Ring1, Singular constructional: NominalFeatureSet // inh. constituents self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /riŋ/ // inh. meaning: RD // inh. constraints self.m.category ← Ring // inh. self.m.natgender ← neuter // inh. self.m.number ← one Fig. 3.77: the Ring1 construction construction Ring1 subcase of Ring constructional: NominalFeatureSet // inh. constituents self.features.grammgender ← male // inh. form: WordForm // inh. constraints self.f.phon ← /riŋ/ // inh. meaning: RD // inh. constraints self.m.category ← Ring self.m.natgender ← neuter // inh. 181 Grounding Meaning in Everyday Experience in the World Fig. 3.78: the Da2b construction construction Da2b subcase of PathPreposition, Da form: WordForm // inh. constraints self.f.phon ← /da/ // inh. meaning: SPG // inh. evokes BoundedObject as bo constraints self.m.landmark ↔ bo.boundedobject self.m.profiledArea ↔ bo.interior self.m.source ↔ self.m.profiledArea self.m.goal ↔ bo.exterior The meaning of Da2b is constrained to be SPG and it evokes the BoundedObject schema already exploited in the representation of other path prepositions above. Nevertheless, this time the trajector of SPG had to go from the internal region of an object (in this case, a particular ring), represented by the interior role, to the region outside the object, represented by the exterior role. Then, the Da2b construction is combined with the Il construction in the ComplexPathPreposition. The phonological form of the complex preposition is, again, /dal/. The ComplexPathPreposition construction is then combined with the noun ring in the PathPP construction. The second path phrase of the cluster includes the complex path preposition sul and the noun tavolo (which seems to have just one meaning). The construction referring to the noun constituent of the preposition is shown in Fig. 3.79 below, while the preposition sul was already analyzed during the investigation of the sentence in (23) above. The ComplexPathPreposition and the Tavolo construction will then be combined in the PathPP construction. Fig. 3.79: the Tavolo construction construction Tavolo subcase of MaleSingularNoun constructional: NominalFeatureSet // inh. constituents self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'tavolo/ meaning: RD // inh. constraints self.m.category ← Table self.m.natgender ← neuter self.m.number ← one // inh. 182 3. An ECG analysis of Italian caused-motion constructions In order to specify the fact that the two path phrases describe just one motion event, which involves the same trajector (i.e. the mover role of the MotionAlongAPath schema) in the same moment, we will posit the existence of a ComplexPathPP construction, whose meaning is bound to a subcase of the SPG schema labeled ComplexSPG. The ComplexPathPP construction is illustrated in Fig. 3.81, while the ComplexSPG schema is shown in Fig. 3.82 below. ComplexPathPP is a subcase of the PathPP construction (Fig. 3.23 above). The constructional block of ComplexPathPP lists two constituents of the construction: pp1 and pp2, both bound to be PathPP. The ComplexPathPP construction allows us to clarify that the motion event described by the cluster of two PPs dal ring sul tavolo is only one, but with a complex structure. As a matter of fact, the dal ring phrase portrays the fact that the affected role of the CauseMotionAction schema, i.e. the referees, moves from the the interior of the ring to the outer region. On the other hand, the sul tavolo phrase specifies that the mover moves from the exterior of the table to its surface. In other words, the two phrases differ in their landmarks (and, consequently, their profiled areas): the dal ring phrase describes the event with reference to the ring, while the sul tavolo phrase describes it taking the table as a background. In order to illustrate this phenomenon, we proposed to bind the meaning of the ComplexPathPP construction to a subcase of the SPG schema which simply doubles its roles, labeled ComplexSPG. This solution allowed us to specify that the landmarks of the two phrases are different, but the motion event is the same. This was achieved binding the landmark and the profiledArea roles of the ComplexSPG schema to the landmark and the profiledArea roles of its pp1 constituent and the landmark2 and profiledArea2 roles of ComplexSPG to the landmark and the profiledArea roles of pp2. On the other hand, since the trajector and its source and goal are the same, the trajector, source, and goal roles of the ComplexPathPP construction were bound to those of both constituent constructions. Moreover, the trajector, source and goal roles of pp1 were bound to those of pp2. As a consequence, the meaning block of the ComplexPathPP construction specifies that the region external to the table from which the patient participant moves is the internal region of the ring, and that the exterior of the ring to which the patient participant moves is the surface of the table. Finally, we have to deal with the causative PP da Undertaker, which comprises a causative preposition, illustrated by the Da3 construction (see Fig. 3.67 above), and a proper noun, Undertaker, shown in Fig. 3.80 below. The two constructions are then combined in the CausativePP construction, which unifies with the pp2 constituent of the PassiveCauseMotion1 construction, supplying the information that its np constituent represents the filler the causer role. 183 Grounding Meaning in Everyday Experience in the World Fig. 3.80: the Undertaker construction construction Undertaker subcase of ProperNP form: WordForm // inh. constraints self.f.phon ← /'ʌndǝteikǝ/ meaning: RD // inh. constraints self.m.category ← Human self.m.number ← one self.m.accessibility ← inactive self.m.referent ← Undertaker Fig. 3.81: the ComplexPathPP construction construction ComplexPathPP subcase of PathPP constructional constituents pp1: PathPP pp2: PathPP form constraints pp1.f before pp2.f meaning: ComplexSPG constraints self.m ↔ pp1.m // inh. self.m ↔ pp2.m // inh. self.m.trajector ↔ pp1.m.trajector pp1.m.trajector2 ↔ pp2.m.trajector self.m.landmark ↔ pp1.m.landmark self.m.landmark2 ↔ pp2.m.landmark self.m.profiledArea ↔ pp1.m.profiledArea self.m.profiledArea2 ↔ pp2.m.profiledArea self.m.source ↔ pp1.m.source pp1.m.source2 ↔ pp2.m.source self.m.goal ↔ pp1.m.goal pp1.m.goal2 ↔ pp2.m.goal Fig. 3.82: the ComplexSPG schema schema ComplexSPG subcase of SPG roles trajector // inh. Trajector2 landmark // inh. landmark2 profiledArea // inh. profiledArea2 source // inh. source2 path // inh. path2 goal // inh. goal2 The category role of the meaning block of the Undertaker construction is simply 184 3. An ECG analysis of Italian caused-motion constructions bound to be Human, since a proper noun hardly gives more information, in particular in the case of a pseudonym. The fact that the referent of the Undertaker construction is human rather than a more generical animate is due to the fact that such a pseudonym exploits an English common noun which denotes a human profession. The accessibility role is awarded the value inactive because we do not know the discourse context and proper nouns can be used in the absence of any previous mention, but the level of accessibility of a proper noun like Undertaker will depend on the context (see the case of the noun Sara in the example analyzed in §2.3.2). The number role is assegned the filler one because the noun identifies a singular referent. The referent role is clearly assigned the value Undertaker. In the case of the Undertaker construction, there is no constructional block to show, since it does not present any features which cannot be considered as purely formal or semantic. The SemSpec produced by the unification of the constructions instantiated by this sentence will be similar to the one of the sentence analyzed above. The main difference will be represented by the presence of two landmarks for the same action, expressed in the path PP cluster dal ring sul tavolo. Besides that, another difference is represented by the fact that the profiledProcess role of the EventDescriptor schema, is filled by the meaning of ScaraventareMalePlPastParticiple, which maintains a basic meaning of CauseMotionAction, but its x-net is one of flinging. Again, this SemSpec will capture the conceptual core of our sentence, driving the enactment of an event in which a human agent nicknamed Undertaker performs a flinging action on more than one uniquely identifiable referees, obtaining their movement from a uniquely identifiable ring to a uniquely identifiable table. Once more, the event will be enacted from the perspective of the patient participant, the mover (i.e. the referees). The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agent (probably very high in this case), the purpose on which they performed the action, the way in which the action affected the patient participant etc. The co-occurrence of nouns such as arbitri and ring, the verb scaraventare, suggests a wrestling frame. Then, on the basis of their encyclopedic knowledge46 and the (discourse and situational) context, the language understander will be able to mentally reconstruct the event in more detail. The next sentence-level construction we are going to analyze, reported in (28), involves the passive conjugation of a compound verb, which will require the use of an additional device. (28) Sono stato gettato nella cella dall' agente Be: PRES.1SG be: PSTPART.MSG dash: PSTPART.MSG into-the.FSG cell.SG by-the.MSG officer.SG paonazzo. red-faced.MSG. “I have been dashed into the cell by the red-faced officer” This sentence is analogous to the one instantiated in (26), but its verb is in the passive present perfect tense. The most important feature which distinguishes this construction is the addition of a nonfinite auxiliary to the finite one. This means that, while we can 46 The language-understander's level of knowledge of this kind of entertainment will make the difference in their mental reconstruction of the event. In particular, if they are familiar with the world of professional wrestling, then they will know that The Undertaker, pseudonym of Mark Calloway, is one of the most famous American wrestlers ever; as a result, they will be able to precisely identify the agent participant. Otherwise, they will conceptualize the referent of this name much more generically. Note that Italians use to delete the article from English nicknames, so he is normally called simply Undertaker. 185 Grounding Meaning in Everyday Experience in the World still exploit the PassiveCauseMotion construction at the phrasal level and the Declarative construction at the clausal level, the FiniteAuxPlusVP construction will not be sufficient to connect the auxiliaries and the A-S construction. As a result, we need to posit another construction, which occupies an intermediate position between FiniteAuxPlusVP and PassiveCauseMotion. Such a construction is labeled NonFiniteAuxPlusVP and is shown in Fig. 3.83 below. The NonFiniteAuxPlusVP construction allows us to insert a non-finite auxiliary which unifies with the following verb phrase. Then, the FiniteAuxPlusVP construction (already exploited and shown) will combine the finite auxiliary with the following VP (i.e. the NonFiniteAuxPlusVP construction), finally unifying with the fin constituent of the Declarative construction. Now, we can focus on the auxiliaries involved in the sentence under consideration. The passive present perfect involves the use of the simple present form of the auxiliary essere and the past participle form of the same auxiliary verb. In this sentence we can find sono, the first-person singular form of the present tense of the verb essere, and stato, the male singular past participle form this verb. Therefore, the constructions corresponding to these two verb forms of essere are labeled EssereAuxSimplePresent1sg and EssereAuxMaleSgPastParticiple and shown in Fig. 3.84 and Fig. 3.85 below, respectively. Fig. 3.83: the NonFiniteAuxPlusVP construction construction NonFiniteAuxPlusVP subcase of VP constructional constituents aux: Aux vp: VP constraints self.features.verbform ← nonfinite self.features ↔ aux.features aux.features.expectedverbform ↔ vp.features.verbform form constraints aux.f before vp.f meaning evokes EventDescriptor as ed // inh. constraints self.m ↔ vp.m ed ↔ vp.ed self.m ↔ ed.eventType // inh. 186 3. An ECG analysis of Italian caused-motion constructions Fig. 3.84: the EssereAuxSimplePresent1sg construction construction EssereAuxSimplePresent1sg subcase of EssereAux, SimplePresentTense, Singular, 1stPerson constructional: AuxFeatureSet // inh. constraints self.features.verbform ← simplepresent // inh. self.features.number ← singular // inh. self.features.person ← 1 // inh. form: WordForm // inh. constraints self.f.phon ← /'sono/ meaning: Function // inh. constraints self.m.function ← grounding // inh. Fig. 3.85: the EssereAuxMaleSgPastParticiple construction construction EssereAuxMaleSgPastParticiple subcase of EssereAux, PastParticiple, Singular, Male constructional: AuxFeatureSet // inh. constituents self.features.verbform ← pastparticiple // inh. self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'stato/ meaning: Function // inh. evokes RD as rd constraints self.m.function ← grounding // inh. rd.number ←one The constraints in the constructional and the meaning blocks of the EssereAuxMaleSgPastParticiple construction anticipate the grammatical information on the referent of the verb and a part of its semantics (namely, its number). As a consequence, the past participle of the lexical verb just adds the (of course, fundamental) specification regarding the meaning of the verb, bound to to the CauseMotionAction schema, adding information on the correct x-net, which is one of dashing (that is quite similar to the x-nets of the verbs found in the two previous 187 Grounding Meaning in Everyday Experience in the World sentences), while its form is constrained to be a word and is assigned the correct phonological form. The Gettare1MaleSgPastParticiple construction is represented in Fig. 3.86 below. Fig. 3.86: the Gettare1MaleSgPastParticiple construction construction Gettare1MaleSgPastParticiple subcase of Gettare1, PastParticiple, Singular, Male constructional: VerbFeatureSet // inh. constituents self.features.verbform ← pastparticiple // inh. self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /ʤet':ato/ meaning: CauseMotionAction evokes RD as rd constraints self.m.x-net ← @dash // inh. rd.number ← one Now that we have established the general constraints on the fillers of the participant roles, we can focus our attention to the NP and the PPs present in the clause, which provide more specific information on the fillers of these roles for the event described in the sentence in (28). We can find a null subject and two PPs: a spatial phrase (nella cella) and a causative one (dall'agente paonazzo). Because the finite auxiliary verb is conjugated in the first singular person, the null subject construction, illustrated in Fig. 3.87, is labeled DropIo, and it is considered a subcase of the Io construction, shown in Fig. 3.88. DropIo will unify with the referent of Declarative's subj constituent, providing the information that the entity that fills the affected and profiledParticipant roles is one human referent represented by the current speaker. The level of accessibility of a first-person singular pronoun is active, since pronouns are generally used to denote active referents and, moreover, the current speaker is always active in any communicative situation. 188 3. An ECG analysis of Italian caused-motion constructions Fig. 3.87: the DropIo construction. construction DropIo subcase of Io constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh. self.features.number ← singular // inh. self.features.person ← 1 // inh. form: WordForm // inh. constraints ignore: self.f.phon ← /io/ meaning: RD // inh. self.m.category ← Human // inh. self.m.number ← one // inh. self.m.accessibility ← active // inh. self.m.referent ← currentspeaker // inh. Fig. 3.88: the Io construction construction Io subcase of SingularSubjPersPron, 1stperson constructional: NominalFeatureSet // inh. constraints self.features.case ← nominative // inh. self.features.number ← singular // inh. self.features.person ← 1 // inh. form: WordForm // inh. constraints self.f.phon ← /io/ meaning: RD // inh. self.m.category ← Human self.m.number ← one // inh. self.m.accessibility ← active // inh. self.m.referent ← currentspeaker We can now turn to the path PP nella cella, which is made up of a path preposition and an NP, and it is very similar to that observed in (25). The Cella1 construction is shown in Fig. 3.89 below. The complex preposition nella is composed of the simple preposition in, and the determiner la. The In2 and the La constructions (both already shown, in Fig. 3.57 and Fig. 2.29 respectively) will be combined together in the ComplexPathPreposition construction. As usual, we precise that the form of the complex preposition is not a simple agglutination of in + la (which would presumibly be */'inla/), but rather the form /'nel:a/. The ComplexPathPreposition construction and the Cella1 construction are finally combined together in the PathPP construction, which provides information about the kind of motion portrayed by the sentence. Then, we get to the causative PP dall'agente paonazzo. This PP is formed by the complex causative preposition dall' and an NP constituted by a noun and an adjective. The noun agente can be either male or female. As in the case of the noun portacenere encountered above, we posit the existence of two subcases (one male and one female) of the Agente1 construction (subcase of the more general construction Agente), and make use of the relevant one. The constructions corresponding to the noun agente and to the adjective paonazzo are shown in Fig. 3.90 and Fig. 3.91 below. 189 Grounding Meaning in Everyday Experience in the World Fig. 3.89: the Cella1 construction construction Cella1 subcase of Cella constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'ʧel:a/ // inh. meaning: RD // inh. constraints self.m.category ← Prison-Cell self.m.natgender ← neuter self.m.number ← one // inh. Fig. 3.90: the Agente1M construction construction Agente1M subcase of Agente1, Male constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /a'ʤente/ // inh. meaning: RD // inh. constraints self.m.category ← Police-Officer self.m.natgender ← male self.m.number ← one // inh. Fig. 3.91: the Paonazzo1 construction construction Paonazzo1 subcase of Paonazzo constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /pao'nat:so/ // inh. meaning: ScalarModifier // inh. evokes RD as rd // inh. constraints self.m.scalarvalue ← red-faced self.m.scale ← complexion self.m.modifiedEntity ↔ rd // inh. rd. category ← human rd.natgender ← male rd.number ← one // inh. Like in the sentence in (24) taken into consideration above, we can observe a noun modified by a scalar modifier. This time, however, the relevant scale is not age but complexion, which can also be seen as a matter of degree, even though there is (to our 190 3. An ECG analysis of Italian caused-motion constructions knowledge) no conventional way to measure it. In this example, the modifier is also important because it is the element which lets us know the natural gender of the referent. The Agente1M and the Paonazzo constructions are then combined together in the ModifiedNP2 construction shown in Fig. 3.92, which establishes a formal constraint on the order of the constituents (i.e. the noun precedes the modifier), which is the opposite of the pattern captured by the ModifiedNP1 construction (shown in Fig. 3.45 above) while the semantic constraints established by the two constructions are the same. Fig. 3.92: the ModifiedNP2 construction construction ModifiedNP2 subcase of ModifiedNP constructional: NominalFeatureSet constituents mod: Modifier // inh. np: NP // inh. form constraints np.f before mod.f meaning constraints self.m ↔ np.m // inh. self.m ↔ mod.m // inh. With regard to the complex preposition dall', it is made up of the causative preposition da and the abbreviation of the definite article lo which, when it precedes an NP starting with a vowel, is shortened to l'. The construction which corresponds to this preposition is Da3 (shown in Fig. 3.67 above), while construction relative to the article is shown in Fig. 3.93 below. L'Detm is a subcase of the LoDet construction (not shown)47, with the ignore command in the form block allowing us to suppress the phonological form of LoDet and replace it with the relevant phonological structure of L'Detm48. The Da3 construction and the L'Detm construction are then combined in the ComplexCausativePreposition construction (Fig. 3.94). We specify that the fusion of the preposition and the determiner is a bit different from the simple agglutination of da + l' (i.e. */dal/), with the final consonant being a bit longer, /dal:/. Then, the ComplexCausativePreposition construction and the ModifiedNP2 construction are unified by the CausativePP construction (Fig. 3.70 above). 47 This construction is named LoDet instead of simply Lo in order to clarify that it refers to the male singular definite article, because the same form is also shared by a male singular clitic pronoun. 48 This construction is labeled L'1m, instead of simply L'1, because the female singular article la takes the same form when it is shortened before a word which begins with a vowel. 191 Grounding Meaning in Everyday Experience in the World Fig. 3.93: the L'Detm construction construction L'Detm subcase of LoDet constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← male // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints ignore: self.f.phon ← /lo/ self.f.phon ← /l/ meaning evokes RD as rd // inh. constraints rd.number ← one // inh. rd.accessibility ← uniquely-identifiable // inh. Fig. 3.94: the ComplexCausativePreposition construction construction ComplexCausativePreposition subcase of CausativePreposition constructional constituents prep: CausativePreposition s: Determiner form: WordForm // inh. constraints prep.f fuses with s.f meaning: Causation constraints self.m ↔ prep.m self.m ↔ s.m The SemSpec produced by the unification of the constructions instantiated by this sentence will once more be similar to that corresponding to the sentence analyzed above. The main difference will be represented by the fact that the profiledProcess role of the EventDescriptor schema, is filled by the meaning of Gettare1MalePlPastParticiple, which maintains a basic meaning of CauseMotionAction, but its x-net is one of dashing. This SemSpec will drive the enactment of an event in which a red-faced male police officer performs a dashing 192 3. An ECG analysis of Italian caused-motion constructions action on the current speaker (who is male too), obtaining his movement from the external into the internal region a uniquely identifiable cell. The event, as in the previous cases, will be enacted from the perspective of the mover, who in the present sentence is the current speaker. The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agent (probably high in this case), the purpose on which they performed the action, the way in which the action affected the patient participant, and so forth. The co-occurrence of nouns such as cella and agente suggests that the event takes place in a prison. Then, as usual, on the basis of their world knowledge and the (discourse and situational) context, the language understander will manage to achieve a better mental reconstruction of the event. The last example of passive caused-motion construction we are going to take into consideration is analogous to the one shown in (28), but at the same time it may be considered as a counterpart of the active clause illustrated in (25), being slightly less prototypical than the three just analyzed above. Consider the sentence in (29) below: (29) Eravate state tirate nella stanza dalle vostre Be:IMPF.2PL be:PSTPART.FPL pull:PSTPART.FPL into-the.FSG room.SG by-the.FPL your.2PL.FPL amiche ubriache friend.FPL drunk.FPL “You had been pulled into the room by your drunk (female) friends” Like the example in (28), involving the passive conjugation of a compound verb, this sentence requires the use of both the FiniteAuxPlusVP and the NonFiniteAuxPlusVP constructions to connect its auxiliary verbs with its A-S construction. As in the case of the example in (25), the meaning of the verb involved in this sentence-level construction does not exactly correspond to that of the A-S construction. Indeed, it unifies with the ForceApplication schema, the first process of CauseMotionAction, rather than with the CauseMotionAction schema as a whole. As a matter of fact, extraposing the verb tirare from the caused-motion construction it is involved in, we can notice that, exactly as was the case with the verb spingere, tirare denotes an action in which a causer participant exerts force on a patient participant in order to move it, but it does not entail that they succeed. This similiarity between spingere and tirare does not occur by chance, since tirare denotes an exertion of force very similar to that described by spingere, but which goes in the opposite direction (i.e. toward the agent). As a result, both spingere and tirare can also be used to denote an action in which the causer exerts force on the patient, but fails to move it. Again, the same condition does not hold with any of the verbs involved in the three sentences taken into consideration above. Therefore, we shall recur to the use of a subcase of the PassiveCauseMotion construction, labeled PassiveCauseMotion2, which explicitly captures this peculiarity in its meaning block. The PassiveCauseMotion2 construction is shown in Fig. 3.95 below. The sentence under consideration involves a verb in the passive recent pluperfect tense, which is formed combining the imperfect tense and the past participle of the auxiliary verb essere with the past participle of the lexical verb. Confining ourselves to the auxiliary verbs, we can observe the second-person plural form of the imperfect of essere and the female plural past participle form of the same verb, labeled EssereAuxImperfect2pl and EssereAuxFemPlPastParticiple and shown in Fig. 3.96 and Fig. 3.97, respectively. 193 Grounding Meaning in Everyday Experience in the World Fig. 3.95: the PassiveCauseMotion2 construction construction PassiveCauseMotion2 subcase of PassiveCauseMotion constructional constituents v: PassiveVerb pp1: PathPP optional pp2: CausativePP form constraints v.f before pp1.f pp1.f before pp2.f meaning: CauseMotionAction // inh. evokes EventDescriptor as ed // inh. constraints v.m ↔ self.m.process1 self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m.affected ↔ ed.profiledParticipant // inh. self.m.causer ↔ pp2.m.np // inh. ed.profiledParticipant ↔ topic self.m.process2 ↔ pp1.m.spg Fig. 3.96: the EssereAuxImperfect2pl construction construction EssereAuxImperfect2pl subcase of EssereAuxImperfectTense, Plural, 2ndPerson constructional: AuxFeatureSet // inh. constraints self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← imperfect // inh. self.features.number ← plural // inh. self.features.person ← 2 // inh. form: WordForm // inh. constraints self.f.phon ← /era'vate/ meaning: Function // inh. constraints self.m.function ← grounding // inh. 194 3. An ECG analysis of Italian caused-motion constructions Fig. 3.97: the EssereFemPlPastParticiple construction construction EssereAuxFemPlPastParticiple subcase of EssereAux, PastParticiple, Plural, Female constructional: AuxFeatureSet // inh. constituents self.features.expectedverbform ← nonfinite // inh. self.features.verbform ← pastparticiple // inh. self.features.grammgender ← female // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /'state/ meaning: Function // inh. evokes RD as rd constraints self.m.function ← grounding // inh. rd.number ←morethanone As in the case of the preceding sentence, the constraints in the constructional and the meaning blocks of the EssereAuxMaleSgPastParticiple construction anticipate the grammatical information on the referent of the verb and a part of its semantics (namely, its number). As a consequence, the past participle of the lexical verb just clarifies the meaning of the verb, bound to the ForceApplication schema, also telling us that the correct x-net is one of pulling, while its form is given the corresponding phonological form. The Tirare1FemPlPastParticiple construction is represented in Fig. 3.98 below. Fig. 3.98: the Tirare1FemPlPastParticiple construction construction Tirare1FemPlPastParticiple subcase of Tirare1, PastParticiple, Plural, Female constructional: VerbFeatureSet // inh. constituents self.features.verbform ← pastparticiple // inh. self.features.grammgender ← female // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /ti'rate/ meaning: ForceApplication // inh. evokes RD as rd constraints self.m.x-net ← @pull // inh. rd.number ← morethanone 195 Grounding Meaning in Everyday Experience in the World The most important consequence of the binding between the lexical verb and the ForceApplication schema is the unification between the profiledProcess role of the EventDescriptor schema with the first process of CauseMotionAction rather than with the schema as a whole (the profiledParticipant remains the same). Apart from that, the sentence under consideration is not dissimilar from the other three analyzed in this section. The meaning of the construction is still bound to the CauseMotionAction schema and, at the clausal level, we still make use of the Declarative construction. We can now turn to the NP and the PPs which are part of the sentence under consideration. We can observe a null subject and, as usual, a path PP (nella stanza) and a causative PP (dalle vostre amiche ubriache). The construction which corresponds to the null subject is labeled DropVoi (Fig. 3.99) and it is a subcase of the Voi construction (Fig. 3.109). Fig. 3.99: the DropVoi construction construction DropVoi subcase of Voi constructional: NominalFeatureSet // inh. constraints self.features.number ← plural // inh. self.features.person ← 2 // inh. form: WordForm // inh. constraints ignore: self.f.phon ← /voi/ meaning: RD // inh. constraints self.m.category ← Animate // inh. self.m.number ← morethanone // inh. self.m.accessibility ← active // inh. self.m.referent ← currentaddresseeandotheranimate // inh. Fig. 3.100: the Voi construction construction Voi subcase of PluralPersPron, 2ndPerson constructional: NominalFeatureSet // inh. constraints self.features.number ← plural // inh. self.features.person ← 2 // inh. form: WordForm // inh. constraints self.f.phon ← /voi/ meaning: RD // inh. constraints self.m.category ← Animate self.m.number ← morethanone // inh. 196 self.m.accessibility ← active // inh. self.m.referent ← currentaddresseeandotheranimate 3. An ECG analysis of Italian caused-motion constructions DropVoi will unify with the referent of Declarative's subj constituent, providing the information that the entity that fills the affected and profiledParticipant roles is morethanone, animate, and the referent is represented by the current addressee plus some other animate being. The level of accessibility of a second-person plural pronoun is active, since the current addressee must be active in any discourse context. The path PP is made up of the complex path preposition nella (already illustrated in the analysis of the previous sentence) and the NP stanza (Fig. 3.101 below). The preposition and the NP are then combined in the PathPP construction. Fig. 3.101: the Stanza1 construction construction Stanza1 subcase of Stanza constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← singular // inh. form: WordForm // inh. constraints self.f.phon ← /'stanʦa/ // inh. meaning: RD // inh. constraints self.m.category ← Room self.m.natgender ← neuter // inh. self.m.number ← one // inh. The causative PP is composed by the complex causative preposition dalle and the NP vostre amiche ubriache, which involves a noun, a scalar modifier and a possessive49. We can begin the analysis of the NP considering the noun amiche and the adjective ubriache, whose corresponding constructions are shown in Fig. 3.102 and Fig. 3.103 below. The amiche construction supplies some information on the nature of the referent of the causative PP (i.e. the causer role of the CauseMotionAction schema). The adjective ubriache, which modifies the noun, is considered as a scalar modifier like giovane and paonazzo: indeed, though drunkness is probably less conventionally thought of as a scale than age but can be considered as a matter of degree. Moreover, unlike the case of complexion, instruments to measure drunkenness do exist. The Amiche and the Ubriache constructions are then combined together in the ModifiedNP2 construction. Then, we have to deal with the possessive vostre, which modifies the NP contained in the ModifiedNP2 construction. 49 While possessives are normally considered determiners, in Italian their status is more similar to that of modifiers. Indeed, determiners are generally seen as mutually exclusive, but in Italian a possessive is normally (though not always) preceeded by the article (see §3.3). Nevertheless, we are going to treat possessives as particular rather than regular modifiers, as will be seen in the next pages. 197 Grounding Meaning in Everyday Experience in the World Fig. 3.102: the Amiche1 construction construction Amiche1 subcase of Amiche constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /a'mike/ // inh. meaning: RD // inh. constraints self.m.category ← Friends self.m.natgender ← female // inh. self.m.number ← morethanone // inh. Fig. 3.103: the Ubriache1 construction construction Ubriache1 subcase of Ubriache constructional: NominalFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /u'brjake/ // inh. meaning: ScalarModifier // inh. evokes RD as rd // inh. constraints self.m.scalarvalue ← drunk self.m.scale ← drunkness self.modifiedEntity ↔ rd // inh. rd.category ← Human // inh. rd.natgender ← female // inh. rd.number ← morethanone // inh. Vostre is more complex than the other modifiers we have seen so far, for possessives in Romance languages are made up of a stem50 (in this case, vostr-) providing some information on the possessor, and a inflectional morpheme (in this case, -e) supplying some information on the possessed entity. For clarity's sake, we had better represent these two parts as separate constructions, and then posit a construction which unifies them. The Vostr- construction, shown in Fig. 3.106 below, is a subcase of the Possessor construction, shown in Fig. 3.104. The meaning of the Possessor constructions is bound to the Possession schema, represented in Fig. 3.105. Fig. 3.104: the Possessor construction construction Possessor subcase of Stem constructional: AgreementFeatureSet // inh. form: StemForm meaning: Possession evokes RD as possessor constraints possessor ↔ self.m.possessor Fig. 3.105: the Possession schema schema Possession roles possessor possessed 50 Note that the term stem here is used in the sense defined in Langacker's CG, rather than its traditional sense (see Langacker 1987). 198 3. An ECG analysis of Italian caused-motion constructions Fig. 3.106: the Vostr- construction construction Vostrsubcase of Possessor, Plural, 2ndPerson constructional: AgreementFeatureSet // inh. constraints self.features.number ← plural // inh. self.features.person ← 2 // inh. form: StemForm // inh. constraints self.f.phon ← /'vostr/ meaning: Possession // inh. evokes RD as possessor // inh. constraints possessor ↔ self.m.possessor // inh. possessor.category ← Animate possessor.number ← morethanone possessor.accessibility ← active possessor.referent ← currentaddresseeandotheranimate The Vostr- construction lets us know that the possessor is more than one, animate, active, and includes the current addressee, together with at least one animate entity. In Fig. 3.107, you can find a representation of the -e construction, which is a subcase of the Possessed construction, shown in Fig. 3.108. Of course, the meaning of Possessed is bound to the Possession schema as well. Fig. 3.107: the -e construction construction -e subcase of Possessed, Female, Plural constructional: AgreementFeatureSet // inh. constraints self.features.grammgender ← female // inh. self.features.number ← plural // inh. form: SuffixForm // inh. constraints self.f.phon ← /e/ meaning: Possession // inh. evokes RD as possessed // inh. constraints possessed ↔ self.m.possessed // inh. possessed.m.number ← morethanone Fig. 3.108: the Possessed construction construction Possessed subcase of Inflection constructional: AgreementlFeatureSet // inh. form: SuffixForm meaning: Possession evokes RD as possessed constraints possessed ↔ self.m.possessed 199 Grounding Meaning in Everyday Experience in the World The two morphemes are then combined together in the Possessive construction, shown in Fig. 3.109 below. Fig. 3.109: the Possessive construction construction Possessive subcase of Word, HasAgreementFeatures constructional: AgreementFeatureSet // inh. constituents possessor: Possessor possessed: Possessed constraints self.features ↔ possessor.features self.features ↔ possessed.features form: WordForm // inh. constraints possessor.f fuses with possessed.f meaning: Possession constraints self.m ↔ possessor.m self.m ↔ possessed.m The -e construction adds the information that the possessed entity is plural51. The possessive construction is then combined with the ModifiedNP2 construction in the PossessivePlusNP construction, shown in Fig. 3.110 below. The preposition dalle represents the unification of the DaCause construction and the Le construction, combined together by the ComplexCausativePreposition construction. Once again, we have to specify that the fusion of the preposition and the determiner is different from a simple agglutination of da + le (i.e. */'dale/), but rather /'dal:e/. Finally, the ComplexCausativePreposition construction and the PossessivePlusNP1 construction are unified by the CausativePP construction (Fig. 3.70 above). 51As the reader will have noticed, the -e construction overlaps with the inflectional morpheme which is exploited by Italian speakers to create the female plural form of most (yet not all) adjectives. In fact, the same mechanism of composition illustrated for the Possessive construction may also be applied to the majority of Italian adjectives (and to many nouns, as well). However, in the present study we prefer to exploit it just when its explicitation proved to be necessary, i.e. with lexical constructions which provide information on two distinct referents. 200 3. An ECG analysis of Italian caused-motion constructions Fig. 3.110: the PossessivePlusNP1 construction construction PossessivePlusNP1 subcase of PossessivePlusNP constructional constituents poss: Possessive // inh. np: np // inh. form constraints poss.f before np.f meaning constraints self.m ↔ poss.m // inh. self.m ↔ np.m // inh. The SemSpec produced by the unification of the constructions instantiated in this sentence will be just a little bit different from the one of the sentences analyzed above. The main difference will be represented by the profiledProcess role of the EventDescriptor schema, filled by the meaning of TirareFemPlPastParticiple, which unifies with ForceApplication, and its x-net is one of pulling. This is a very important point, since it highlights that the enactment process will be especially focused on ForceApplication. Also, the verb specifies the setting time of the event: in Italian (as in other Romance languages), the recent pluperfect is normally used to achieve one of the following two goals: to express an action in the past that happened before another past action expressed by the past participle or the imperfect, or in an independent clause when the later action is implied but not stated (see Tognozzi and Cavatorta 2009: 6970). Again, this SemSpec will capture the conceptual core of our sentence, driving the enactment of an event in which at least two drunk female friends of the current addressee (and at least one more female animate being)'s perform a pulling action on the current addressee and at least another female animate entity, obtaining their movement into a uniquely identifiable room. Once more, the event will be enacted from the perspective of the mover, i.e. the current addressee and somebody else. The process of enactment of this event will produce more inferences, about the amount of energy supplied by the agent, the purpose on which they performed the action, the kind of room the mover entered into etc. The presence of the words amiche, ubriache, stanza will hint that probably the action during a social event (e.g. a party) in a domestic environment. Then, as usual, on the basis of their encyclopedic knowledge and the (discourse and situational) context, the language understander will be able to mentally reconstruct the scene much more in detail. In the present subsection, we have briefly analyzed four passive caused-motion constructions, showing that the CauseMotionAction schema and the subcases of the PassiveCauseMotion A-S construction allows us to define a certain degree of generalization over distinct verbs of caused-motion, similarly to what was observed for active sentences in the previous subsection. As in the previous subsection, we can assert that the CauseMotionAction schema captures the fact that all the sentences examined 201 Grounding Meaning in Everyday Experience in the World above describe events in which an agent makes a patient move in a certain direction by volitionally exerting force on it. At the same time, the PassiveCauseMotion1 construction captures the fact that all these events are described by the patient's perspective. More specifically, we analyzed sentences involving the verbs scagliare (“to hurl”), scaraventare (“to fling”), gettare (“to dash”), tirare (“to pull”): the first three of these four verbs describe quite similar processes, but the sentences they appear in suggest quite different scenarios. In the case of the sentence involving the verb tirare, however, we noticed that such a verb presents the same semantic peculiarity shown by spingere: the meaning of the verb fuses with the ForceApplication process rather than the CauseMotionAction schema as a whole; therefore, we preferred to analyze such a sentence using another subcase of the PassiveCauseMotion construction, labeled PassiveCauseMotion2, able to capture this slight difference. The path phrases used in the sentences analyzed in this subsection do not instantiate kinds of motion distinct from those illustrated in the active clauses considered in §3.2.1. On the other hand, we introduced a causative phrase which captured the generalizations over the agent participants quite easily, since in Italian there is only one causative preposition. While the first sentence we faced was fairly simple, the following three were a bit trickier. In one of our sentences we described the motion from the internal region of a place to the uppermost layer of another, positing the existence of two different landmarks. In the last two sentences, we dealt with the passive conjugation of two compound verbs adding a construction able to link the non finite auxiliary to the A-S construction. Moreover, in the last of our sentences we introduced the uneasy issue of Italian possessives. Before discussing the issues raised by the analysis of the sentence-level construction provided above, in §3.2.3 we want to briefly outline a couple of extensions (along the lines of those offered for English in Feldman et al. 2009), regarding sentences which can be better approached using constructions alternative or complementary to Declarative. 3.2.3. Two examples of possible extensions In the previous two sections, we dealt with sentences which could all be handled by positing the same clause-level construction, namely Declarative. In this section we are going to briefly address some cases whose analysis requires the use of either an alternative construction at the clause-level or an additional construction at a higherlevel. For the sake of both simplicity and clarity, we shall make use of modified versions of one of the sentences analyzed above, namely the one represented in (22), repeated in (30) below: (30) I bambini lanceranno i sassi al bersaglio. The.MPL child.PL throw:FUT.3PL the.MPL stone.PL to-the.MSG target.SG “The children will throw the stones to the target” Let us begin taking into the consideration the simple wh-question in (31): (31) Quali bambini lanceranno i sassi al bersaglio ? Which.PL child.MPL throw:FUT.3PL the.MPL stone.PL to-the.MSG target.SG “Which children will throw the stones the the target?” 202 3. An ECG analysis of Italian caused-motion constructions At the lexical level, the only difference between the declarative clause and the whquestion is the presence in the latter of the specifier quali, which replaces the definite article i. The construction corresponding to quali can be found in Fig. 3.113 below. The Quali construction is a subcase of the PluralWhSpecifier construction (Fig. 3.111), in turn a subcase of WhSpecifier (Fig. 3.112). Fig. 3.111: the WhSpecifier construction construction WhSpecifier subcase of NPSpecifier constructional: NominalFeatureSet // inh. form: WordForm // inh. meaning evokes RD as rd // inh. constraints rd.accessibility ← unidentifiable Fig. 3.112: the PluralWhSpecifier construction construction PluralWhSpecifier subcase of WhSpecifier, Plural constructional: NominalFeatureSet // inh. constraints self.features.number ← plural // inh. form: WordForm // inh. meaning evokes RD as RD // inh. constraints rd.number ↔ morethanone rd.accessibility ← unidentifiable // inh. Fig. 3.113: the Quali construction construction Quali subcase of PluralWhSpecifier constructional: NominalFeatureSet // inh. constraints self.features.number ← plural // inh. form: WordForm // inh. constraints self.f.phon ← /'kwali/ meaning evokes RD as rd // inh. constraints rd.number ← morethanone // inh. rd.accessibility ← unidentifiable // inh. The Quali construction is then combined with the Bambini1 construction (Fig. 3.11 above) in the WhSpecifierPlusNP construction shown in Fig. 3.114 below, specifying that the level of accessibility of the referent of the NP is unidentifiable, since whspecifiers are used to express the fact that the language user is not able to identify the referent of the phrase. 203 Grounding Meaning in Everyday Experience in the World Fig. 3.114: the WhSpecifierPlusNP construction construction WhSpecifierPlusNP subcase of NPSpecifierPlusNP constructional: NominalFeatureSet // inh. constituents npspec: WhSpecifier np: NP // inh. constraints self.features ↔ np.features // inh. self.features ↔ npspec.features // inh. form constraints npspec.f before np.f // inh. meaning: RD constraints self.m ↔ np.m // inh. self.m ↔ npspec.m // inh. While at the phrase-level we can still exploit the A-S construction used in the declarative clause in (30) (i.e. ActiveCauseMotion1), at the clause-level Determiner should be replaced by a more adequate construction. In this case, the SubjWhSpecQuestion construction (shown in Fig. 3.115 below) will carry out the duty satisfactorily. Fig. 3.115: the SubjWh-Question construction construction SubjWhSpec-Question subcase of S-With-Subj constructional constituents subj: WhSpecifierPlusNP fin: VP constraints self.features.mood ← wh-question form constraints subj.f before fin.f meaning: EventDescriptor constraints self.m. ↔ fin.ed subj.m ↔ fin.ed.profiledParticipant // inh. 204 3. An ECG analysis of Italian caused-motion constructions The SubjWhSpec-Question construction constrains its subj constituent to be a WhSpecifierPlusNP phrase, and its mood role is filled by the value wh-question. The slight modifications just illustrated represent all we need to handle the differences between the wh-question in (31) and the declarative sentence in (30). Next, we shall illustrate the modifications necessary to carry out a correct analysis of the wh-question in (32) below, which is less simple than the previous one, since we have to deal with a fronted object: (32) Quali sassi i bambini lanceranno al bersaglio ? Which.PL stone.PL the.MPL child.PL throw:FUT.3PL to-the.MSG target.SG “Which stones will the children throw to the target?” At the lexical level, this example does not present any problems. The only difference between this sentence and the previous one has to do with the fact that the WhSpecifierPlusNP construction (Fig. 3.114) will combine the Quali construction (Fig. 3.115) with the Sassi1 construction (Fig. 3.13) rather than with the Bambini1 construction. On the other hand, we need to introduce some changes at the phrasal and clausal levels. With regard to the former, we shall posit a subcase of the ActiveCauseMotion constructions, labeled ActiveCauseMotionX, which just comprises two constituents: a verb and a path phrase. Such a construction is shown in Fig. 3.116 below. Fig. 3.116: the ActiveCauseMotionX construction construction ActiveCauseMotionX subcase of ActiveCauseMotion constructional constituents ignore: np: NP v: Verb // inh. pp: PathPP // inh. form constraints v.f before pp.f meaning: CauseMotionAction // inh. evokes EventDescriptor as ed // inh. constraints self.m.affected ↔ ed.topic self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m ↔ v.m self.m.causer ↔ ed.profiledParticipant // inh. ed.profiledParticipant ↔ ed.topic self.m.process2 ↔ pp.m.spg 205 Grounding Meaning in Everyday Experience in the World The use of the ignore command in the constructional block allows us to posit an active caused-motion construction made up of just a verb and a path PP, while its use in the meaning block allows us to ignore the constraint which binds the affected role of the CauseMotionAction schema to the np constituent of the A-S construction (which in this case is absent) and does not prevent us from positing an identification constraint between this role and the topic role of the EventDescriptor schema. This constraint denotes the fact that, while the profiled participant of event is the causer role of CauseMotionAction, the topic of the larger chunk is the affected role. We also need to change the clause-level construction, replacing it with one comprising three different constituents: along with the subject and the finite VP, this time we also have to handle a fronted direct object. We shall make use of the FrontedObjWh-Question construction shown in Fig. 3.117 below. Fig. 3.117: the FrontedObjWh-Question construction construction FrontedObjWh-Question subcase of S-With-Subj constructional constituents subj: NP // inh. fin: VP obj: WhSpecifierPlusNP constraints self.features.mood ← frontedobjwh-question form constraints obj.f before subj.f subj.f before fin.f meaning: EventDescriptor constraints self.m. ↔ fin.ed obj.m ↔ fin.ed.topic subj.m ↔ fin.ed.profiledParticipant // inh. The constructional block of FrontedObjWhQuestion constrains its subj constituent to be an NP, its fin constituent to be a VP and its obj constituent to be WhSpecifierPlusNP. The mood role of the construction filled by the wh-question value. Form constraints specify that the obj constituent precedes the subj constituent, and that the fin constituent comes last. The meaning of a construction is bound to an EventDescriptor schema, which is constrained to be the same as that of the ActiveCauseMotionX construction. Moreover, meaning constraints identify the obj constituent with the topic role of the EventDescriptor schema, so with the affected role of the CauseMotionAction schema, while the subj constituent is identified with the topic role of EventDescriptor and, consequently, with the causer role of 206 3. An ECG analysis of Italian caused-motion constructions CauseMotionAction. Exploiting the ActiveCauseMotionX and the FrontedObjWhQuestion, we are able to handle the wh-question in (32) quite comfortably. Next, we are going to consider the phenomenon of dislocations, beginning with the left dislocation in the sentence in (33) below: (33) I sassi i bambini li lanceranno al bersaglio. the.MPL stone.PL the.MPL child.PL CL.3MPL throw:FUT.3PL to-the.MSG target.SG. “The stones, the children will throw them to the target” At the lexical level, the only “innovation” is the introduction of the third-person male plural clitic personal pronoun li in the direct object position, represented in the LiClitic construction in Fig. 3.118 below. Fig. 3.118: the LiClitic construction construction LiClitic subcase of Male3rdPersonPluralCliticPersPron constructional: NominalFeatureSet // inh. constraints self.features.case ← accusative // inh. self.features.grammgender ← male // inh. self.features.number ← plural // inh. self.features.person ← 3 // inh. form: WordForm // inh. constraints self.f.phon ← /li/ meaning: RD evokes EventDescriptor as ed constraints self.m.referent ↔ ed.topic self.m.number ← morethanone // inh. self.m.accessibility ← active // inh. The Li construction provides the information that the referent is more than one and its status of accessibility is active and its referent is bound to be the topic of EventDescriptor. Actually, clitic pronouns are normally used to denote topical referents, just mentioned or contextually active52. The constructional block of the LiClitic construction specifies that the case role of the Li construction is filled by the accusative value. Indeed, this kind of clitic pronoun can only denote objects. With regard to the phrasal level, we can posit another subcase of the ActiveCauseMotion construction, labeled ActiveCauseMotionY, shown in Fig. 3.119 below. 52 In colloquial Italian, they may also be exploited to denote less active referents, but this strategy can lead to misunderstandings. 207 Grounding Meaning in Everyday Experience in the World Fig. 3.119: the ActiveCauseMotionY construction construction ActiveCauseMotionY subcase of ActiveCauseMotion constructional constituents v: Verb // inh. np: NP // inh. pp: PathPP // inh. form constraints np.f before v.f v.f before pp.f meaning: CauseMotionAction // inh. evokes EventDescriptor as ed // inh. constraints self.m.affected ↔ ed.topic self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. self.m ↔ v.m self.m.affected ↔ np.m // inh. self.m.causer ↔ ed.profiledParticipant // inh. self.m.process2 ↔ pp.m.spg The ActiveCauseMotionY construction inherits its constituents from the ActiveCauseMotion construction, together with the its semantic constraints. Nevertheless, it shows a different word order and an additional meaning constraints which identify the affected role of the CauseMotionAction schema the topic role of the EventDescriptor schema, besides the np constituent of the construction. When we turn to the clause-level construction, this time we can use Declarative again, with its subj and fin constituents, the latter unifying with the ActiveCauseMotionY construction, while the former unifies with the profiledParticipant role of EventDescriptor schema. Nevertheless we should also posit the existence of a higherlevel construction, in order to connect the clause to the dislocation. This is the duty of the DeclarativePlusLeftDislocation construction, shown in Fig. 3.120 below, while the Declarative construction (Fig. 2.18) is repeated in Fig. 3.121. 208 3. An ECG analysis of Italian caused-motion constructions Fig. 3.120: the DeclarativePlusLeftDislocation construction construction DeclarativePlusLeftDislocation subcase of S-With-SubjPlusNP constructional constituents leftdisl: NP decl: Declarative optional: c: comma constraints self.features.mood ↔ decl.features.mood form constraints leftdisl.f before c.f c.f before decl.f meaning: EventDescriptor // inh. constraints self.m ↔ decl.ed leftdisl.m ↔ decl.ed.topic Fig. 3.121: the Declarative construction construction Declarative subcase of S-With-Subj constructional constituents subj: NP // inh. fin: VP constraints self.features.mood ← declarative form constraints subj.f before fin.f meaning: EventDescriptor // inh. constraints self.m. ↔ fin.ed subj.m ↔ fin.ed.profiledParticipant inh. // The meaning of the leftdisl constituent of the DeclarativePlusLeftDislocation construction unifies with the topic role of the EventDescriptor schema (the affected role of CauseMotionAction). The resort to this higher-level construction is sufficient to allow us to analyze the example in (33)53. A similar solution may be adopted to deal with right dislocations. For instance, consider the example in (34) below: (34) I bambini li lanceranno al bersaglio, i sassi. The.MPL child.PL CL.3MPL throw:FUT.3PL to-the.MSG target.SG the.MPL stone.PL. “The children will throw them to the target, the stones” In the example (34) above we have just turned the left dislocation in (33) into a right dislocation. In order to account for this sentence, we have just to recur to a construction which is slightly different from that shown in 3.120. Indeed, the difference between the current example and the previous one just concerns the formal features of the two sentences, namely word order54. At the phrasal level, we can exploit the ActiveCauseMotionY construction just used in the analysis of the previous example, and at the clause-level we will make use of Declarative again. The only difference is to be found at the higher-level, where we can posit the existence of a construction which unifies the sentence with the right dislocation, labeled DeclarativePlusRightDislocation, shown in Fig. 3.122 below. 53 Constructions of this type may help us analyze similar but different utterances. Therefore, it may be auspicable to build an ECG ontology of this kind of constructions which currently, to our knowledge, is not available. 54 Actually, there are some slight functional differences between left and right dislocations. See Andorno (2003: 90-92) for a concise exposition of the main points of divergence. 209 Grounding Meaning in Everyday Experience in the World Fig. 3.122: the DeclarativePlusRightDislocation construction construction DeclarativePlusRightDislocation subcase of S-With-SubjPlusNP constructional constituents rightdisl: NP decl: Declarative optional: c: comma constraints self.features.mood ↔ decl.features.mood form constraints decl.f before c.f c.f before rightdisl.f meaning: EventDescriptor // inh. constraints self.m ↔ decl.ed rightdisl.m ↔ decl.ed.topic In the four examples just illustrated above, we slightly modified the sentence in (30), creating two sentence-level and two higher-level constructions whose syntactic and/or pragmatic structure diverges from that of the original example (sharing, nevertheless, the same semantic core), just to show that the grammar used in §3.2.1 and §3.2.2 can also be employed to handle cases other than simple declarative sentences. The choice of the sentence in (30) was arbitrary, in a sense driven by the simplicity of its structure and the prototypical caused-motion meaning of its verb, but analogous analyses could also be carried out of modified versions of all the sentences taken into consideration in the two previous subsections, as well as of other sentence-level constructions. Clearly, the two phenomena considered in the present subsection, wh-questions and dislocations, just represent two instances of the large number of existing grammatical constructions beyond simple declaratives, but the limited scope of the present subsection was just to provide some evidence of the fact that the analytical potential of ECG grammars is not limited to simple declarative sentences, but these grammars can also be exploited to deal with a wider range of grammatical phenomena. 3.3. Discussion In the previous section, we carried out an analysis of a handful of Italian causedmotion constructions, adopting an ECG approach similar to Dodge and Bryant's (forthcoming), but adjusted in order to cope with the properties of the Italian language and with the theoretical premises of Langacker's network model of polysemy. Overall, 210 3. An ECG analysis of Italian caused-motion constructions the results of our analysis seem to be satisfying, allowing us to make an inventory of the constructions involved in the process of comprehension of the sentence-level constructions taken into consideration, applying several notions and concepts developed in the Cognitive Linguistics framework during the last decades, consistent with the theoretical assumptions of the NTL research program. While §3.2 was virtually completely devoted to the analysis of our sentences, in the present section we shall embark in a brief discussion of the main issues observed in such an analysis. This section will be divided in three subsection. Since, as anticipated at the end of §3.1, the main topic under examination has to do with compositionality, the first section will be dedicated to the compositional aspects of some especially interesting syntactic constructions examined in the previous section. While compositionality has a paramount importance in ECG, a deep, thorough comprehension of the phenomenon can hardly be achieved without taking polysemy into consideration; as a consequence, the interaction between compositionality and polysemy in the language understanding process will be the topic of the second subsection. After discussing the main subjects of our analysis, in the final subsection we shall be concerned with some more aspects of the analysis carried out above which may necessitate some explanatory remarks. 3.3.1. Compositionality As briefly introduced in §3.1, compositionality is a hot topic in the linguistic debate. We already outlined the fact that scholars working within the standard (generative) paradigm tend to adopt an entirely compositional approach to the study of meaning. We also mentioned that this approach has been challenged by some cognitive linguists, who advocate the need for constraining the compositional principle, in order to account for those linguistic constructs whose meaning cannot be considered to be simply a sum of the meanings of its parts. Then, we also stated that recently some scholars pushed this proposal one step further, advocating a fully noncompositional approach, and we briefly outlined one of this entirely top-down approaches (Cristofaro 2008). We clarified that most scholars working from an ECG perspective tend to label their approach “compositional”, but the kind of compositionality characteristic of the ECG approach is quite different from the strict (one would be tempted to say “reductionist”) compositional approaches often adopted in the generative tradition (as recently reasserted by Feldman 2010). Indeed, we briefly outlined Ettlinger's (2005) theory of “constructional compositionality” which is not incompatible with Cristofaro's (2008) holistic approach, a compositionality which is constantly informed by the usage-based model of grammar and requires a simultaneously top-down and bottom-up analysis of data, meeting the requirements of a constructionist approach built to model language processing. As highlighted by Dodge (2010a, 2010b), one of the pillars of ECG compositionality is the combination of the verb and the A-S construction of the sentence under consideration. Following Goldberg (1995, 2006), in ECG argument structures are considered to be constructions in their own right and to play an important role in the construction of the meaning of a sentence: as a matter of fact, the A-S pattern is considered to convey a general, quite abstract meaning (corresponding to a determined schema), which is further elaborated by the verb (whose meaning not only corresponds 211 Grounding Meaning in Everyday Experience in the World to a determined schema, but also exhibits a particular x-net). Since our study focused on caused-motion constructions, one of our initial assumptions was that the meaning of all our example sentences was bound to the CauseMotionAction schema, and that all of them could be handled positing a limited number of A-S constructions. This is indeed what emerged from the analysis provided in the previous section. We saw that three out of four active sentence-level constructions could be handled by the same A-S construction (ActiveCauseMotion1), while the last one required that we resorted to a slightly different phrase-level construction (ActiveCauseMotion2), and the same pattern held with regard to our passive sentences. Both ActiveCauseMotion1 and ActiveCauseMotion2 are subcases of a more general ActiveCauseMotion construction. Conceptualizing A-S constructions as radial categories, as suggested by Dodge and Bryant (forthcoming), we can say that ActiveCauseMotion1 represents the prototypical center of the ActiveCauseMotion category, with ActiveCauseMotion2 being a radial extension. Let us now have a closer look at the relationship between verb and A-S construction in the example sentences analyzed in §3.2. The first three active sentences we handled in the previous sections show the verbs lanciare (“to throw), posare (“to lay”), and sollevare (“to lift”). In these cases, the semantics of the verb fits quite straightforwardly in the semantics of the A-S construction: since the meanings of these verbs is bound to the CauseMotionAction schema, exactly as the meaning of the ActiveCauseMotion construction and its subcases, these cases are rather unproblematic. The situation is a bit different with regard to the fourth sentence, which involves the verb spingere (“to push”). Indeed, the meaning of spingere does not reflect the meaning of the A-S construction, since its meaning is not bound to the CauseMotionAction schema as a whole, but rather to its first part, ForceApplication. The reader should remember that the CauseMotionAction schema, represented in Fig. 3.4 above, is a subcase of the CauseEffectAction schema (§2.3.2), which in turn is a subcase of the ComplexProcess schema, and therefore comprises two subprocess. The constituents of the CauseMotionAction schema are ForceApplication (Fig. 2.12) and MotionAlongAPath (Fig. 3.5), whose combination captures the fact that a causedmotion action implies a causer's exertion of force on a patient, resulting in the latter's motion through space. While the verbs lanciare, posare, and sollevare all denote the whole caused-motion action, the verb spingere just profiles the force-exertion action, not saying anything about the result of this process, leaving open two possibilities: the causer can either succeed or fail in moving the patient. We can find evidence of the difference between these two kinds of verb by using a couple of simple logical tests along the lines of those illustrated by Lewandowska-Tomaszczyk (2007: 141-142). Consider the following sentences: (35) *Ho lanciato/posato/sollevato il masso ma non si è Have:PRES.1SG throw/lay/lift:PSTPART the.MSG rock.SG but NEG REFL.3 be:PRES.3SG mosso. move:PSTPART.MSG “I have thrown/laid/lifted the rock but it has not moved” Ho spinto il masso ma non si è mosso. Have:PRES.1SG push:PSTPART the.MSG rock.SG but NEG REFL.3 be:PRES.3SG move:PSTPART.MSG “I have pushed the rock but it has not moved” (36) While the second sentence is perfectly plausible in Italian, the first one is not, since the 212 3. An ECG analysis of Italian caused-motion constructions verbs lanciare, posare, and sollevare, unlike spingere, imply the movement of the patient participant, which is not allowed to resist the force application process performed by the causer participant. If the causer's force exertion is not sufficient to make the patient move, Italian speakers exploit a circumlocution with the verb provare (“to try”). See the sentence in (37) below: (37) Ho provato a lanciare/posare/sollevare il masso Have:PRES.1SG try:PSTPART to throw/lay/lift:INF the.MSG rock.SG è mosso. be:PRES.3SG move:PSTPART.MSG “I have tried to throw/lay/lift the rock but it has not moved” ma non si but NEG REFL.3 The difference between spingere and the other three verbs, taking also into consideration their relationship with the A-S construction, may be characterized in cognitive semantic terms exploiting Talmy's notion of force-dynamics. Talmy (e.g. 2000) denotes four basic types of possible force-dynamic patterns55. These forcedynamic patterns can be summarized as below: a) b) c) d) the causer forces the patient to move, overcoming its intrinsic tendency to rest; the patient's intrinsic tendency toward rest is stronger than the force applied by the causers; therefore, the patient does not move; the patient's tendency toward motion overcomes the causer's opposition, so the patient moves; the causer blocks the patient, overcoming its tendency toward motion. The semantics of the ActiveCauseMotion construction and its subcases can be described by the first of the four conditions listed above: the force-application performed by the causer is stronger than the patient's inherent tendency to stand still, resulting in the patient's movement through space. The meanings of the verbs lanciare, posare, and sollevare are convergent with the meaning of the A-S construction, denoting a successful process of caused-motion. From this point of view, the semantics of the verb spingere is less specific; indeed, this verb does not imply that the caused-motion process was successful. As a result, we can say that the meaning of spingere encompasses both the a) and the b) conditions mentioned above: it is the integration with the A-S construction which makes the verb assume the the sense captured in a). The combination of verb and A-S construction in this case proves to be a useful cue for the language-understander to grasp the meaning of the sentence. Of course, the same remarks apply with regard to the passive sentence-level constructions analyzed in §3.2.2: the first three sentences involve the verbs scagliare (“to hurl”), scaraventare (“to fling”) and gettare (“to dash”)56, whose semantics fits exactly that of the PassiveCauseMotion construction and its subcases. As a matter of fact, the semantics of such verbs is bound to the CauseMotionAction schema. As a consequence, their 55 For a brief overview of the force-dynamics notion and its applications in cognitive semantics, the reader is referred to De Mulder (2007). See also Croft and Cruse (2004: §3.5) 56 Though in §3.2 we translated scagliare as “to hurl”, scaraventare as “to fling”, and gettare as “to dash”, and will be keeping this convention in this section, this choice is rather arbitrary: indeed, these three verbs are, to a great extent, interchangeable. 213 Grounding Meaning in Everyday Experience in the World meaning elaborates (through the values filling their x-net role roles) that of the A-S construction. This situation is captured by the PassiveCauseMotion1 construction. On the other hand, the semantics of the verb tirare (“to pull”) corresponds to the first subprocess of the CauseMotionAction schema only, so it simply reflects a part of the meaning of the A-S construction, and this relationship is captured by the PassiveCauseMotion2 construction. As a result, from the perspective of the language understander, the importance of the interaction between the A-S construction is rather clear. Indeed, on the one hand, the verb elaborates the meaning of the argument structure pattern; on the other hand, when the meaning of the verb is not an elaboration of the whole meaning of the A-S construction, the A-S construction guides the understander to a more specific interpretation of the verb's meaning. This is exactly what outlined above for the verbs spingere and tirare when they combine with the ActionCauseMotion2 and the PassiveCauseMotion2 constructions, respectively. Casting a glance at Dodge's (2010b) analysis of English caused-motion constructions, we must notice that constructions of this kind in English allow more variety than their Italian counterparts. Indeed, while in Italian we can just find verbs whose meanings overlap with either the whole CauseMotionAction schema or its first component process (i.e. ForceApplication), in English we can also find a third kind of verb, whose semantics profiles the second component process of this schema, MotionAlongAPath. This situation seems to be related to Talmy's typological distinction between verb-framed languages (like Italian) and satellite-framed languages (like English) mentioned in §3.1. For instance, consider the following sentence, which involves the verb slide, already found in the similar example in (8) above: (38) Jenny slid John's hand off her leg. The sentence in (38), analogous to that in (8) above, describes the following event: Jenny performs an indefinite act, causing John's hand to slide off her leg: the attention is focused on the effect of the process, not specifying anything about how this result was achieved. In this case, the verb represents the opposite of the Italian verbs spingere and tirare, with its meaning reflecting the second process which is part of the CauseMotionAction schema rather than the whole schema or its first process. Exploiting Talmy's force-dynamics patterns, we can say that the English verb slide does not imply that the motion process was the result of somebody else's force application. As a result, we can say that the meaning of slide encompasses both the a) and the c) conditions mentioned above: it is the integration with the A-S construction which makes the verb assume the the sense captured in a). In Italian, on the contrary, a sentence exemplified in (38) above would be translated by using a circumlocution with the verb fare, as in (39) below: (39) Jenny fece scivolare la mano di John dalla sua gamba. Jenny make:PRT.3SG slide:INF the.FSG hand.SG of John from-the.FSG her.F3SG.SG leg.SG “Jenny made John's hand slide off her leg” Getting back to our Italian data, a further construction which plays an important role in ECG is the clause-level construction, since it allows the undertander to grasp the mood of the sentence, distinguishing a statement from, e.g., a question, a request, or an order. The clause-level construction, moreover, unifies the A-S construction with the constituents of the sentences which are not part of the scope of the VP. Being our study 214 3. An ECG analysis of Italian caused-motion constructions restricted to declarative sentences, in our examples the subject was the only constituent occurring in such a particular position; however, in §3.2.3 we briefly outlined the FrontedObjWh-Question construction, a kind of clause-level constructions where the object is also outside the VP. We can therefore say that the clause-level construction imposes a pragmatic reading on the understander's possible interpretations of a sentence and it also helps them to grasp the role of the constituents not handled by the A-S constructions. For instance, in a sentence like (22) above, Declarative makes it clear that i bambini is the subject of the clause. The same condition holds when the subject of a clause is not overtly expressed. Indeed, since in Italian the verb agrees with the subject in person and number, it is rather effortless for the understander to retrieve the subject. This is the reason why in §3.2 we considered null subjects as subcases of the personal pronoun with which the verb agrees, without a phonological realization but still with the same grammatical and semantic features. Obviously, this “pro-dropping” strategy can only be used to denote referents who are active in the (discourse and/or situational) context. When the sentence under consideration involves a compound verb, i.e. it shows the presence of a finite auxiliary verb which precedes a nonfinite lexical verb, a further construction is used (FiniteAuxPlusVP), in order to account for the fact that the auxiliary verb is part of the fin constituent of the Declarative construction and so unifies with the ActiveCauseMotion1 construction. Positing this kind of construction let us capture the fact the auxiliary verb is connected to the A-S construction, performing a grounding function but without having effects on its semantic content. Consider for instance the example in (24) above, repeated in (40) below: (40) Il giovane netturbino ha sollevato un bidone dal The.MSG young.SG dustman.SG have:PRES.3SG lift:PSTPART a.MSG garbage.can.SG from-the.MSG marciapiede. sidewalk.SG. “The young dustman has lifted a garbage can from the sidewalk” You can observe that the auxiliary verb avere inflects for tense, person, and number, situating the utterance with respect to the moment of speaking, the discourse context, and the speech participants. On the other hand, the semantic load of the compound verb is still on the lexical verb, which appears in its (base) past participle form and on the A-S construction. Then, after the finite auxiliary verb avere, the understander will expect for there to be a nonfinite lexical verb. In the case of compound verbs in the passive voice, the finite auxiliary will be a form of the verb essere and, moreover, a nonfinite form of the auxiliary verb essere is also to be found in an intermediate position between the finite auxiliary and the nonfinite lexical verb. This second auxiliary, which appears in a past participle forms but inflects for gender and number (see §3.3.3 below), performs the function of specifying the fact that the event is construed from the patient participant's perspective. Like its finite counterpart, the nonfinite auxiliary has just a grounding function, without affecting the semantic content of the sentence. Consider the following sentence in (29) above, repeated in (41) below: (41) Eravate state tirate nella stanza dalle vostre Be:IMPF.2PL be:PSTPART.FPL pull:PSTPART.FPL into-the.FSG room.SG by-the.FPL your.2PL.FPL amiche ubriache friend.PL drunk.FPL 215 Grounding Meaning in Everyday Experience in the World “You had been pulled into the room by your drunk (female) friends” In the sentence under consideration, the finite auxiliary grounds the utterance in the discourse context and speech situation, the nonfinite auxiliary conveys the hint that the clause is passive, while the interaction between the lexical verb (which occurs, like the nonfinite verb, in a past participle form inflecting for gender and number) and the A-S construction determine the event being described. After the finite auxiliary essere, the understander will normally be led to expect either a nonfinite lexical verb or a nonfinite form of the auxiliary essere followed by a nonfinite lexical verb. We can now observe the “lower level” phrases found in our example-sentences, starting from the NPs which occur in the syntactic roles of subject and object. We have already had a look at null subject pronouns above, but most of the NPs found in our data comprise two lexical items: a noun preceded by a definite or indefinite article. It has already been observed that Italian determiners are inflected for gender and number, while nouns are inflected for number only, but gender can be changed through the process of derivation, providing a picture which is substantially different from the situation observable in English, where nouns just inflect for number and determiners do not inflect at all. Moreover, in the previous chapter, it was also underlined that the Italian language has no neuter gender; therefore all nouns in this language are either male or female, even when they denote inanimate referents. The two-gender system is often a source of controversy even when denoting animate referents, nonetheless. In the analysis of the sentence in (22), repeated in (42) below, we saw that the phrase i bambini could be interpreted in two ways: either referring to a group of male children or to mixed group of male and female children. (42) I bambini lanceranno i sassi al bersaglio. The.MPL child.PL throw:FUT.3PL the.MPL stone.PL to-the.MSG target.SG “The children will throw the stones to the target” This problem holds for all plural nouns whose ontological category does not exclude possible female referents57 and it is not possible to resolve it without any contextual information. Nevertheless, determiners and NPs are, to a great extent, more informative in Italian than in English. In §3.1 we briefly outlined the grammatical and semantic information conveyed by the NP il divano. The observations made with regard to that example holds for the majority of the examples analyzed above. A bit different case is the NP illustrated by the subject role of the construction in (40) above, where we can find a determiner plus a modified NP, i.e. the combination of a modifier and a noun, in a certain order58: il giovane netturbino. In this case, after the information given by the definite article, the adjective giovane provides additional semantic information on the status of the referent which is not provided by the following noun, and may also give rise to some statistically-driven expectations on the ontological category of the 57 Though these nouns seem to cover the vast majority of Italian nouns, it is still possible to find some which do not conform to this tendency, such as those related to priesthood, e.g. prete (“priest”), parroco (“parish priest”), vescovo (“bishop”), etc . 58 It Italian, word order is less strictly constrained than it is in English. Often, inside a modified NP, the modifier can either precede or follow the noun, without resulting in dramatic semantic shifts. Nevertheless, different orders are often preferred over others. Furthermore, the choice of a particular order may have pragmatic implications. We are not going to discuss these pragmatic differences here, but the interested reader will find a detailed account in Andorno (2003). 216 3. An ECG analysis of Italian caused-motion constructions referent denoted by the noun59 (Allen and Seidenberg 1998; Narayanan and Jurafsky 1998, 2001; Christiansen et al. 1999; Bicknell and Levy 2009). When we turn to PPs, a slightly more interesting picture can be observed. Dealing with caused-motion constructions, in our study, we focused on path PPs, which are made up of an allative preposition and an NP, and (in passive sentences only) CausativePPs, made up of a causative preposition and an NP. In Italian, if the NP following a “proper” preposition begins with a definite article, this determiner is fused with the preposition, producing a complex preposition, which assumes the semantic features of both its constituents but a new phonological form. For instance, al, observable in the sentence reported in (22) above, takes the meaning of motion toward (from the preposition a) a single uniquely-identifiable referent (from the article il). Moreover, the complex preposition receives from its article constituent the ability to trigger the inference that the grammatical gender of the noun denoting the referent will be male and singular. With this information available, the following noun just adds information about its ontological category. This is not to say that the rest of the information given by the noun is lost: simply, it is redundant because the speaker will have already inferred it from the previous constituent of the phrase and, since redundancy in Cognitive Linguistics is allowed (if not expected), this does not represent a problem. It may also be worth spending a few words on the sentence in (27) above, repeated as (43) below, since it shows a peculiarity, namely a cluster of two consecutive path PPs: (43) Gli arbitri vennero scaraventati dal ring sul tavolo da The.MPL referee.PL come:PRT.3PL fling:PSTPART.MPL from-the.MSG ring on-the.MSG table.SG by Undertaker. Undertaker. “The referees were flung from the ring to the table by The Undertaker” The first PP is made up of the complex preposition dal, which takes the meaning of motion from the internal part of an object to the exterior (from its preposition constituent da) and the information that such a landmark is a single and uniquely identifiable object (from its determiner constituent il) and the the noun ring, which specifies the ontological nature of the this object. The second PP comprises the complex preposition sul, which takes the meaning of motion from the outside to the uppermost layer of an object (from its preposition constituent su) and the information that it is a single and uniquely identifiable object (from its determiner constituent il). There is no substantial difference between the processing of each of the PPs in this cluster and a simple PP, excepting the fact that these two consecutive PPs are related by the fact that they convey the same motion event with reference to two different landmarks: the position inside the ring is external to the table, and the position on the surface of the table is outside the ring. This particular relationship is captured by the ComplexPathPP construction. Causative PPs represent an optional device exploited to express the causer participant of an event in a passive sentence. Unlike path PPs, they occur only with the preposition da, but are formed by preposition + NP as path PPs, and they show the same 59 In the present case, the reader will possibly have a tendency to expect a noun denoting a human being, on the basis of the distributional tendency of the adjective giovane in Italian, which normally (but not always) collocates with nouns describing human referents. 217 Grounding Meaning in Everyday Experience in the World behavior. Nevertheless, those found in the last three sentences analyzed in §3.2.2 are particularly relevant to our current purposes. This is because they show some peculiarities with respect ot the first passive sentence examined in §3.2.2. On the one hand, the NP embedded in the causative PP included in the sentence in (43) above includes a proper noun; on the other hand, the last two show the presence of modifiers which are a bit more interesting than giovane, briefly seen above. The first case is interesting because a causative PP characterized by the presence of a preposition and a proper noun is rather different from the PPs seen so far: it lacks the article and its noun does not only denote a uniquely identifiable entity, but a particular being that is unique in the world. The nickname Undertaker picks a specific individual, and so it does not take any determiners or modifiers. As a proper noun, it clarifies that its referent is one animate being and its level of accessibility depends on the discourse and situational context (though the value of proper nouns can be set as “inactive” by default). As already noted in §3.2.2, the case of the name Undertaker is curious: since proper nouns are not peculiar to human beings, the meaning of the word undertaker represents a cue to the interpretation (out of context) of the noun as referring to a human being, for this word denotes an exclusively human activity60. If we now turn to the sentence in (28) above, repeated in (44) below, we will have to deal with the causative PP dall'agente paonazzo: (44) Sono stato gettato nella cella dall' agente Be: PRES.1SG be: PSTPART.MSG dash: PSTPART.MSG into-the.FSG cell.SG by-the.MSG officer.SG paonazzo. red-faced.MSG. “I have been dashed into the cell by the red-faced officer” Not only does the adjective paonazzo provide us with information on the complexion of the referent of the noun it modifies, but its semantic content also entails that it is a human being (which is not clear out of context), while its form tells us the referent's natural gender, which is not evident from the form of the noun agente. Therefore, this PP provides us with the information that the event under consideration was performed by a single, uniquely identifiable red-faced male agent. A similar situation can be observed in the sentence in (41) above. The major peculiarity of the NP contained in the causative PP found in this sentence is the presence of a possessive. Unlike languages like English, while Italian possessives provide quite rich information about the possessor, they are also explicit with regard to the number of the possessed entity. Nevertheless, in this language possessives act like a particular kind of modifiers rather than determiners and they are preceded by the definite article (see §3.3.3 below); as a consequence, the information about the possessed entity's grammatical gender and number is not new. Then, it is the information on the possessor the most important contribution of the Italian possessive to the meaning of the NP (and, consequently, of the PP it is embedded in): vostre provides that the possessor of the referent denoted by the following noun is the addressee together with some other animate entity, entailing as a result that the possessor's number is plural, it belongs to the category of animate beings, and it is obviously active, involving the addressee of the message. The following noun amiche adds information on the ontological category of the referent and the 60 It should be noted, to be more precise, that undertaker is usually improperly translated into Italian as becchino (“gravedigger”), while the correct translation would be the (somewhat heavy) NP impresario di pompe funebri. 218 3. An ECG analysis of Italian caused-motion constructions postnominal adjective ubriache specifies its semantic properties, notably that it is a human referent, since drunkenness is a characteristic which can only be associated with human beings. So far, we have been casting a glance at the different kinds of constructions involved in the handful of example-sentences analyzed in §3.2, briefly discussing their semantic weight. However, we should not lose sight of the fact that these constructions derive from the segmentation of the whole message, and that they are not fully meaningful considered in isolation, i.e. their interpretation is partial and largely unspecific. Furthermore, it should always be kept in mind that linguistic information is provided to the understander sequentially, leading them to process utterances incrementally. Therefore, now we have to consider our example-sentences as wholes and try to account for the understander's online processing of these constructions. In other words, it is time to discuss the process of analysis illustrated in §3.2. Nevertheless, we cannot carry out such a duty without taking into consideration the other main issue of the present study, polysemy. This task will be performed in the next subsection. 3.3.2. The intersection between compositionality and polysemy in processing As mentioned in the introductory section of this chapter, compositionality and polysemy can hardly be parted from one another without a significant impoverishment of the account. Indeed, these aspects are unavoidably bound to intersect in the process of language understanding. Keeping in mind that we adopted Langacker's network model for the representation of polysemy, which is compatible with Lakoff's radial category model, we can now try to provide an account of the language understander's incremental processing of the example-sentences discussed in §3.2. Let us begin with the first sentence analyzed in §3.2.1: I bambini lanceranno i sassi al bersaglio. The definite article i shows the grammatical features of being male and singular, leading the understander to the inference that the following item will show the same characteristics; on the semantic side, it provides the information that the referent being denoted is a uniquely identifiable plural entity. The noun bambini, as expected, shows the same grammatical properties of the preceding article and adds the information that the uniquely identifiable entity being described belongs to the ontological category “child”, while without knowing anything about the context, it is not possible to know if the children being denoted are all males or include a mixed group of male and female children. The default word order of unmarked Italian sentences will also suggest to the understander that the phrase i bambini should probably (though not surely) be taken as the subject of the clause. Lanceranno is the third-person plural form of the simple future tense of the verb lanciare, whose interpretation is still hanging in the balance: while the most probable meaning for the verb lanciare is that of throwing (an object), there is still room for alternative interpretations depending on the following information. However, the use of this verb makes it clear that we are dealing with a transitive construction, and also that it is in the active voice. The following NP i sassi is crucial for the selection of the relevant sense of the verb and the interpretation of the sentence: on the grammatical side, the constituents of the NP are both male and singular, while with regard to meaning they together 219 Grounding Meaning in Everyday Experience in the World convey the information that the referent being denoted is a uniquely-identifiable plural entity whose natural gender is neuter; the ontological category of this referent, “stone” (rather than “rock”), is suggested by the presence of the verb lanciare (throwing a stone is of course more plausible than throwing a rock) and perhaps also a bit by the fact that the subject of the sentence denotes children (whose strength is supposed to be limited). Moreover, denoting inanimate referents, this NP confirms the understander's supposition that this sentence shows an unmarked word order and i bambini is the subject of the sentence, while i sassi correspond to the syntactic role of direct object. Then, we can find the complex preposition al, which includes the prepositional constituent a and the definite article il. While il is quite unproblematic, showing the grammatical features of being male and singular and the meaning constraints of denoting a single uniquely identifiable referent, prepositions are by their very nature highly polysemous and unlikely to be understood out of context. The path reading of the preposition is largely facilitated by the presence of the verb lanciare, which makes the allative interpretation of the preposition quite straightforward, together with the provisional structure of the clause. At the same time, the presence of the noun bersaglio as the other constituent of the PP also strengthens this reading, since a target is, by definition, an entity you throw something at. Indeed, though the meaning of bersaglio can undergo metaphorical extention, the presence of the verb lanciare and the NP i sassi in direct object position clarify that this is not the case. At the same time, the presence of this PP allows the understander to “sweep away” any doubts on the category of the referent of i sassi. While on the grammatical side bersaglio is male and singular, its semantics denotes a single referent, but its gender cannot be known out of context, since targets can be male, female, or inanimate, though the scene evoked by the sentence as a whole would suggest an inanimate reading. When the whole clause is available for processing, the understander will be fully able to definitively segment it into constituents, recognizing the structure of an abstract formal and semantic patterns, termed Declarative in ECG, which divides the clause into a subject, bound to be an NP, and a finite VP61. Another abstract pattern will be recognized with regard to the inner structure of the VP, which drives the interpretation of such a phrase putting formal and semantic contraints on its constituents (bound to be a verb, an NP, and a path PP, respectively, and occurring exactly in this order), and is endowed with a meaning in its own right: it describes an event in which a causer participant exerts force on a patient participant, obtaining its motion through space. This meaning is fully shared by the verb, which further elaborates it specifying how the caused-motion action is carried out. The event is described from the perspective of the causer participant, which is profiled. In summary, we consider the understander's incremental processing as a constant interaction between different levels of abstractness: concrete linguistic items provide general, surface bottom-up cues for the interpretation of a sentence, while simultaneously the more abstract structure instantiated by the concrete constituents supplies top-down constraints on these cues. For instance, we have just seen that the preposition a is a very generic cue, since it is highly polysemous. Nevertheless, the structure of the ActiveCauseMotion1 construction binds it to be a path PP, and the presence of the verb lanciare immediately preceding the preposition helps to make this constraint explicit. This bidirectional interplay is steady and involves all the linguistic items in a complex construction. Together with world knowledge, the Declarative construction, 61 In a sense, we may say that the Declarative construction roughly embodies the classical subjectpredicate dichotomy. 220 3. An ECG analysis of Italian caused-motion constructions the ActiveCauseMotion1 construction, and their constrained constituents present in this sentence permit the evocation of a scene portraying a future situation in which children who play a game whose scope is hitting a target by throwing stones to it, and they will probably trigger inferences on the details of the event (e.g. probably the event will take place in the open air, the children will enjoy it, and so forth). The second sentence-level construction analyzed in §3.2, Posasti il portacenere sul comodino, is rather similar to the previous one. Nevertheless, there are some differences which are worth signaling. First of all, the sentence begins with the verb posasti, which is the second-person singular form of the simple past tense of the polysemous verb posare. As such, it allows the understander to infer the subject of the sentence, which is the current addressee, so an active, singular, animate entity, and to know that the event being described took place in the past and has no effect on the time when the message is conveyed. Nevertheless, the correct meaning of the verb cannot be worked out, yet. This problem is solved when the following item, the determiner il, is processed. The grammatical and semantic characteristics of il have already been illustrated, but in this sentence it is particularly important since it introduces a referent which is going to appear in the syntactic role of direct object, and this fact allows the understander to know that the sentence is transitive and, since the verb posare can be used transitively only when it means “to lay (down)”, the range of possible meanings of the verb is peremptorily restricted. The monosemous noun portacenere (whose grammatical features are being male and singular, while on the semantic side it belongs to the ontological category “ashtray”, and it denotes a single referent whose gender is neuter) rules out the possibility of a figurative use of the verb posare; therefore the meaning of the verb is made clear. And it is the selected meaning of the verb, together with the rising structure of the clause, which makes it possible for the understander to get the meaning of the complex preposition sul right. Indeed, as a caused-motion verb, posare directs the understander to the allative reading of the preposition, which denotes a motion from the exterior to the surface of a landmark. In the end, the monosemous male singular noun comodino completes the PP providing the landmark toward whose surface the ashtray is moved: a single, uniquely identifiable bedside-table. In this case, Declarative, ActiveCauseMotion1, and the other lexical and syntactic constructions instantiated in the sentence-level construction, together with world knowledge, evoke a scene which describes the current addressee, in the past, who lays down an ashtray on a bedside-table (triggering some inferences, including the fact that probably the scene is set in a bedroom). The next sentence, Il giovane netturbino ha sollevato un bidone dal marciapiede, differs from the previous two in the presence of a modifier in the subject NP and an auxiliary verb. The first lexical item, the male definite article il, provides the grammatical hint that the referent of the following noun is male and singular, and the semantic information that it is a single uniquely identifiable entity. The modifier giovane is singular, but it is vague with regard to gender; nevertheless, the preceding article enables the understander to interpret it as male (at least from the grammatical point of view). From the semantic perspective, giovane is a polysemous noun, therefore we have to rely on the following noun to determine its actual meaning in the sentence. The monosemous noun netturbino, which shows the grammatical features of being male and singular, provides the understander with the information that its referent, besides being a single male being, belongs to the ontological category “dustman”. Therefore, the understander can identify the exact meaning of the adjective giovane, which 221 Grounding Meaning in Everyday Experience in the World assumes the scalar value “young” in the age continuum. As a result, they will be able to recognize that the phrase depicts a single, uniquely identifiable, young dustman. Again, on the basis of statistical factors, the phrase will be provisionally considered as the subject of the clause. The following lexical item ha, third-person singular form of the simple present of the verb avere, is by itself rather vague, since this verb can either be a lexical or an auxiliary verb, but the presence of a subsequent past participle let us infer that the latter condition is the relevant one. Therefore, we come into possession of the information that ha performs a grounding function, setting the event described by the utterance in the past. Sollevato, the unmarked past participle form of the verb sollevare, is polysemous, but the presence of the noun netturbino in subject position enables the understand to guess that the verb will denote an action related to the dustman profession (“lift up”), an impression which will be confirmed by the NP which occurs in object position. The male singular indefinite article un tells us that the referent of the following noun is a single entity which is not uniquely identifiable, but whose ontological category is instead accessible. Then, the male singular noun bidone adds information on the ontological category and the natural gender of the referent, respectively “garbage can” and “neuter”. To be precise, bidone is a polysemous noun, but the presence of the previous mention of the noun netturbino allows the understander to capture the meaning of bidone quite easily. The following lexical item is the complex preposition dal. As usual with complex prepositions, the kind of information provided by the determiner constituent is rather straightforward: the subsequent noun is male and singular, and the referent it denotes is a single uniquely-identifiable entity; on the other hand, the meaning supplied by its preposition constituent, in this case motion from the surface to the exterior of a landmark, is accessed thanks to the provisional structure of the clause and the presence of the verb sollevare. Finally, the noun marciapiede specifies the nature of this landmark. Marciapiede, a male singular noun, has at least three different meanings, but the presence of the nouns netturbino in subject position and bidone in object position suggests that the “sidewalk” meaning is the best fitting (even though the meaning “platform” may fit better in certain particular contexts, but it is a kind of “marked”, nonprototypical situation). In the case of this sentence, the understander will have to recognize another abstract construction, which schedules the presence of a finite auxiliary verb which is part of the VP in the present perfect tense. This construction is labeled FiniteAuxPlusVP. Here, the combination of world knowledge with Declarative, ActiveCauseMotion1, FiniteAuxPlusVP, and the other lexical and syntactic constructions instantiated in the sentence, evokes a scene describing a young dustman, in the past, who lifts a garbage can up from a sidewalk, activating some inferences (e.g. the event was set in a street and the scope of the action was emptying the can). As already specified in the previous subsection, the main peculiarity of the last active sentence analyzed in §3.2 above, Spingevamo le ceste nel ripostiglio, is the presence of a different A-S construction, which allows the verb to mirror only the first part of the complex schema to which its meaning is bound. The sentence begins with the verb spingevamo, the first-person plural form of the imperfect tense of the polysemous verb spingere, which immediately lets us infer the referent of the subject, i.e. the current speaker plus some other animate entity, whose status of accessibility is (inevitably) active. Then, the female plural definite article le provides the information that the referent of the upcoming noun is a plural uniquely identifiably entity. The following noun, ceste, is polysemous, but all its meanings have to do with physical inanimate 222 3. An ECG analysis of Italian caused-motion constructions objects (except one, which is nonetheless a highly specialized term incompatible with all the possible sense of the verb spingere, i.e. the equipment of an actor for a show), allowing the understander to grasp the relevant meaning of the verb spingere, i.e. the literal one “push” (while its other meanings are figurative extensions). The following item is the complex preposition nel. Again, while the meaning of its article constituent is straightforward (and already illustrated several times above), the meaning of its preposition constituent in is due to the presence of the verb spingere, whose meaning is bound to the ForceApplication schema, and suggests the allative interpretation of the preposition, describing a motion from the exterior to the interior of a landmark. Such a landmark is represented by the referent of the male singular monosemous noun ripostiglio, which denotes a single, uniquely identifiable lumber-room (whose natural gender is obviously neuter). The presence of ripostiglio in the path PP allows the understander to make some guesses on the relevant meaning of the noun ceste (“baskets”) previously found in object position. In the case of this sentence-level construction, when the whole sentence is available for processing, together with Declarative, the understander will recognize an A-S construction a bit different from the one observed in the previous examples, since its meaning is only shared by the verb only with regard to the first process of the CauseMotionAction schema, which corresponds the ForceApplication schema. As illustrated in the previous subsection, the A-S construction forces a caused-motion reading of the sentence, yet the presence of a verb of this kind profiles the ForceApplication process, which is (in terms of perspective) privileged with regard to the other process, MotionAlongAPath, which is part of the complex CauseMotionAction process. Declarative, ActiveCauseMotion2, and the other lexical and syntactic constructions instantiated in the sentence-level construction, together with world knowledge, evoke a (possibly ongoing or repeated) scene involving the current speaker together with some other animate being pushing two or more uniquely identifiable baskets, into a uniquely identifiable lumber-room. The interpretation of the sentence will give rise to some inferences on what is not explicitly stated (e.g. the action took place in the room of an apartment or a hotel...). We can now turn to the passive sentences analyzed in §3.2.2. The first of this sentences was Una pietra sarà scagliata contro la folla da un infiltrato. The female singular indefinite article una supplies the hint that the referent of the subsequent name is a single type-identifiable entity. Pietra as a noun can either denote stone as a material or a block of such a material: being preceded by the indefinite article una, it is quite a straightforward conclusion that the relevant condition in this case is the latter. Sarà is the third-person singular form of the simple future of the verb essere, which can be used as either a full verb or as an auxiliary. The following past participle clarifies that in this case sarà performs the auxiliary function, grounding the event described by the sentence in the future. Scagliata is the female singular past participle form of the verb scagliare, which can be used either literally or figuratively: since the NP in subject position denotes a stone, the first reading (“hurl”) is effortlessly recognized as the relevant one. The following lexical item is contro, a word which can be used (with more than one sense) as either a preposition or an adverb. The presence of the verb scagliare and the subject NP una pietra strongly biases the interpretation of contro in favor of the former option; moreover, the action denoted by the verb scagliare suggests an allative reading of the preposition. The female singular definite article la provides the hint that the referent of the subsequent noun is a single, uniquely identifiable entity. The following female singular mass noun folla shows, besides its literal meaning, a 223 Grounding Meaning in Everyday Experience in the World metaphorical extension, but the partial interpretation of the sentence drawn up to now suggests that the noun be interpreted in its literal sense (“crowd”). The preposition da is highly polysemous, but given the partial structure of the sentence, the suitable options seem to be restricted to two: causation or motion from. The doubt will be solved by the following NP: while the male singular indefinite article un simply tells us that the referent of the following noun is a single type-identifiable entity, the noun infiltrato is polysemous, but its relevant meaning (“infiltrator”) is accessed straightforwardly, given the rest of the clause and especially the noun folla. In turn, the recognition of the meaning of infiltrato simultaneously allows the interpretation of da as a causative preposition and sweeps away any doubt on the literal reading of folla. As in the case of the corresponding active sentences, when the whole clause is available for processing, the understander will be able to segment it into constituents, recognizing the structure of an abstract formal and semantic patterns, termed Declarative in ECG, which divides the clause into a subject, bound to be an NP, and a finite VP. However, the abstract pattern relative to the structure of the VP will be different: the constituent constructions will be four (a verb, an NP, a path PP, and a causative PP, occurring in this order) and, while the meaning of this A-S construction is the same of its active counterpart (the description of an event in which a causer participant's force-exertion action causes the motion of a patient participant through space) and is elaborated by the lexical verb, the event is described from the perspective of the patient participant, which is profiled. In this case, the combination of world knowledge with Declarative, PassiveCauseMotion1, FiniteAuxPlusVP, and the other lexical and syntactic constructions instantiated in the sentence, evokes a future scene describing an infiltrator, who hurls a stone against a uniquely identifiable crowd, activating some inferences (e.g. the event was set in a particularly crowded place, perhaps in some particular occasion). The second passive sentence, Gli arbitri vennero scaraventati dal ring al tavolo da Undertaker, shows the substitution of the auxiliary verb essere with venire, a common phenomenon in Italian when it is followed by the past participle form of a lexical verb in the passive voice of simple tenses. Moreover, it involves the path PP cluster outlined in §3.2.2 above and the presence of a resolved referent corresponding to the (unprofiled) causer participant of the action. The male plural definite article gli signals that the following noun's referent is a plural, uniquely identifiable entity. The following noun arbitri is polysemous, so it cannot be interpreted until some more information is added. Vennero is the third-person plural form of the simple past of the verb venire, which has several possible meanings. Scaraventati is the male plural past participle of the verb scaraventare, which can be interpreted either literally or metaphorically; in any case, its presence in this position clarifies that venire in this sentence performs the auxiliary function, grounding the event being described in a remote past. The interpretation of the complex preposition dal is driven toward an allative reading (motion from a landmark) by the presence of the verb scaraventare (whose both interpretations would influence the understander toward this interpretation of dal). At this point, the sentence is still fairly vague, but the subsequent presence of the noun ring strongly drive us toward a more specific reading of the sentence. Indeed, though the borrowed term ring has several meanings in Italian (although less than it has in English, of course), but the cooccurrence of the noun arbitri in subject position and the lexical verb scaraventare offer the hint that in this case the word denotes an arena where a boxe fight or a wrestling contest is going on (in particular, world knowledge would suggest wrestling, since it is arguably the only sport were beating referees is a common practice); simultaneously, the 224 3. An ECG analysis of Italian caused-motion constructions presence of ring in the PP suggests that arbitri should be interpreted as the referee of a sports game, and that the literal meaning of scaraventare (“fling”) is the relevant one. The subsequent complex preposition sul can be easily be recognized as a path preposition, because it is preceded by the verb scaraventare plus a complex PP, and it can be interpreted as referring to the same path motion denoted by such previous PP seen from another perspective, expressing motion onto a landmark. The following monosemous noun tavolo (“table”) can quite easily be interpreted as the relevant landmark. Given the provisional structure of the clause, the interpretation of the following preposition da as a causative should be rather straightforward, as should be the reading of Undertaker as the causer of the event described by the sentence. We can observe that the combination of world knowledge with Declarative, PassiveCauseMotion1, FiniteAuxPlusVP, and the other lexical and syntactic constructions instantiated in the sentence, evokes a past scene describing a wrestler called Undertaker, who flings two or more uniquely identifiable referees from a uniquely identifiable ring to a uniquely identifiable table, triggering some further inferences (e.g. the wrestler was not very happy with the referees' decisions, the wrestler was very strong and probably dressed in black, etc.). The subsequent sentence, Sono stato gettato nella cella dall'agente paonazzo, involves the presence of the passive voice form of a verb conjugated in a compound tense. The first word in the sentence is highly vague: sono is the form of both the firstperson singular form of the simple present of the verb essere, but it is also the thirdperson plural form of the same tense of the same verb. Moreover, we saw above that essere can be either a lexical verb (with many possible meanings) or perform the auxiliary function. Both problems are resolved when the second word is processed: stato is the male singular past participle of the verb essere, and thus the understander can strightforwardly conclude that sono in this sentence is the first-person singular form of essere, which is simply grounding the event in the past. Now, we know that the subject of the sentence is the current speaker, whose gender is male and its accessibility level is inevitably active. On the other hand, a new problem turns up: stato in this position can in turn be either a lexical verb or a passive auxiliary verb. The situation is disambiguated by the subsequent occurrence of gettato, the male singular past participle form of the verb gettare, which have several meanings but just one (“dash”) makes sense in combination with a human patient participant. The path meaning of the following complex preposition nella (motion into a landmark) is easily captured because it fits in perfectly with the meaning of the verb gettare. The female singular noun cella is strongly polysemous, but the arising structure of the clause rules out some senses, restricting the range of possibilities. The complex preposition dall' is problematic per se; indeed, while the causative reading of its preposition constituent is fairly straightforward, its determiner constituent is present in its short form, which provides us with the useful hint that the following noun begins with a vowel, but since it is shared by the male and female definite article, it does not provide any information on its grammatical gender. This problem persists when we get to the subsequent constituent, the singular noun agente, whose form is the same for referents of both genders. Agente is also a polysemous noun, but the presence of the noun cella in the path PP (and, to some extent, the occurrence of the lexical verb gettare) influences the understander toward a preference for the “police-officer” reading. Simultaneously, the occurrence of the noun agente suggests the kind of referent denoted by the word cella in this sentence (i.e. a prison cell). The male singular adjective paonazzo, in the end, 225 Grounding Meaning in Everyday Experience in the World informs us that the causer of the event described by the sentence is male, also specifying that the short form of the determiner constituent of the complex preposition dall' is to be considered as male as well. This modifier can take two possible senses, but one is a highly specialized and old-fashioned meaning (“purple”, with regard to the bishop's frock, but this use is very rarely encountered in contemporary Italian) which does not make sense with regard to the noun paonazzo modifies, tellings us that the alternative sense (“red-faced”) is the relevant one. In sentences like this (and the following), one more abstract construction is necessary, which schedules the presence of a nonfinite auxiliary verb between the finite auxiliary verb and the rest of the VP in the passive conjugation of compound verbs. This duty is comfortably carried out by the NonFiniteAuxPlusVP construction. The combination of Declarative, ActiveCauseMotion1, FiniteAuxPlusVP, NonFiniteAuxPlusVP, and the other lexical and syntactic constructions instantiated in the sentence, evokes a past situation in which the speaker is dashed into a prison cell by a red-faced policeman, activating some further inferences (e.g. the current speaker had been arrested, the police officer was high-handed, etc.). Our final passive sentence analyzed in §3.2.2, eravate state tirate nella stanza dalle vostre amiche ubriache, like the last active one examined in §3.2.1, shows a different AS construction with respect to the previous three, which allows its verb to mirror only a part of its constructional meaning. This sentence begins with eravate, the second-person plural of the imperfect tense of the verb essere, which lets us know that the subject of the sentence is the current addressee together with some other animate being, and its status is obviously active. As usual, it is necessary for further linguistic stuff to be available to know if essere is to be interpreted as an auxiliary or a full verb. The presence of the female plural past participle of the verb essere solves the problem, signaling that the auxiliary reading of eravate is the correct one. Moreover, the following participle state lets us know that the referents of the null subject are all female beings. With regard to the status of state is necessary to wait for the following linguistic item to know if state should be considered as a lexical verb or it just performs the grounding function. The subsequent occurrence of tirate, the female plural past participle of the verb tirare, tells us that the latter hypothesis is the relevant one. Tirare can take various meanings, but the presence of animate beings in subject position of a passive clause directs the interpretation of the verb to the meaning “pull”. The forceexertion meaning of this verb leads the understander to the allative reading (motion into a landmark) of the following complex preposition nella, which provides the hint that the following noun is grammatically female and singular. The female singular noun stanza has various distinct meanings, but the nature of the subjects and the meaning of the verb just allows a particular interpretation (“room”). Given the meaning of the verb and the provisional structure of the sentence, it is straightforward for the understander to capture the causative meaning of the complex preposition dalle, which also provides the hint that the NP constituent of the PP is grammatically female and plural while, on the semantic side, it denotes a uniquely identifiable plural referent. Vostre expresses a relationship of possession, also informing the understander that the possessor is the current addressee together with at least one more animate entity, adding that the possessed entity is grammatically female and plural and denotes a plural referent (see §3.4 below). The modified female and plural noun amiche can be interpreted in more than one way, but the possible readings of the noun are restricted to its most frequent meaning (unless the noun is used with irony) by the linguistic context, then informing 226 3. An ECG analysis of Italian caused-motion constructions us about the ontological category of the referent, together with its gender (female) and its number (more than one). Finally, the modifier ubriache provides information about the state of the referent of the noun, taking the scalar value “drunk” in the drunkenness cline. There exists some extensions to this default meaning, but since it fits perfectly in the clause, then there is no reason to opt for an alternative interpretation. As already mentioned, this sentence resembles the last active sentence outlined above, showing an A-S construction which is different from that of the other three passive sentences, allowing the meaning of the verb to converge with its only partially. As a matter of fact, the meaning of the verb tirare is bound to the ForceApplication schema rather than the CauseMotionAction schema. As a consequence, the sentence is still understood as depicting a caused-motion action, but the use of a verb like tirare profiles the initial part of the first process of the (complex) event, rather than the whole event. Declarative, ActiveCauseMotion2, FiniteAuxPlusVP, NonFiniteAuxPlusVP, and the other lexical and syntactic constructions instantiated in the sentence-level construction, together with world knowledge, evoke an event in which the current addressee and at least one more female animate being are pulled into a room by two or more uniquely identifiable friends of theirs, who are in a drunk state. The interpretation of the sentence will give rise to some inferences on what is not explicitly stated (e.g. the action happened in the context of a social event, which is taking place in an apartment or a hotel, etc.). At this point, it may be useful to remind the reader that the schematic phrase-level and clause-level constructions used in the analysis of example sentences carried out in §3.2 are not the result of a “building block” process. Instead, they are patterns which the understander had previously stored in their mental inventory of constructions learned through use, whose recognition takes place as the incremental processing of a sentence develops. Therefore, ECG cannot be considered as a strictly compositional approach (like those adopted in several strands of Generative Grammar); rather, it represents a usage-based model of language processing where formal and semantic analysis proceeds in a top-down and bottom-up fashion at the same time, supposing a mutual information process whereby a message sequentially presented to the understander is dynamically and incrementally processed making use of provisional representations of their partial structures, which also give rise to statistically-driven expectations on the basis of their grammatical and semantic local coherence (see e.g. Elman 1993; Christiansen et al. 1998; Allen and Seidenberg 1999; Narayanan and Jurafsky 2001; Tomasello 2003: ch.5; Seidenberg 2007, and with specific reference to ECG, Bryant and Gilardi forthcoming). These provisional representations are motivated as they seem suitable to instantiate entrenched and conventionalized schematic constructions. It should be highlighted that from this perspective, as seen in the discussion of our example sentences just outlined above, hints given at a certain point in time can be informative about different components of the utterance62. However, it is worth noting that, due to some critical properties of the language understanding process (it is carried out online, being to some extent dependent on variable factors such as the understander's level of attention, the fact that the message is supplied sequentially, language is sometimes ambiguous, etc.), the provisional structures internally built by the understander may turn out to be nonsensical when the whole sentence is processed; therefore, the understander may be forced to “backtrack” and reanalyze the sentence, searching for another solution less automatically and more consciously. It has to be stressed that often, especially when the linguistic input is ambiguous and/or the 62 A converging view can be found in Massaro and Jesse (2007). 227 Grounding Meaning in Everyday Experience in the World understander is disturbed by some noise, the final representation they draw may not correspond to the meaning conveyed by the message. This phenomenon is not explored in the present study, but the reader is referred to Ferreira et al. (2001) for a general overview of the problem as has been tackled in the fields of psycholinguistics and computational linguistics. In any case, it must also be remembered that when the the language understander is faced with a structure which is not part of their mental inventory of linguistic units, the understanding process will be slower and more demanding, and the final result is not always successful, i.e. the understander may fail to get the meaning of the utterance right. In the end of this subsection, it seems relevant to spend some words on constructional polysemy. In §3.1 we specified that in ECG the label “caused-motion construction” only applies to constructions which effectively denote a caused-motion action, thus solely to what Goldberg's (1995) study considers as prototypical and excluding what she defines as radial extensions of the category. Nevertheless, we also mentioned that this is not necessarily equivalent to the denial of the fact that these different kinds of construction share some common semantic (as well as syntactic) features, but rather suggests that this common ground is to be captured at a more schematic level. This duty is not straightforward and, so far, the problem has not been dealt with in detail. Nevertheless, a first little step has been taken. Dodge and Bryant (forthcoming) raise the problem that, in current ECG ontologies, there are no intermediate constructions between VerbPlusArguments (the most abstract construction comprising all VPs which include at least one argument, together with the verb) and the various kinds of possible syntactic constructions. Then, they advance a proposal to unify all active transitive constructions, building on the notion of profiling63. Dodge and Bryant propose the adoption of an additional schema, the FocalPair schema, with two roles: head and tail. These two roles are meant to be used to indicate that one of the participant, labeled head, is profiled with respect to the other, named tail (cfr. Langacker 1991). The FocalPair schema is shown in Fig. 3.123 below. Dodge and Bryant point out that this schema can be used to define a general active transitive A-S construction, which would fit between the VerbPlusArgument construction and the various active transitive constructions (such as ActionCauseMotion). This intermediate construction inherits a verb constituent from VerbPlusArguments, and adds an NP constituent, along with form constraints specifying the canonical word order of the construction. In the meaning block, it evokes the FocalPair schema and defines two binding constraints: the head role is bound to profiledParticipant, and tail is bound to the meaning of the np constituent (namely, NP)64. The ActiveTransitive construction is shown in Fig. 3.124 below. 63 The importance of conceptual asymmetry in ECG is apparent in the distinction between active and passive clause, and in the relationship between verb and A-S construction. 64 Note that the FocalPair schema can also be used to posit the corresponding passive transitive construction. It will be sufficient to bind the tail role to profiledParticipant and the head role to the meaning of the optional causative PP. 228 3. An ECG analysis of Italian caused-motion constructions Fig. 3.123: the FocalPair schema schema FocalPair roles head tail Fig. 3.124: the ActiveTransitive construction construction ActiveTransitive subcase of VerbPlusArguments constructional constituents v: Verb // inh. np: NP form constraints v.f before np.f meaning: Process // inh. evokes EventDescriptor as ed // inh. evokes FocalPair as fp constraints self.m ↔ ed.eventType // inh. v.m ↔ ed.profiledProcess // inh. fp.head ↔ ed.profiledParticipant fp.tail ↔ np.m This general construction for active transitive sentences seems a good starting point; nevertheless, more specific constructions have to be posited, in order to identify and classify the different types of event that can be described by this pattern. Dodge and Bryant point out that, as subcases of ActiveTransitive, these constructions will inherit its structure and will have to specify their event type, the relation between their meaning and that of the verb, and the bindings between FocalPair roles and participant roles associated with that particular type of event. In order to capture the connection between the constructions taken into consideration in the present study (Goldberg's prototypical caused-motion construction) and Goldberg's radial extensions, it will intuitively be necessary to follow a two-step precedure. First, an ECG A-S construction for any of these constructions will have to be posited. Second, it will be necessary to posit a construction able to abstract the semantic commonalities between these different constructions. This is not a straightforward procedure, since the formalization of an A-S construction is often problematic and can require a lot of time to be satisfactorily worked out. Moreover, in this case the creation of a new schema for each A-S construction will be necessary (if we stick to Goldberg's radial category, four new schemas will have to be posited in an English grammar, while probably an Italian grammar will require only three). Then, the need for a new superordinate schema will arise. As a result, the schema lattice as well as the construction network will become strongly more complicated. However, this is an enterprise which is in the ECG agenda for the near future, since its undertaking is essential in order to enable the model to scale up and be used to handle a growing number of different constructions, providing at the same time a finer-grained description of the construction network which 229 Grounding Meaning in Everyday Experience in the World constitute the grammar of a language. As already underlined in the previous chapter, ECG is a relatively young enterprise, and these constructions are not available at present; on the other hand, ECG is also a lively project and a lot of (especially computational) work is under way in this framework (see Boas forthcoming). Therefore, there is (at least a priori) no reason to believe that ECG scholars will not be able to overcome the problem outlined in these few lines and scale the formalism up to intermediate levels of abstraction between highly general and schematic constructions and low-level, concrete form-meaning pairings. 3.3.3. Some more issues related to our analysis In the present subsection, we shall be dealing with some issues which have been touched upon during the analysis of the caused-motion constructions taken into consideration in §3.2. We are going to dedicate some lines to a few features related to the ECG model and/or the Italian language which are not properly the center of attention in the present study, but nonetheless deserve to be briefly addressed for clarity's sake. First of all, it seems worth mentioning an ontological problem concerning the status of the events captured by our example sentences. In the present study, the concept of “caused-motion” have been used in a broad sense, denoting all force-application actions performed by a causer participant on a patient participant resulting in the latter's motion through space. Nevertheless, this point is not strictly speaking uncontroversial. As shown in §3.2 above, the CauseMotionAction schema, which in ECG is taken as the basic meaning of all caused-motion constructions, denotes a complex process, made up of two different processes, whose meaning is bound to the ForceApplication and the MotionAlongAPath schemas, respectively. The former schema represents the “cause” part of the motion event, while the latter corresponds to its “effect”. "so far, so good." Nevertheless, an ontological problem arises if we take into consideration the fact that the process of caused-motion can also be construed with regard to another dimension along with “cause-effect”, namely the temporal dimension. Indeed, if we examine the verbs present in our sentences taking into account the cause-effect parameter only, we will notice a substantial similarity between the events denoted by all of them (even though spingere and tirare privilege the force-application process while the other six verbs profile the complex event as a whole). However, if we analyze these verbs with regard to the other parameter, we can observe that posare and sollevare show a peculiarity compared with the other verbs involved in our examples: while other verbs imply that the two processes comprising the CauseMotionAction schema take place, to some extent, sequentially, posare and sollevare describe events in which the actions described by the ForceApplication and the MotionAlongAPath schemas take place simultaneously. As a matter of fact, the verbs posare and sollevare denote that the effector exploited to perform the action moves together with the object being moved. Therefore, in this case ForceApplication and MotionAlongAPath may be thought of as collapsing in a single process, while this is less plausible in the case of the other verbs examined in §3.2. However, while this issue has an empirical reality and should not be underestimated, it is not a major issue in our perspective. Since each verb has some (greater or lesser) 230 3. An ECG analysis of Italian caused-motion constructions degree of peculiarity which distinguishes it from others, binding the meanings a verb to the same schema of another or not is primarily a matter of levels of specificity. In other words, it entails that the selection of the parameter of choice can take place according to either broader or stricter criteria. It is plausible that adding a further schema (and, consequently, a different A-S construction) able to capture the (rather minimal) differences which distinguish posare and sollevare from the other verbs involved in our example sentences may allow us to realize a finer-grained grammar. Nevertheless, it would be considerably costly and, arguably, unnecessary. As a matter of fact, the core meaning of these two verbs is the same as the other verbs' (with the exception of spingere and tirare, of course): a forceful action whereby a causer participant causes a patient participant's motion with regard to a determined landmark. For instance, consider the sentence in (23) above, repeated in (45) below: (45) Posasti il portacenere sul comodino. Lay:PRT.2SG the.MSG ashtray on-the.MSG bedside-table.SG “You layed the ashtray on the bedside table” As illustrated in §3.2, this sentence describes the motion of an ashtray from the exterior onto the uppermost layer of a bedside table, caused by a force-application act performed by the current addressee of the message. The slightly different nature of the action depicted by this verb and that described by a verb like, for instance, lanciare can be simply captured by the filler of the x-net role specified in the meaning block of the construction corresponding to the verb. One may object that the difference between spingere and tirare and the other verbs is also fairly minimal, and nonetheless we distinguished their meanings, positing two new A-S constructions to deal with them (ActiveCauseMotion2 and PassiveCauseMotion2). However, this case is a bit different. Indeed, posare and sollevare diverge from the other verbs with regard to the temporal relation between the two processes which comprise the schema to which their meaning is bound, whereas spingere and tirare show a difference from the other verbs in terms of focal prominence, for they focus on one of them at the expenses of the other. In our opinion, while the value assigned to the x-net role in the meaning block can comfortably account for the peculiarity of verbs like posare and sollevare, it is not sufficient to deal with problems concerning conceptual asymmetry. As mentioned above, however, there does not exist any definite criterion to enable us to establish a cut-off point for the assignment of a schema to the meaning of a verb, and much depends on the analyst's subjective construal of the situation. As a result, it is possible (maybe even likely) that somebody else working with ECG grammars opts for a solution which diverges from the one adopted in this study. This is not a big problem, in any case, for the high level of flexibility of the ECG formalism allows us to build distinct grammars to carry out different analysis of the same phenomena, without implying that some must necessarily be wrong65. Indeed, as pointed out by Bergen and Chang (2005: 174), while the search for an ECG solution of a problem can often be exacting and time-consuming, it is also possible to find other cases where, on the contrary, “the number of constraints to be satisfied – and the number of ways in which to satisfy them – may (…) make it difficult to choose among competing analysis”. Another point which deserves to be briefly outlined regards the status of Italian 65 Obviously, this does not mean that analyst's “creativity” is unconstrained. While being rather flexible, the ECG formalism is at the same time precise enough to rule out unreasonable strategies. See Bergen and Chang (2005). 231 Grounding Meaning in Everyday Experience in the World possessives, whose particular inner structure suggests some morphological and constrastive observations. We have already observed that the last sentence analyzed in §3.2.2 shows the presence of an NP including a scalar modifier and a possessive. While the representation of the scalar adjective ubriache was rather straightforward, dealing with the possessive vostre was more problematic and required a different, compositional analysis. Indeed, as already specified above, Italian possessives, besides offering several hints on the identity of the possessor, also provide information on the grammatical gender and number of the possessed entity, a property shared (though to different extents) by possessives in several other languages, but absent in English (the language on which most ECG work has been carried out). Therefore, the formalization of Italian possessives is more complicated than that of their English counterparts. Since in Italian the relationship of possession is expressed making use of morphological means, we found it convenient (perhaps unavoidable) to momentarily shift our focus from the lexical-syntactic level and “pop into” morphology in order to find a satisfactory solution to the problem. Indeed, as observable in Tab. 3.1 below, Italian possessives are made up of two morphemes: a stem (in the CG sense of the word), which carries information with regard to the grammatical and semantic features of the possessor, and an inflectional suffix, which provides information on the grammatical gender and number of the possessed entity. The only exception is found when the possessor is in the third-person plural: in that case, the possessive always takes the same form, loro, providing information on the possessor only. Tab. 3.1: Italian possessives Possessor's person and number Singular I II III I II III All singular singular singular plural plural plural Mi-o Tu-o Su-o Nostr-o Vostr-o Loro Propr-io66 Possessed entity's gender and number Male Plural Mi-ei Tu-oi Su-oi Nostr-i Vostr-i Loro Propr-i Singular Mi-a Tu-a Su-a Nostr-a Vostr-a Loro Propr-ia Female Plural Mi-e Tu-e Su-e Nostr-e Vostr-e Loro Propr-ie As a consequence, analyzing the possessive vostre as an instantiation of a complex construction (Possessive, see Fig. 3.109 above) appeared as a suitable (though perhaps a bit unorthodox), solution in order to formally represent its somewhat rich semantic load. After all, this solution is rather similar to the process adopted to handle complex prepositions, from a formal and methodological point of view. Nevertheless, there is an important theoretical difference which might spoil the picture: while a complex preposition is the fusion of two independent lexical units, a possessive is a simple lexical unit. Therefore, in order to account for the fact that a single simple lexical unit provides information about two different referents, we had to deal with its “anatomy”, 66 Note that proprio is a reflexive form which is used as an alternative to the canonical form in the thirdperson singular and plural persons, and as an intensifier in the other persons. 232 3. An ECG analysis of Italian caused-motion constructions supplying a formal representation of the two morphemes included in this word. A similar condition could be satisfied only shifting to a different level of linguistic analysis. This is exactly the duty carried out by the Possessive construction. As a result, sublexical units turned out to deserve not to be ruled out a priori as a suitable resource to explain some particular lexical or syntactic phenomena. Therefore, although morphological analysis is not the scope of the present study67, the formal representation of morphemes proved useful to our analysis of one of the sentences examined in §3.2, and this fact compels us to make explicit a point hitherto taken for granted, namely that sublexical cues do play a role in the construction of partial structures which take part in the process of (especially spoken) language understanding (for experimental psycholinguistic evidence, see e.g. Christiansen and Chater 2001; Rohde and Plaut 2003; Pisoni and Levi 2007; Gaskell 2007b). There is also another reason to have a closer look at possessives. Indeed, the Italian language shows a peculiarity with regard to this part of speech: while in most other languages they are considered determiners, in Italian they seem to be closer to the category of modifiers. This claim is due to the fact that Italian possessives behave in several (though not all) respects as ordinary adjectives. The most peculiar of their properties is that they must (virtually always) be preceded by an article68: as an example the NP occuring in the causative PP in the last sentence analyzed in §3.2.2, the use of the definite article in the NP le vostre amiche ubriache is mandatory, since dropping the article would lead to an unacceptable phrase (*vostre amiche ubriache). The same condition holds in regard to NPs including an indefinite determiner: una vostra amica ubriaca is the correct form. This is exactly the opposite of what happens in English, where the possessive is considered a determiner, and since English determiners are mutually exclusive, possessives cannot follow articles (see Berk 1999: ch. 2). What is more, unlike English, in Italian there is a single form common for both possessives which modify an NP (termed “possessive adjectives”) and for those which stand as heads of NPs in their own right (labeled “possessive pronouns”); furthermore, the latter as the former must be preceded by the determiner (e.g. le vostre), while in English they stand on their own. When they function as predicate adjectives, possessives are preceded by the definite article when they need to specify that their referent is uniquely identifiable, while they are not when this specification is unnecessary. Beyond the evident contrast between Italian and English possessives, we have to precise that there are also languages which adopt different (and often to some extent intermediate) strategies. As an example, we can cast a glance at the properties of Spanish possessives (for a more detailed exposition, see Gómez Torrego 2002: 76-79). In Spanish, possessives are considered determiners and they cannot be preceded by articles 69, as in English (nuestras amigas borrachas / unas amigas nuestras borrachas vs. *las nuestras amigas borrachas / *unas nuestras amigas borrachas). Moreover, when the forms they take when they head an NP (labeled “full forms”) are different from those they take when they modify one (“short forms”), excepting the first two plural persons. 67 Although most work on ECG do not directly deal with morphology, a systematic application of the ECG formalism to morphological phenomena can be found in Schneider (2010). See also Chang and Gurevich (2004). 68 A notable exception can be found when they modify kinship terms: in that case, the article has to be dropped. For instance, people have to say mia madre (“my mother”), rather than *la mia madre (an unacceptable phrase). 69 Actually, the use of possessives preceded by determiners can be found in some varieties of Peninsular Spanish. Nevertheless, it is not accepted in Standard Spanish (Gómez Torrego 2002: 77). 233 Grounding Meaning in Everyday Experience in the World Nevertheless, Spanish possessives also show some properties which overlap with those of Italian possessives: full forms of Spanish possessives inflect for gender and number, though the third-person singular and the third-person plural share the same forms. On the other hand, short forms just inflect for number, excepting the first and second-person plural (again, the third-person singular form coincides with the third-person plural). Moreover, like in Italian, full forms are preceded by determiners when they head an NP. Finally, they behave like Italian possessives when they are in predicate adjective position. Given the properties of Italian possessives, in our grammar we preferred to posit the Possessive construction as a sort of part of speech in its own right, describing it as a direct subcase of the Word and the HasAgreementFeatures constructions, rather than depict it as a subcase of the Adjective construction. The observations made in the latest few lines suggest an intuitive remark which may not be totally irrelevant. While we have just observed this problem “in passing”, our brief sketch of the different properties of possessives in three different languages, though superficial and so without expectations, seems to provide us with some little cues about the status of what are traditionally called “grammatical categories”. The fact that possessives in these three languages (which belong to the same linguistic family) behave quite differently seems to supply a (very) little bit of support the position advocated by some scholars (and implicitly adopted in the present study, along with most ECG work), who claim that these categories are in fact convenient labels, but they are not likely to denote universal properties of language (see e.g. Croft 2001, 2003). On the contrary, it seems a hint of the fact that they are both language-specific, since their semantic features and distributional behavior across languages do not always overlap, and construction-specific, behaving differently across constructions within the same language. Another point which was touched upon during our analysis concerns the way of dealing with scalar concepts. In three of the sentence-level constructions analyzed in §3.2 we briefly addressed the issue of modification, since in these sentences we could observe the presence of some adjectives. These adjectives belong to the category of scalar modifiers, which represents a rather interesting class of adjectives, since their semantic properties cannot be considered absolute values, but rather should be seen as gradable concepts. Actually, it is probable that the semantic content of most adjectives can be seen, to a some extent, as a matter of degree70. We represented the meaning of these adjectives as assuming a certain value with regard to a defined reference scale. In the case of giovane, for instance, we postulated that it takes a “young” value in a cline referred to age. This characterization can be sufficient for the present purpose, since the quantitative dimension of scales is more a matter for representations of the process of enactment rather than analysis (as remarked by Feldman 2002: 12). Nevertheless, the given description is somewhat vague, and some more information may be useful for the reader to understand the ECG conception of scales; therefore, the following lines will (very) briefly sketch the general idea. We have just underlined that a scalar concept, such as age, could be thought of as a continuum delimited by two opposite poles. In a natural language like Italian, there are several adjectives related to age. The most natural move following from these two basis conditions would be the assignment of each of these adjectives to a corresponding position along the continuum, according to its similarity to the two extremes of the 70 Notable exceptions are adjectives which denote basic ontological features such as animate/inanimate, dead/alive, etc. 234 3. An ECG analysis of Italian caused-motion constructions cline. Additionally, a numerical score may be associated to each position. Actually, this is the metaphor lying at the foundations of the ECG notion of scale, but it does not tell us the whole story. As illustrated by Feldman (2002), ECG scholars assimilate scales to the interval between the two real numbers 0 and 1; therefore, one pole of the cline will be given the “0” value, whereas the other will be assigned the “1” value, and all the points lying in between will take intermediate values (e.g. 0.01, 0.02, etc.). Nevertheless, it would be rather unrealistic to deal with the meaning of a vague adjective like giovane (“young”) as if it were a single, discrete number71. On the other hand, conceptualizing its semantic structure as covering a range of numerical values included between the two extreme poles of the scale seems much more suitable. Indeed, adjectives like the one under consideration show a broad meaning, whose interpretation is highly dependent on the ontological nature of the referent denoted by the noun it modifies. For instance, a fifteen-year-old human being would probably be considered “young” by most people, while a cat of the same age definitely would not. Moreover, the range of values covered by the meaning of this kind of adjective can partially overlap with the semantic content of others. As an example, the meaning of adolescente (“teenage”) can be considered as covering a smaller range of values which corresponds to a section of the range of values related to the meaning of giovane. Therefore, viewing the meaning of this kind of modifiers as covering a range of values seems more realistic than considering it as a single value. However, while the strategy just outlined above seems a suitable solution to carry out a formal representation of the semantic value of scalar adjectives, it is appropriate to explicit the fact that the formalization of this kind of concepts represents an approximation whose rate of precision is especially unpredictable. Indeed, while a model of language processing (and, more generally, cognitive activities) is inevitably bound to oversimplify the picture, this is a topic which is particularly dependent on a number of variables related to the situational context, including sociocultural factors and, more importantly, the subjective beliefs of the addressee. Obviously, this condition does not belittle the enterprise, but it just makes one of its limits explicit72. The solution proposed by Feldman (2002) seems convenient and consistent with the assumptions of the NTL program, and during the last years the realization of formal models of language processing have proved to be helpful for the development of the ECG enterprise. Nevertheless, it should always be kept in mind that there are some issues whose vagueness and/or complexity posit particular challenges and constraints, and whose satisfactory solution may be more demanding and time-consuming. In other words, while certain aspects can be rather properly captured by existing models and others seem likely to be managed in the near future, there are also several problems whose solution is still very much under way and may need a long time to be fully worked out, provided that some intrinsic limits are unlikely to be completely overcome at all (for a representative point on the computational virtues and vices of ECG, see Feldman and Gilardi forthcoming).. Finally, we want to jot down an observation on the status of Italian participles. In §3.2, we observed five Italian past participles: one appears in an active sentence, while the remaining four are all included in passive sentences. It was argued that in this language, when past participles are part of compound forms of verbs occurring in active 71 Such a solution would be more plausible for more precise adjectives such as, for instance, diciottenne (“eighteen-year-old”). 72 Probably, this is not a problem specific to the ECG enterprise, but rather an intrinsic limit of any linguistic formalism. 235 Grounding Meaning in Everyday Experience in the World transitive sentences, they are preceded by the lexical verb avere as in the case of the third sentence analyzed in §3.2.1. It was also remarked that past participle in active sentences do not inflect for gender and number, but appear in a common base form instead. Indeed, the only agreement (in number, along with person) we can find in this kind of sentence is that between the subject of the sentence and the auxiliary verb. Consequently, if the sentence in (40) above were replaced by one with female subjects and objects, the past participle would not be affected by this shift, as can be observed in (46): (46) La giovane cliente ha sollevato una bottiglia dallo The.FSG young.SG customer.SG have:PRES.3SG lift:PSTPART a:FSG bottle.SG from-the.MSG scaffale. shelf.MSG “The young (female) customer have lifted a bottle from the shelf” On the contrary, it was observed that past participles which take part in compound forms of verbs involved in passive transitive sentences, which follow the auxiliary verb essere, do agree in gender and number with the subject of the clause (providing information on the grammatical gender of the subject of the sentence, when it is not explicitly expressed). We also asserted that, when they occur in passive constructions, the status of Italian past participles can be considered closer to that of predicate adjectives then when they appear in active sentences. Nevertheless, this is an oversimplification: indeed, if we take a step away from the canonical, unmarked declarative sentence, we can find some contexts in which past participles inflect for gender and number even when occurring in active constructions. It is exactly the case of sentences in which the verb is preceded by a third-person direct object clitic pronoun73. As an example, the past participle of (left or right) dislocation is inflected for gender and number. Consider the following variations of the example in (42) above: (47) I sassi, i bambini li hanno lanciati al the.MPL stone.PL the.MPL child.PL CL.3MPL have:PRES.3PL throw:.PSTPART.MPL to-the.MSG bersaglio. target.SG “The stones, the children have thrown them to the target” I bambini li hanno lanciati al bersaglio, i The.MPL child.PL CL.3MPL have:PRES.3PL throw:.PSTPART.MPL to-the.MSG target.SG the.MPL sassi. stone.PL “The children have thrown them to the target, the stones” (48) Dislocations are not the only kind of phenomenon which triggers the inflection for gender and number of the past participle. An analogous situation can be found when the object of a sentence already occurred (in either subject or object position) in a previous sentence, as in the examples in (49) and (50) below: (49) Marta ha spinto Luigi contro il muro. L' aveva Marta have:PRES.3SG shoved:PSTPART Luigi against the.MSG wall.SG CL.3FSG have:IMPF.3SG insultata. 73 When the verb is preceded by other person direct object pronouns, the agreement between object and verb is optional (see Moneti and Lazzarino 2006: 89-90). 236 3. An ECG analysis of Italian caused-motion constructions INSULT:PSTPART.FSG “Marta has shoved Luigi against the wall. (He) insulted her” (50) Carlo ha spinto la ladra contro il muro. Carlo have:PRES.3SG shove:PSTPART the:FSG thief.FSG against the.MSG wall.SG ha stordita. have:PRES.3SG stun:PSTPART.3FSG “Carlo has shoved the (female) thief against the wall. (He) has stunned her” L' CL.3MPL As a consequence, the inflection of Italian past participles for gender and number is not a peculiarity of passive sentences, but also some active sentences. Nevertheless, there is still a difference between the two cases. In passive sentences, past participles actually seem more similar to adjectives, while in the particular active sentences seen in the examples above, this does not appear to be the case. Indeed, intuitively the motivation for the inflection of past participles seem different. While in passive sentences past participles actually seem to represent an intersection point between adjectives and verbs (i.e., they appear to denote an action, but in a sense also a property)74, in dislocations the lexical verb has a strong connection with the topicalized constituent of the sentence, while in sentences like those in (49) and (50) above, they seem to perform a disambiguating function, allowing to distinguish the subject from the object of the sentence. It seems relevant, therefore, to specify that the situation cannot be simply described as concerning a clear-cut distinction between active and passive sentences, but the picture is somewhat more complex. While the investigation of the bundle of issues just outlined above was to a certain extent neglected in the present study, where we found it more appropriate to direct our attention to the phenomena of compositionality and polysemy, we considered it opportune to conclude the present section with a list of remarks on these phenomena, which may be useful for future work. 3.4. Summary and remarks In the previous sections, we carried out an investigation of some simple active and passive Italian caused-motion constructions, exploiting the ECG model under development at Berkeley within the NTL research program supervised by the computer scientist Jerome A. Feldman with the collaboration of the linguist George Lakoff. The purpose of this chapter was to apply the ECG formalism (usually used to analyze data from the English language) to Italian data, providing for the adaptation and modification required by the grammatical system of this language. In §3.1, we briefly introduced the type of data chosen and the criteria whereby they were selected, along with the methodology we were going to adopt in our study. First of all, we specified the reason why we decided to investigate a phenomenon of the Italian language, i.e. the combination of two main factors: the lack of ECG studies on this language (and typologically related languages as well) and the fact that it is the language we know best, being our native language. The choice of the caused-motion construction drew inspiration from Goldberg's (1995) influential study on English, since its consultation made it straightforward for us to reach two conclusions: on the one 74 It should be noted that past participles can also be used as attributive modifiers (see Croft 2001: 88). 237 Grounding Meaning in Everyday Experience in the World hand, several types of construction frequently found in English have no corresponding construction in Italian (e.g. the “way” construction) or such a construction is either attested only with particular syntactic constituents (e.g. the ditransitive construction) or simply rare (e.g. the resultative construction, whose status in Italian is also sometimes unclear, see Broccias 2003), whereas the caused-motion construction, though less productive then in English, can be observed with a certain regularity in Italian; on the other hand, the Italian caused-motion construction shows some differences from its English counterpart (e.g. it is less obiquitous), a fact which seems to be connected to Talmy's typological distinction between verb-framed and satellite-framed languages (see Ochsenbauer and Hickmann 2010). Then, we clarified that, though caused-motion constructions are often used metaphorically, we wanted to deal with constructions showing literal meaning, i.e. denoting actual caused-motion events, in order to provide an ECG representation of their “basic” semantic content. We also specified that the example sentences to be examined were built on analogy with real utterances drawn from ItWaC (a very large corpus of written Italian which crawls real usage data from the world wide web), being careful to meet some formal and semantic requirements. From a theoretical point of view, we asserted that our two background assumptions were the following: a radial category of A-S constructions would be able to capture generalizations about Italian caused-motion constructions, and a small set of schemas would account for their basic meaning. Taking these assumptions as a starting point, our main target phenomenon was semantic compositionality, since it has long been a matter of debate in linguistic studies: following Feldman (2010), we specified that the kind of compositionality assumed in ECG has little to do with the strict principle of compositionality often adopted in traditional semantics, according to which the meaning of a complex construction is made up of smaller building blocks. On the contrary, being ECG a usage-based model of grammar, the (often multiple) meanings of a “block” are abstracted from real occurrences in larger linguistic units attested in everyday language use. In other words, following some radical approaches advocated by scholars working in the Cognitive Linguistics framework (Croft 2001; Taylor 2003; Cristofaro 2008), we accepted the larger construction to be the semantic primitive, with the meaning of its components being derived via a process of abstraction; nevertheless, we specified that, once abstracted, this meaning can be used in new and creative ways. Since ECG is a formalism meant to describe the process of language understanding and that such a process is incremental, we posited that the analysis of a linguistic utterance should proceed in both top-down and bottom-up manner, with smaller surface forms supplying bottom-up cues and the larger schematic unit providing top-down constraints on their constituents (Bergen and Chang 2005). Moreover, since compositionality is tightly related to polysemy, we also adopted a precise model of polysemy, namely Langacker's network model, according to which different senses of a unit are separately stored (as a result of entrenchment and conventionalization), with an encompassing higher-level abstraction emerging later. This model is perfectly consistent with the usage-based approach to grammar and largely compatible with Lakoff's radial category model adopted by Dodge and Bryant (forthcoming) in the paper on which we based the analysis of our data. In §3.2, our selected example sentences were analyzed making use of the relevant ECG “operational machinery”, in order to specify all the constructions involved in each of them, including concrete instances and more schematic constructions, at the lexical, phrasal, and clausal levels. As just mentioned, we took Dodge and Bryant's 238 3. An ECG analysis of Italian caused-motion constructions (forthcoming) study as our main reference text for the analysis, making also use of the ontology found in Gilardi's Starter2 grammar. Obviously, since there are several factors on which the Italian language diverges from English, we built our ECG grammar drawing a lattice of constructions which aimed to deal with this language (it is relevant to remind the reader that ECG grammars are language-specific), for instance including information on the grammatical gender of nouns, which has to be distinguished from their natural gender for the sake of formal and semantic accuracy. Overall, the adaptation of the ECG formalism to provide a formal representation of Italian lexical and syntactic constructions was carried out rather effortlessly, with a few notable exceptions whose arrangement turned out to be a bit more difficult and timeconsuming, such as possessives and some vague nouns and modifiers, but without any substantial, unsolvable problems. In general, we can remark that, since Italian word order is somewhat looser than English word order, some phrase-level and clause-level constructions adopted in our study do not reflect the full range of possible patterns which can be found in Italian (e.g. the order of path PP and causative PP may in some cases be reversed without any substantial semantic or pragmatic change), but this is not a major concern, since it can be easily handled by adding some constructions to the lattice. As a result, the ECG formalism performed quite well in the representation of the constructions involved in Italian caused-motion constructions taken into account, providing a satisfactory picture of the event described by each sentence. As for the specific grammar we designed in order to handle our data, even though it was not “acidtested” on computer software, there does not seem to be any reason why it should not pass the test (at least in principle). It may be necessary for it to undergo some slight modifications, in order to meet the requirements of the software, but there should not be any need for radical changes. In §3.3, we discussed the role of compositionality and its interaction with polysemy in the processing of an utterance undertaken by a language understander, given the incremental nature of this activity which underlies the process of constructional analysis portrayed in the previous section. First of all, we briefly addressed the relationship between verbs and A-S constructions, in order to show how their mutual interaction influences the interpretation of each sentence as a whole, remarking that this interplay has an important role in the process of understanding. Later, we cast a glance at some NPs and PPs encountered in our target sentences, showing the contribution of each lexical item to the mental representation of the entities or events depicted by such phrases. Then, we addressed our whole sentences one at a time, examining the role of the grammatical and semantic features of each lexical item, together with the phrase and clause-level constructions, in building representations incrementally, in order to finally arrive at the most plausible interpretation of the sentence. It is important to reassert that, in order to mentally build these representations, the higher levels cooperate with the lower levels in a simultaneously top-down and bottom-up fashion, with word-level constructions doing most work at the beginning of the sentence and schematic constructions coming into the picture as more formal and semantic information is given. At the end of the section, we briefly overviewed some issues related to ECG and/or the Italian language touched upon in §3.2 which, though not the center of attention of the present study, were (in our opinion) worth discussing a little bit more. On the basis of the points made so far, we can quite confidently state that the descriptive formalism of ECG is a suitable tool to investigate Italian data, since its flexibility allows the analyst to build a grammar of their target language adapting its 239 Grounding Meaning in Everyday Experience in the World conceptual primitives to the properties of this language. As a consequence, we can move on to asserting that the ECG formalism represents a powerful asset to perform at least the first two basic functions it was designed for75: specifying the shared grammar and conventions of a language community and carrying out a technical analysis of linguistic data. Indeed, as pointed out above, the ECG model provides a computational representation to several conceptual notions developed within the Cognitive Linguistics tradition during the last decades (image schemas, force dynamics, etc.), allowing us to make use of them in order to analyze and formally represent the semantic content of linguistic constructions of various size (a recent review is provided in Feldman 2010). Nevertheless, providing a list of the positive sides of this study does not tell us the whole story. Indeed, notwithstanding the fact that the application of the ECG formalism to Italian data was (all in all) successful, the present study is not free from important limitations (which stem from both theoretical and operational issues), mainly shared by most other ECG studies. First of all, the range of data taken into consideration is rather limited and strongly homogenous. This was a methodological choice, since we wanted to focus on a very specific topic, in order to put all our efforts into its description, explanation, and representation in the perspective of ECG. Indeed, this decision is well motivated by the need to concentrate on a precise target phenomenon, not to mention the fact that carrying out an ECG analysis of any sentence is enormously costly in terms of time and space. Nevertheless, this decision is not costless, for it consequently constrains the range of conclusions and generalizations which can be drawn on the basis of its results. This problem is directly related to the issue of constructional polysemy sketched in §3.3.3. As a matter of fact, we highlighted that taking into consideration other kinds of construction such as Goldberg's extensions to “prototypical” caused-motion constructions would require the tweak of a common higher-level construction, which is in the schedule of ECG scholars but is not currently available and, though this goal seems reachable in the long term, the problem is unlikely to be resolved in the near future (see Dodge and Bryant forthcoming). A major problem which is related to ECG but also quite generally shared by other linguistic theories and models, regards the (relative) neglect of the role of context. Indeed, handling linguistic data out of context makes any effort toward their semantic explanation rather incomplete and artificial, since communication can hardly take place out of context. As recently restated by Feldman: Recall that compositionality of language is manifest if the definition explicitly includes the discourse and situational context of an utterance and also that strict context-free compositionality is demonstrably false. Therefore, any adequate formalization of compositionality entails formalizing context-dependence in language understanding. (Feldman 2010: 18) In a sense, the interplay between schematic syntactic patterns and word-level constructions driving the interpretation of a sentence and its constituent may be understood as a sort of minimal, local discourse context. Nevertheless, sentence-level constructions are rarely (if ever) self-contained. On the contrary, they are supposed to be related to other language chunks along with they are part of specific language events. 75 The other functions it is supposed to serve are far beyond the purpose of the present study. However, it is supposed to perform also as a computer specification for the implementation of linguistic theories, a front end system for applied language understanding tasks, a representation for models and theories of language acquisition, and a high-level functional description for biological and behavioral experiments (see e.g. Feldman et al. 2009; Feldman and Gilardi forthcoming). 240 3. An ECG analysis of Italian caused-motion constructions In particular, the interpretation of an utterance is influenced by previous linguistic units, which indeed provide a general frame for the interpretation of new information. In other words, discourse context drives the interpretation of an utterance, making the process of meaning selection illustrated in §3.3.2 easier and more straightforward. Besides discourse context, situational context plays a very important role as well. Indeed, extralinguistic information provides many cues to the language understander, contributing to the interpretation of an utterance. Therefore, both types of context affect the performance of the understander in the language processing task76. To cut a long story short, we can say that the interpretation of an utterance relies on the combination of probabilistic considerations driven by syntactic, semantic, and contextual hints, the latter including both linguistic and extralinguistic information. In oral communication, the picture can be complicated even further by the presence of nonverbal elements, such as gestures and facial expressions, whose mental representation precedes that of linguistic units (for an overview, see Tomasello 2008: chs. 1-3). Currently, some attempt to deal with situational context have been made by ECG scholars (Chang and Mok 2006; Mok and Bryant 2006), but these studies are still at the very beginning and they have to face a number of theoretical and computational problems, whose solution is likely to be found only in the very long term 77 (Feldman 2010; Feldman and Gilardi forthcoming). A similar observation holds for discourse context, whose handling is at least as difficult. We can nonetheless mention a remarkable achievement obtained by some scholars working within the ECG enterprise: the introduction of another conceptual primitive (besides schema and construction), labeled situation, developed in order to capture the functioning of mental spaces (taking Fauconnier 1994 as a reference text) and provide a formal representation for them. The notion of mental space78, quite influential in cognitive semantics, can be used to account for phenomena such as counterfactuals, conditionals, thought of others, reported speech, etc. Situations were successfully exploited to analyze some simple sentences (Chang et al. 2004; Mok et al. 2004; Feldman and Gilardi forthcoming), but further efforts are needed to make this primitive suitable to be used in a broader range of cases. Nevertheless, the fact that some steps have been taken in this direction is on its own positive, for it clarifies that ECG scholars are perfectly conscious of the importance of context in language understanding (and, consequently, of the current limits of the ECG model) and of the fact that scaling up the formalism to provide a satisfactory formal representation of discourse and situational context will be a demanding and time-consuming undertaking. However, being a team effort by its very nature (and strongly interconnected with advances in parallel projects underway within the NTL paradigm), ECG may gradually develop into a more robust model, inclusive of a (relatively) fine-grained representation of context in the following years. Indeed, if we take into consideration the fact that ECG is a recent approach, it has been growing fast during last years, obtaining remarkable results, and there is no reason to think that 76 Comprehension is influenced by context in all languages, to some extent. Nevertheless, the lack of context shows more dramatic consequences in some languages than in others. In particular, contextual information plays an especially crucial role in those languages which license pro-dropping, allow the productive omission of arguments, and cannot rely on morphological cues, such as Mandarin Chinese (Chang and Mok 2006; Mok and Bryant 2006). 77 Furthermore, It must always be kept in mind that the availability of information on discourse and situational context can vary according to the type of data being analyzed. 78 A brief and concise definition of mental space is provided in Mok et al. (2004): “Mental spaces (Fauconnier 1985) are partial cognitive structures built up during discourse that keep track of entities and relations in different contexts.” (emphasis original). 241 Grounding Meaning in Everyday Experience in the World this positive trend should not go on in the future, though the ECG formalism may undergo substantial, even dramatic modifications in order to reach a higher level of adequacy. With reference to future work, the first observation we find natural to make regards the fact that the present study may be taken as a starting point for further ECG investigation of phenomena concerning the Italian language. For instance, in the previous two sections we constrained our analysis to caused-motion constructions which show the presence of path PPs, but there are also caused-motion constructions which express path motion by means of adverbial phrases. Sometimes, this kind of phrases denotes motion in rather vague terms, without reference to a specific landmark, and such a phenomenon may posit different problems for their formal representation. Moreover, it may also be interesting to investigate different motion-events: for instance, Italian intransitive motion constructions may be analyzed in order to compare the different ways in which self-propelled motion and caused-motion are expressed in this language. Furthermore, it may also be interesting to work on phenomena of related languages and/or language variaties. Indeed, to be accurate, very little has been done on all Romance languages from an ECG perspective. Therefore, it may be interesting to investigate a corpus of another language of this family, for instance Spanish79 and select a sample of sentences suitable to be used as a reference to build sentence-level constructions similar to our Italian data. Then, these caused-motion constructions could be analyzed and a contrastive study of Italian and Spanish data may be carried out, in order to compare all the lexical and syntactic constructions which make up the lattices exploited in the analysis of caused-motion constructions in the two languages. Finally, the commonalities and divergences between Italian and Spanish caused-motion constructions may be discussed in detail. The same kind of study may be carried out with regard to other languages80. Another phenomenon whose investigation may be interesting is represented by figurative language, but ECG work in this direction is still very much “in progress”. Within the ECG program, metaphor has been addressed in a handful of papers (see e.g. Chang et al. 2002; Feldman and Gilardi forthcoming): taking Lakoff and Johnson's (1980) book as their reference point, ECG scholars have worked out a fourth conceptual primitive, labeled map, which serves to capture metaphorical mappings. Following Lakoff and Johnson, maps carry out this duty creating a link between two different schemas. Nevertheless, while some instances of metaphor have been successfully analyzed using maps (in particular, those taken from Narayanan's 1997 dissertation), a lot of work is still necessary to refine the primitive, in order to make it suitable for the analysis of a broader range of data. As a result, an ECG analysis of figurative language is possible to a limited extent, approximately in so far as work on mental spaces is. A more suitable target may be morphology, a branch of linguistics to which ECG has payed little attention until very recently (though Chang and Gurevich 2004 dedicated a presentation to Russian case morphology). However, since Schneider's (2010) study on Hebrew verbal morphology showed that the ECG formalism can be used to model morphosemantic phenomena satisfactorily, it may be interesting to extend the 79 Being Spanish an international idiom spoken in many different countries, it would be wise to search for a corpus which satisfies either of the following two requirements: it should focus on just one linguistic variety or otherwise it should be statistically well-balanced, in order to respect the proportion between the different varieties of Spanish. 80 A comparison between different languages of the same family may also serve as a starting point for diachronic studies. 242 3. An ECG analysis of Italian caused-motion constructions application of the ECG formalism to the analysis of morphological phenomena in other languages. As underlined in the previous sections, during the analysis of the last of our example sentences we had to shift from the lexical-syntactic to the morphological level, in order to deal with the possessive vostre, and this change of perspective did not have particular consequences on the overall analysis of the sentence. Of course, there are substantial differences between languages like Hebrew, which belong to the morphological type of introflexive languages, and languages like Italian, whose morphology is generally considered flexive, but these are essentially formal differences, which should not be difficult to adapt, given the fact that ECG grammars are languagespecific. Rather, the main threat to a thorough analysis of morphological issue may be related to the language user's consciousness of the status of the linguistic unit under consideration. Indeed, while we can be quite confident that words, phrases, and clauses have independent representations in the mind of the user, it is not that assured that the same condition holds for most morphemes. First of all, while the notion of “morpheme” is often taken for granted by linguists, most language users simply ignore it. In fact, people are generally used to segment words into prosodic units (syllables), but they often do not seem to pay attention to the distribution of meaning inside a word. We are not suggesting that people have no consciousness of all morphemes, which is clearly not the case, but simply that, lacking an explicit knowledge of this linguistic unit, they fail to recognize most of them as meaningful units. In other words, we suggest that people have an implicit knowledge of the morphemes they use most and those they often find in association with a specific domain of knowledge only81. This implies that the analyst willing to provide an account of morphological issues of a given language should be particularly careful in the choice of the phenomenon to investigate and, above all, in the analysis of the data collected, in order to avoid positing implausible mental representations, which would spoil the soundness of their work. In conclusion, the present chapter seems to confirm that the ECG formalism is suitable to provide a usage-based, cognitively motivated formal representation of the constructional analysis involved in the language understanding process, exploiting notions developed in cognitive semantics and cognitive approaches to grammar and endowing them with a computational dimension82. The result of this integration is a formal mechanism of representation based on a robust theoretical apparatus which grounds its roots in the heterogeneous world of (embodied) cognitive science, taking insights from different fields which range from experientialist philosophy to connectionist computational models of the brain. The ECG model allows us to account for the processing of a number of phenomena of language use, but since the long-term goal of the enterprise is ambitious and extremely demanding, there are several various problems which are still to be addressed properly. Indeed, the remarkable results obtained by the scholars involved in the project can just be considered as first little steps moved toward a detailed representation of the mental mechanisms at work during the process of language comprehension. 81 This condition would favor the view brought forward by some functional and cognitive linguists (see e.g. Croft 2001, 2007; Cristofaro 2008), according to whom the larger construction represents the primitive unit of meaning, while the meaning of its components can (but need not) be derived by means of abstraction. 82 In general, cognitive linguists have always rejected to supply a formal representation of their models, on the basis of the conviction that an attempt to formalize cognitive processes would be either too difficult and time-consuming or simply doomed to failure (see e.g. Langacker 2007). 243 4. Conclusion: ECG and beyond 4. Conclusion: ECG and beyond The main point of the present study was to carry out an analysis of a circumscribed phenomenon of the Italian language, adopting the perspective of Embodied Construction Grammar, a computational formalism designed to provide a precise representation of the semantic content of linguistic expressions, in order to supply a detailed description of the process of online language comprehension carried out by individuals involved in acts of verbal communication. Nevertheless, to be more accurate, the enterprise undertaken here can be characterized at two different levels of scientific interest. More precisely, in ch. 1 we offered an overview of the present study underlining the fact that it could be described as showing a double-layered nature, a nested structure, and also several connections between the two. We also suggested that the reader should approach this piece of work keeping this point in mind, for this kind of duality would be a recurrent theme throughout our study as a whole, though usually not explicitly mentioned. It is no surprise, therefore, that the two main chapters of the present thesis performed two different, yet deeply intertwined functions. Ch. 2 provided the reader with a detailed account of the theoretical background which underlies the choice of the ECG model, a relatively recent approach whose development is still very much under way1. Indeed, as mentioned in §2.3.2, the epistemological roots of ECG represent the ultimate reason for picking this particular perspective. In other words, in order to understand the motivation for choosing the ECG model instead of others, it is necessary to take into consideration a much broader scientific framework developed during the last decades. Ch. 2 represents our attempt to situate ECG in a broader enterprise in the world of linguistics and cognitive science. First of all, it was clarified that our work is aligned with the emergent paradigm of Embodied Cognition, which rejects the basic tenets of Mainstream Cognitive Science (§2.2.1). While the dominant theoretical framework takes cognition to be the result of the intracranial process of manipulation of arbitrary amodal symbols and considers the human mind as a device similar to a piece of computer software running in the hardware, theorists of embodiment see cognition as the result of the constant interaction between organism and environment, thus arising from the interplay between the structure of our brains and bodies and our everyday bodily experience in our physical and sociocultural world. Therefore, at the very core of the present study lies the belief that the body and the mind are strictly interdependent, rather than being two separate, autonomous systems. Since Embodied Cognition is a highly fragmented world which covers a number of different approaches, we specified that our particular reference point is to be found in Gibbs' (2005) book, whose position on Embodied Cognition is mostly, though not completely, overlapping with the point of view of advocates of the “philosophy of embodied realism”, a perspective which has been largely influential during the last decades (see e.g. Lakoff and Johnson 1980, 1999; Rohrer 2007). Along with our commitment to the tenets of Embodied Cognition, in ch. 2 we specified that our work aligns itself with the Cognitive Linguistics tradition (§2.1.2). In fact, it was argued that the Cognitive Linguistics enterprise and the Embodied Cognition 1 Indeed, even within the family of cognitive approaches to grammar, ECG is often relegated to a position of secondary importance. See e.g. Croft (2007: footnote 3). 245 Grounding Meaning in Everyday Experience in the World paradigm are deeply related, and it is no suprise that these frameworks developed in a somewhat parallel way. As a matter of fact, both programs emerged as a reaction against the dominant perspectives in linguistics and cognitive science, namely Generative Grammar and Mainstream Cognitive Science, which are in turn causally linked. Indeed, Chomsky's theory of language can be seen, in a sense, as the application of the assumptions of Mainstream Cognitive Science to linguistic theory: in a nutshell, the view of the mind as a device for the manipulation of abstract symbols led to the view of language as a formal system, made up of syntactic rules operating on semantic primitives. The rejection of this characterization of language by the side of cognitive linguists can be seen as parallel to the rejection of the view of the mind as a computer on the part of the theorists of Embodied Cognition. As a result, during the last decades it has been possible to observe an increasing intersection between scholars working in the two fields, and nowadays Embodied Cognition and Cognitive Linguistics can mostly be seen as mutually informative. Pushing this observation one step further, we might even say that these two paradigms can be seen as mutually constitutive, since more and more often Embodied Cognition is seen as a pillar of the theoretical building of Cognitive Linguistics, and at the same time Cognitive Linguistics is a remarkable source of data for embodiment theories. In the light of our commitment to these two research programs, Embodied Cognition and Cognitive Linguistics, it follows that from our perspective language is considered an emergent system, which arises from domain-general principles, rather than an innate, autonomous faculty. Moreover, while more traditional approaches view grammar as a system of formal rules to combine semantic primitives, we take the opposite stand, awarding meaning a central role in linguistic organization. Meaning is seen as arising from bodily experience in the world, then being shaped and constrained by the structure of the human body and its interactions with the (both physical and sociocultural) environment. As a result, language is considered to be learned from actual usage events: sufficiently frequent units of language will be stored in the learner's mind. Items showing a high type frequency will also give rise to abstract, schematic patterns, which will then be used productively to create novel expressions. Therefore, we reject the traditional characterization (generally advocated in generative approaches) of language as a system comprising a list of words and a set of rules to put them together; rather, we see language as a structured inventory of form-meaning pairs of different size, considering syntax and the lexicon as forming a continuum. Carrying out the present study, we focused in particular on the research work developed within a specific research program, the NTL paradigm, an interdisciplinary effort which involves the work of several scholars, most of which based in California (§2.2.1). The NTL program represents an enterprise committed to the theoretical assumptions of Embodied Cognition and Cognitive Linguistics and whose main goal is to uncover the neural correlates of the human manifest behavior, with particular reference to language. The various projects underway within the NTL program aim to integrate insights from different academic disciplines, especially linguistics, psychology, neuroscience, and computer science. Particularly significant for the scope of our study is the theoretical motivation and experimental support provided to simulation semantics, i.e. the idea that the comprehension of a linguistic construction involves the active simulation of its content (§2.2.2). From this perspective, understanding a sentence such as Henry kissed Sara involves the recognition of all the constructions involved in the sentence, the binding of their relevant roles, and the internal (mental) simulation of the 246 4. Conclusion: ECG and beyond scene as a whole on the part of the understander. Therefore, according to the simulation hypothesis, grasping the meaning of a sentence like Henry kissed Sara will involve the activation of (at least part of) the same neural circuitry which is active when the current understander is actually engaged in the action of kissing somebody2. The notion of simulation semantics is crucial for ECG, which can be seen as (basically) a simulationbased model of language understanding. Another aspect of the NTL paradigm which is important for the present study is the massive exploitation of computational models to simulate human cognitive processes (§2.2.3). This is a crucial point because ECG makes use of a computational formalism, which aims to reach two goals: on the one hand, it was developed to provide a detailed description of the process of online language understanding in the most precise manner possible; on the other hand, it was also thought to be implemented on machines. The implementation of models on computers has often proved useful in cognitive science, and though it does not provide ultimate evidence for the reliability of a theory, it can supply useful hints on its plausibility. While we did not implement any grammar on machine, insights from several studies carried out using some probabilistic and connectionist computational models provided us with some direct feedback on the plausibility of some hypothesis of ours on the online comprehension of some Italian clauses. At the same time, some studies carried out making use of some connectionist computational models allowed us to explain to the reader how inferences are supposed to spread in ECG. Before providing the reader with an outline of the ECG model, we found it relevant to briefly introduce the common characteristics of cognitive approaches to grammar, also sketching a summary of the main points on which these approaches differ from the mainstream generative model, and to outline three among the most developed and influential of these models which exerted a considerable influence on the design of the ECG approach: CG, CxGx, and RCG. Several concepts exploited in ECG are borrowed (and sometimes adapted) from these three models. Together with a brief summary of the main points which characterize these models, we highlighted some notions which are commonly exploited in ECG. Among these, we can mention Langacker's notion of “profiling”, which has a paramount importance in ECG since it specifies from which perspective a certain event has to be simulated. Also, the choice to maximize the number of constructions at expenses of verbal polysemy derived from CxGx is important, since it avoids to posit implausible senses for verbs found in several different patterns3. Moreover, some suggestions from RCG were also approved of, such as the application of the label “construction” to all pairings of form and meaning and the idea to consider grammatical relations as both language- and construction-specific. Coming to a characterization of the ECG model proper, it was stated that, while this model is part of the family of cognitive approaches to grammar and share the basic tenets of Cognitive Linguistics, there are some divergences between ECG and other cognitive models. These differences are prominently motivated by the fact that ECG 2 It is important to clarify that the simulation of the corresponding passive sentence would be remarkably different: Henry kissed Sara will be simulated from Henry's perspective, thus involving the same structures active when the current understander kisses somebody, whereas Sara was kissed by Henry would be simulated from Sara's perspective, then involving the structures active when the understander is kissed by somebody. 3 It is true, as pointed out by e.g. Broccias (2006), that positing a cut-off point between lexical and constructional meaning is somewhat arbitrary. Nevertheless, it is relevant to consider that using a formalism requires to look for a balance between precision and flexibility which can sometimes lead to draw rather artificial boundaries. 247 Grounding Meaning in Everyday Experience in the World attempts to model online language processing, thus showing a certain number of peculiar theoretical and methodological features. First of all, ECG is more deeply related to the tenets of Embodied Cognition, since it assumes that all our linguistic and conceptual knowledge is grounded in our neural structures and arise through bodily experience and constant interaction in and with our environment. Adopting an ECG approach, language is studied from the point of view of the connections between mind and enviroment, relying on a remarkable amount of evidence obtained by research in different fields (see Feldman 2006). Focusing on online language processing, ECG aims to supply an analysis of the stages involved in the comprehension of linguistic units and to unravel the nature of the embodied knowledge with which the system interacts, rather than modeling linguistic knowledge itself (see Evans and Green 2006: §20.3). According to ECG scholars, indeed, meaning is represented by a series of parameterizations of the speakers' perceptual and sensorimotor representations, which constitutes a dynamic complex. Such schematizations are made possible by a process of generalization which selects the most important aspects of actions and percepts required to perform a simulation. These important aspects are then expressed in relatively discrete linguistic units. ECG aims to describe the process of online language comprehension through a process of analysis, which consists in the recognition of all the constructions instantiated in a message. The process of analysis supplies parameters for the simulation of the content of the message, which is performed in the following phase, labeled enactment. In order to provide a plausible representation of online language processing, the ECG model makes use of a formal language, which is designed in order to express the semantic structure of constructions explicitly, unifying the roles of the relevant schemas involved. The ECG formalism, whose functioning was preliminary introduced at the very end of ch. 2, was exploited in ch. 3 to provide an account of a circumscribed phenomenon of the Italian language, namely caused-motion constructions involving verbs of forceexertion. Eight declarative sentences (four active and four passive) were selected from an amount of data extracted from a large corpus of written Italian based on the web, and adapted according to some, rather strict, syntactic and semantic constraints. Then, these sentences were analyzed making use of an ECG grammar based on that adopted by Dodge and Bryant (forthcoming), slightly modified in order to cope with the peculiarities of the Italian language and to be more in line with Langacker's network model of polysemy. The starting point of our case-study rests on two basic hypotheses: 1) a single A-S construction would be able to capture generalizations about prototypical Italian caused-motion constructions, while possible radial extensions might be posited to deal with less prototypical constructions, and 2) a limited set of schemas would be able to account for the central meaning of Italian caused-motion constructions. The adoption of the ECG formalism allowed us to provide a representation of the several constructions involved in our sentences, at three different levels of analysis: lexical, phrasal, and clausal, permitting us to speculate, with a reasonable degree of confidence, on the mental processes active during the activity of understanding an utterance. The two main parameters of analysis of our case-study have to do with compositionality and polysemy, two issues which can hardly be kept apart when dealing with language processing. Compositionality was approached from a usagebased perspective: a construction is considered to be stored via entrenchment when it occurs with a high degree of token frequency, and it is considered productive when it 248 4. Conclusion: ECG and beyond appears with a sufficient level of type frequency. In this way, the meaning of constituent constructions (e.g. words) is abstracted from the meaning of the larger construction (e.g. a clause) they appear in; then, once their meaning and context of use is stored, they can be used in novel ways and appear in other larger constructions. In other words, we posit a constant interaction between the component constructions and the composite construction: the meaning of the whole construction is seen as the result of the interplay of the component constructions, which fill the slots in the larger construction and provide bottom-up cues, and the larger construction itself, which provides top-down constraints, restricting the range of its possible constituents. This point of view seems favorable when dealing with language understanding, since utterances are considered to be processed incrementally, through the interpretation of partial structures and expectations for future input, in a simultaneously bottom-up and top-down manner. In this process, polysemy plays a crucial role, for most words in a language displays more than one meaning. This issue was dealt with adopting Langacker's (1987) network model, according to which the senses of a linguistic unit represent the nodes of a network, linked horizontally by relations of similarity and vertically by a sort of “subcase” relation, with the lowest nodes representing the most specific senses of a unit (which are the first to arise, consistent with the usage-based approach), while the highest node represents the most abstract representation of the meaning of a unit (which is the result of a later process of abstraction). The adoption of a simultaneously bottom-up and top-down approach allowed us to capture the interplay between compositionality and polysemy in the process of understanding an utterance4. The results of the present study can be evaluated, in a sense, at two different levels. At a micro level, we can draw some conclusions about our analysis of Italian causedmotion constructions illustrated in ch. 3, attempting to assess the adequacy of our grammar and, more generally, of the ECG model to deal with Italian data. At a macro level, on the other hand, we can propose some observations on the nature of the ECG model overall, considering its function within the NTL paradigm and, more generally, in the world of embodied approaches to cognitive science and linguistics. At both levels, some points were already raised in §3.4, but in the following pages we are going to discuss some additional issues, trying to situate our concluding remarks in a wider context. Let us begin with a couple of observations at the “micro” level. With regard to the analysis of the Italian data we carried out in the previous chapter, it seems that we can overall be satisfied: the ECG grammar we built and exploited in order to provide a detailed representation of all the constructions coming into play during the online processing of our sentences seems to work well. Using Dodge and Bryant's grammar as a reference point, the development of a similar grammar able to account for Italian data was relatively straightforward. As a matter of fact, while the realization of an ECG grammar to explain an observed phenomenon in a given language may at first seem a rather exacting effort, the compromise between precision and flexibility on which the model rests makes it feasible to build such a grammar without too much of an effort. As already mentioned, the representation of Italian lexical and syntactic constructions was not particularly difficult, with some notable exceptions like the possessive vostre and some vague nouns and adjectives, which resulted remarkably more time-consuming. Overall, the unification requirements of the various constructions were met, showing 4 This approach is supported by a range of studies carried out making use of probabilistic models (see e.g. Narayanan and Jurafsky 2001). 249 Grounding Meaning in Everyday Experience in the World that the ECG formalism can be extended to analyze Italian data with no particular problems. Moreover, our study tried to explicitly deal with the problem of lexical polysemy, in the case in point with reference to prepositions, nouns, verbs, and adjectives, making use of an adapted version of Langacker's network model. Polysemy is an issue that, while generally recognized, has most often been taken for granted and seldom directly dealt with in ECG studies. Nevertheless, since in our view an account of language processing could hardly be effective without a representation of lexical polysemy, we found it relevant to address this topic and show its intersection with compositionality, specifying that it plays an important role in the interaction between the composite construction and its constituent constructions. While this may represent a strong point of our analysis, on the other hand the solution we adopted may cause the reader to raise some doubts. Indeed, if we look at the highest nodes of the network, we shall see that all verbs show the same meaning, connected to the Process schema, differing from one another just in terms of their form. An analogous situation can be observed in regard to nouns (RD schema), adjectives (Modifier schema), prepositions (Relation schema), etc.: what are traditionally labeled “parts of speech” at their highest node have no specific meaning, exactly sharing the schematic meaning of their grammatical category, being distinguished from it just on the basis of their phonological form. Compare the two constructions in Fig. 4.1 (repeated from Fig. 3.3) and Fig. 4.2 below: Fig. 4.1: the Lanciare construction construction Lanciare subcase of LexVerb constructional: VerbFeatureSet // inh. constraints self.features.verbform ← base form: WordForm // inh. constraints self.f.phon ← /lan'ʧare/ meaning: Process // inh. Fig. 4.2: the Posare construction construction Posare subcase of LexVerb constructional: VerbFeatureSet // inh. constraints self.features.verbform ← base form: WordForm // inh. constraints self.f.phon ← /po'zare/ meaning: Process // inh. The reader will have noticed that the only difference between the two constructions above has to do with their particular form. Now, compare both of these constructions with their superordinate construction shown below: 250 4. Conclusion: ECG and beyond Fig. 4.3: the LexVerb construction construction LexVerb subcase of Verb constructional: VerbFeatureSet // inh. constraints self.features.verbform ← base form: WordForm // inh. meaning: Process It is quite straightforward to notice that the only feature which distinguishes this construction from those illustrated in Fig. 4.1 and Fig. 4.2 is the absence of form constraints. It may seem unreasonable to posit that even the most schematic characterizations of polysemous verbs such as lanciare or posare are represented in the speaker's mental grammar as having as little meaning as the category “lexical verb” (i.e. just the fact that these verbs denote processes). We admit that it is indeed somewhat implausible: if we asked an Italian speaker to explain the meaning of a verb like lanciare (out of context), they will probably pick a particular sense (most probably, “to throw”), rather than answer that they need some context in order to choose the right sense. Nevertheless, it does not follow that our characterization of the highest node of the network as being almost semantically empty is untenable. As a matter of fact, there appear to be a couple of reasons why our solution may not be disregarded. The first of this reasons has a theoretical nature: we mentioned in §3.1 that speakers do not need to access the most schematic node of the network whenever they happen to encounter a particular linguistic unit; in fact, there is experimental evidence that they can directly pick the relevant sense which is required in the specific context, without having to review all the possible senses of the item. As a result, positing the existence of constructions like those illustrated in Fig. 4.1 and Fig. 4.2 above does not imply that speakers usually access them when processing language; nevertheless, these constructions are useful to highlight the vagueness of polysemous verbs out of context. The second reason is a more practical one: these constructions serve as “linking nodes” in the network between the generical LexVerb construction, which has no form constraints, and the various senses a polysemous verb can take in different contexts. A related point is represented by the formal representation of semantic content. While the notation of the ECG formalism provides us with tools which allowed us to represent the main features of the semantic pole of most of our constructions, the situation is rather different for the meaning of auxiliary verbs. As a matter of fact, while it is quite easy to denote the meaning of a lexical verbs exploiting the x-net role of the Process schema, the same condition does not hold for auxiliary verbs. In our study presented in ch. 3, we followed Langacker claiming that the traditional view according to which auxiliary verbs are merely empty grammatical markers should be rejected, in favor of a perspective which recognizes that “function words” are inherently meaningful. Indeed, in Cognitive Linguistics auxiliary verbs are said to serve a grounding function for the linguistic utterance, linking it to the time and place of the conversation, the participants in the exchange, and their shared knowledge. Nevertheless, binding the meaning of auxiliary verbs to a Function schema, whose 251 Grounding Meaning in Everyday Experience in the World function role is filled by the value grounding might be (at the very least) a bit too vague. Unfortunately, it is by no means obvious how to find a satisfactory solution to the problem. To our knowledge, a finer-grained (formal) characterization of the semantic structure of auxiliaries is not currently available in the ECG camp. In any case, we should point out that this is not a structural issue but only a temporary solution, which is likely to be replaced with a better one as soon as it is worked out. The vague solution chosen in our grammar is of course due to the complexity of the problem, but is also tightly linked to the current deficiency of a better option in all the ECG grammars we are aware of. This point suggests that we should temporarily shift from the “micro” to the “macro” level of discussion, casting a glance at the ECG formalism more generally, trying to weigh the pros and cons of the use of a formal model to give an explicit representation to the notions developed in the Cognitive Linguistics literature. Overall, as observed above with regard to our Italian data, the application of the ECG formalism to the analysis of several linguistic phenomena can be seen as satisfactory: the ECG model allows the analyst to carry out a quite detailed analysis of the mechanisms involved in the process of understanding of linguistic utterances, accounting for a range of phenomena by means of a series of structures carefully built in order to be consistent with both the bulk of notions carried out within the Cognitive Linguistics enterprise and the results of research from other disciplines. One of the main assets of the ECG formalism is that it provides Cognitive Linguistics with a computational dimension, attempting to integrate different insights about the structural and semantic properties of language (Dodge and Bryant forthcoming), also allowing analysts with advanced computer skills to implement this collection of integrated insights on machine. To reach this goal, the formalism is designed in order to satisfy two distinct (and, perhaps, somewhat competing) requirements at the same time: the need for precision, which is necessary in order for the formalism to be a reliable and useful tool, and flexibility, needed to enable the analyst to build the best grammar to analyze different phenomena from different languages. Meeting these two conditions at once represents a remarkable success, making the formalism a quite powerful instrument for the analysis of linguistic data. Nevertheless, there are still a couple of drawbacks which might (although to a minor extent) “spoil the picture”. First of all, an inherent limitation of the model is represented by the fact that, for all its flexibility, no linguistic formalism can perfectly simulate the dynamic and extremely context-sensitive process of language understanding in the speaker's mind, as argued in Feldman et al. (2009). Besides this obvious limitation, we also consider relevant to remind the reader that the actual relation between verbal and A-S constructions is probably less clear-cut than is represented in the formalism, which generally tend to follow Goldberg in minimizing lexical polysemy, positing instead a large number of different constructions (see Langacker 2009); probably, this is again an intrinsic problem of all formalisms: several aspects of language can be seen as continua, and this continuous (rather dynamic) nature is very difficult to be captured in formal terms5. Furthermore, the high level of flexibility of the ECG formalism can make the analyst face some doubts regarding the architecture of their grammar. As a matter of fact it is not too rare to find out that more than one possible lattices can be posited in order to deal with a particular linguistic phenomenon. In general, this is a positive side of the model, allowing the analyst to choose among 5 Nevertheless, taking into account how many steps ECG scholars have taken in the last few years, it is not implausible that some improvements will be eventually achieved also in this respect. 252 4. Conclusion: ECG and beyond different solutions; nonetheless, while often the benefits of a possible grammar are clearly superior to those of its competitors, sometimes the pros and cons of the alternative grammars are roughly equivalent; therefore, making a choice may become quite difficult and time-consuming (as asserted in Bergen and Chang 2005). Although any linguistic formalism has some intrinsic limitations, we have to remark that having the ECG formalism at our disposal was a valuable asset, since exploiting a tool designed to provide a formal representation of mental processes (loosely) inspired by the neurobiological structure of the brain makes it possible to investigate the phenomenon of language processing resting on relatively solid basis, which go beyond the mere theoretical speculation. As a matter of fact, the ECG model fits with the other projects developed within the NTL paradigm, which attempt to some extent to simulate the functioning of the brain, building on what is known from academic disciplines such as neurobiology and neuroscience, along with psycholinguistics and various strands of computational modeling (§2.2). The results of psycholinguistic experiments and computational simulations carried out using probabilistic models can also be used as help to carry out the constructional analysis of utterances, making use of their results as hints in order to find the most likely interpretations of these linguistic units on the part of the language user engaged in the activity of processing linguistic information. Indeed, ECG together with other projects developed within the NTL program represents a noteworthy effort to integrate the theoretical and empirical background of Cognitive Linguistics with the massive potential of computer science in order for the two disciplines to benefit from the assets of each other: on the one hand, the cognitive linguist can exploit computational tools in order to gain formal precision and robustness, and can also draw inspiration from the performance of computational models of cognitive processes6 as insights to be exploited in their further studies; on the other hand, the computer scientist can take cues from the work of cognitive linguists to improve their computer simulations. Another important aspect of ECG which seems appropriate to emphasize is the fact that, though embodied theories of meaning are often charged with the accusation of being exclusively interested in the intrapersonal dimension of language disregarding its social nature, since its background is rooted in the work of scholars who are mostly interested in the relation between language, mind, and brain, ECG cannot be convicted on this charge: the fact that most work on the ECG model, as more generally on the several projects which comprise the NTL paradigm, focuses on the neurological and psychological aspect of language, does not imply that ECG scholars disregard the importance of its intersubjective dimension. Indeed, the scholars who work on ECG have a clear idea of what shared, sociocultural meaning is, as recently emphasized by Feldman (2010: 22), “Shared meaning is partial and is a social construction of a language community based on similar genetics and experience” (italics original). The recognition of the importance of the social aspect of language is manifest in the assumptions of the ECG model itself. Indeed, a foundational concept of ECG is the notion that language evokes rather than represent ideas. That is, linguistic expression is only a conventionally schematic map of thought: in verbal communication, a message is conveyed even though most details of the information are left unexpressed. As Feldman points out, everybody is member of several language communities, which can be seen as lattices with partial overlaps (we speak differently in different situations). In ECG, 6 The reader is referred to Christiansen and MacDonald (2010) for a recent demonstration of the use of connectionist models in the field of linguistic studies. 253 Grounding Meaning in Everyday Experience in the World the meaning pole of constructions is expressed in terms of skeletal versions of universal and cultural experience – language is considered inseparable from culture. As already explained throughout the present study, the ECG model assumes two processes, analysis and enactment. The former, which is the one we focused on in the previous chapter, precisely reflects the social aspect of language, since the analysis process converts an utterance in context to what is called the SemSpec, which is constant across a linguistic community. Only then, the enactment process (people understands language by simulating its content), which depends entirely on the particular beliefs of the individual, is run. In Feldman's (2010: 23) words, “The NTL/ECG separation of language understanding into analysis and enactment phases provides a clean and well motivated distinction between the shared meaning of an LC and embodied individual meaning” (italics original). This point deserves to be highlighted because it serves to clarify that the accusation of “physico-biological reductionism” often directed to embodied theories of language and cognition can (at least with regard to ECG) reasonably be rejected. In summary, the ECG model can currently be considered as a valuable tool for the investigation of the phenomenon of online language processing and the study of the relation between the formal structures of a given language and the amount of conceptual, embodied knowledge they schematically evoke. Chang (2008) showed that the formalism can also be used as the basis to build more precise models of facts related to phenomena such as child language acquisition. It must nevertheless be kept in mind that ECG is still in its infancy, and while it can allow the analyst to carry out a satisfactory analysis of several linguistic phenomena, the formalism has to be improved in order to be able to deal with a greater variety of language facts. As a matter of fact, at present, the ECG formalism has been employed mainly to provide a representation of the processing of literal language expressions describing quite basic scenes (see e.g. Bergen and Chang 2005; Feldman et al. 2009; Feldman 2010; Dodge and Bryant forthcoming), with the exception of more specific studies dealing with particular phenomena such as quantification and classification (e.g. Dodge and Wright 2002), the behavior of language-specific constructions (e.g. Bretones Calleja et al. 2001; Ettlinger 2005), or more abstract processes such as conceptual representations (e.g. Bretones Calleja 2004). While these studies represent necessary attempts to account for certain aspects of human language, together they currently cover the facts of language in a somewhat piecemeal, nonhomogeneous fashion, primarily devoting their attention to specific phenomena at the expenses of others. As a consequence, it is still early for ECG to attempt to provide a more comprehensive picture of the grammatical system of a language. Nevertheless, the ECG enterprise looks quite promising: in §3.4, we listed some of the issues which are on the agenda of ECG scholars for the near future, and whose coverage is currently under way. We mentioned the fact that the formalism should be extended to provide a unified representation of both central cases and extensions (adopting the “radial category” perspective of Lakoff 1987, Goldberg 1995, and Dodge and Bryant forthcoming) of grammatical constructions, and also pointed out that some initial steps in this direction have been taken. While this objective is rather ambitious, its achievement seems possible in the long term, and it would be really important, for it would endow the ECG model with a significantly stronger descriptive and explanatory power. Moreover, we mentioned the fact that ECG scholars are working on improvements which would allow the formalism to deal with cases of 254 4. Conclusion: ECG and beyond figurative language and to include discourse and situational context in the analysis. Again, this is an exacting goal to be reached, but some significant steps in this direction have been moved, and though the solution to this problem is unlikely to be found in the short term, it seems feasible to be reached in the long term. This would be another impressive improvement which would increase the power of the formalism, allowing the analyst to deal with a remarkably broader range of linguistic utterances. After the brief evaluation of the (current) assets and liabilities of the ECG enterprise supplied above, we find it appropriate to conclude the present study going back to our own analysis of Italian data and see in what direction future work might be functional in order to take both the work developed in this study and more generally the ECG model (thus encompassing the “micro” level as well as the “macro” level of discussion) one step further, permitting research to advance in a direction which might allow ECG to approach and intersect other emergentist approaches to the study of language and cognition in order to join the attempt to develop a more unified paradigm able to rise as a strong alternative framework (the foundation of such an enterprise has already being laid, see The Five Graces Group 2010) to the standard nativist/generative tradition. As already specified in §3.4, the case study illustrated in ch. 3 focused on a very circumscribed phenomenon, making use of a small quantity of data, which are, all in all, rather homogeneous, for they were selected on the basis of quite strict parameters, based on severe constraints. This can be seen as an important limitation of our study, since it seriously constrains the range of conclusions which can be drawn to a few generalizations over the specific phenomenon, i.e. Italian caused-motion constructions involving verbs of force-exertion. Nevertheless, carrying out a rather satisfactory analysis of this phenomenon can be taken as an encouraging indication that the exploitation of the ECG model to investigate Italian data is feasible and may be fruitful. It is from this viewpoint that we interpret the results of our study (as mentioned at the end of ch. 3): as a first (little) step toward a systematic investigation of the online processing of phenomena of the Italian language from a constructionist perspective centered on the embodied nature of human cognition. Since there are currently no other ECG studies on Italian data available, in the conclusion of the previous chapter we underlined the fact that our study may serve as a starting point for further studies on the grammatical phenomena of this language and, to some extent, on other Romance languages, by now a bit neglected among the community of ECG scholars (even though some work has been carried out on Spanish, see e.g. Bretones Calleja et al. 2001), also foreseeing the possibility to carry out contrastive studies between different languages of the same family. Adopting a broader perspective, we may also propose that a contrastive analysis of the same phenomenon in different languages might spread some light with regard to the relation between two fundamental characteristics of conceptualization (and, consequently, cognition): its bodily basis and its experiential basis. From this point of view, working simultaneously on two or more closely related languages and some more remote languages may be functional to this scope. In a sense, a crosslinguistic work of this kind might approach the typological quest for linguistic universals, exploiting the ECG formalism to explicitly represent the commonalities and differences in language processing between speakers of different languages, trying to supply some relevant hints on the possible conceptual universals which underlie the process of language comprehension, quite similarly to the work carried out by Regier (1996) briefly summarized in §2.2.3. In other words, it may represent a little contribution to the 255 Grounding Meaning in Everyday Experience in the World research area which aims to distinguish the properties of human language and cognition which are due to our “(socio)cultural embodiment” (see e.g. Gibbs 2005: ch. 1; Maalej 2008), and thus depend on the particular sociocultural setting we are situated in, from those which simply stem from our “physical embodiment”, i.e. the mere fact that, as human beings, we happen to have a certain (neuro)biological structure, which imposes a determined way to perceive the enviroment around us. It is in the light of this kind of perspective that the explicit formal and semantic representation and the computational potential of the ECG model (and, more generally, the NTL research program) might fruitfully approach other projects which take part in the rising paradigm which conceives language as an emergent system and aims to challenge the dominant nativist approach, calling into question its assumptions and developing a strong alternative framework. The theoretical assumptions and the methodologies of empirical investigation of this alternative framework, whose manifesto is represented by the Five Graces Group's (2010) position paper7 and the other articles collected in Ellis and Larsen-Freeman (2010a), are largely overlapping with those which underlie the ECG enterprise and the NTL program. Proponents of this approach, on the basis of the results of research carried out in various disciplines which focus on human cognition, assert that the structure and knowledge of language are shaped by processes of human interaction along with domain-general cognitive processes. From this perspective, processes of language acquisition, use, and change are all affected by patterns of use and represent facets of the same system, which is then best characterized as a complex adaptive system (CAS henceforth). This construal of language shows the following key features: • the system involves multiple agents interacting with each other; • the system is adaptive: the speakers' behavior is based on their past interactions, and current and past interactions together feed forward into future behavior; • a speaker's behavior is determined by competing factors, ranging from perceptual mechanisms to social motivation; • the structures of language derive from interrelated patterns of experience, social interaction, and cognitive processes. The members of the Five Graces Group (2010) affirm that this conception of language allows for a unified account of seemingly unrelated linguistic phenomena (including variation, the probabilistic nature of linguistic behavior, the continuous change within agents and across communities, the emergence of regularities from the interaction of agents in language use, and the transitions due to nonlinear processes). Moreover, they argue that such an approach permits the analyst to reveal commonalities in various areas of linguistic research (cognition, sociolinguistics, L1 and L2 acquisition, psycholinguistics, historical linguistics, and language evolution). Finally, this approach stands on converging evidence from a plurality of methodologies (corpus analysis, crosslinguistic studies, grammaticalization, psychological and neuroscientific experiments, and computational and/or mathematical modeling). According to proponents of the CAS approach, language is shaped by human cognitive abilities (e.g. categorization, sequential processing, planning) and is used for social interaction, “and so its origins and capacities are dependent on its role in our social life” (Five Graces Group 2010: 4). From this perspective, language evolved in the 7 The Five Graces Group is an interdisciplinary research team constituted by the following ten scholars (in alphabetical order): Clay Beckner, Richard Blythe, Joan Bybee, Morten H. Christiansen, William Croft, Nick C. Ellis, John Holland, Jinyun Ke, Diane Larsen-Freeman, and P. Thomas Schoenemann. 256 4. Conclusion: ECG and beyond context of an already highly interactive social existence, as the result of the interplay of different constraints (including the structure of thought processes, perceptuo-motor biases, cognitive limitations, and sociopragmatic factors). Therefore, more generally, not only language but also culture can be considered as emergent phenomena of an increasingly complex social existence8. Language variation and change are seen as the result of the indeterminacy of communication, since language operates at the following four levels: • producing and attending to the utterance; • formulating and identifying the proposition; • signaling and recognizing the communicative intention; • proposing and taking up the joint interaction. This complex model is nevertheless fragile, as shown by the fact that misunderstandings are a common occurrence. The fragility of the system leads to variation, which is the substrate of language change9. Advocates of the CAS approach adopt a usage-based theory of grammar: grammar is seen as a network built up from categorized instances of language use. Constructions (form-meaning pairs) are the units of language and vary in size and levels of specificity, and grammar contains details of co-occurrence and a record of the probabilities of occurrence and co-occurrence. As stated in the Five Graces Group's (2010: 5) paper, “The evidence for the impact of usage on cognitive organization includes the fact that language users are aware of specific instances of constructions that are conventionalized and the multiple ways in which frequency of use has an impact on structure”. Cognitive changes are seen to occur in response to usage and contribute to the shape of grammar. We are invited to consider the following three phenomena (Five Graces Group 2010: 6): • communication largely consists of prefabricated sequences rather than an “open choice” among all available words; • as words co-occur in speech, they gradually come to be retrieved as chunks (e.g. at the syntactic level, frequent word combinations become encoded as chunks that influence how we process sentences online); • historical changes in language point toward a model in which patterns of occurrence can be taken into account. This detailed knowledge of the interactions of grammar and the lexicon in usage, which includes knowledge of which words commonly co-occur with which syntactic patterns, leads to the conception of lexicon and grammar as deeply intertwined rather than separate. The cognitive representations underlying language use are built up by the categorization of utterances into exemplars and exemplar clusters based on both their linguistic form and their meaning and the context in which have been experienced. Because this categorization is ongoing during language use, adult grammars as well as infant grammars are not fixed and static but have the potential to change as experience changes. General properties of language are seen as coming about through the repeated 8 Since social complexity is correlated with brain size across primates, we are also pointed out that brain size evolution should provide us with some general clues about the evolution of language (see Schoenemann's 2010 contribution). 9 The Five Graces Group's (2010: 5) paper list four reasons for the fragility of the communicative system: people's impossibility to read each other's mind; the looseness of convention as a coordination device; the fact that the choice of words and constructions is based on prior use of these conventions in similar situations, but the hearer's knowledge of prior uses of the conventions is not the same as the speaker's; the uniqueness of the new situation being communicated and its subjection to different construals. 257 Grounding Meaning in Everyday Experience in the World application of general processes of change. Since the same processes happens in all languages, general resemblances develop; nevertheless, the trajectories of change (e.g. paths of grammaticalization) are much more similar than the resulting states. Scholars adopting a CAS approach to language are thus interested in emergent generalizations across languages, specific patterns of use as contributors to change and indicators of linguistic representations, and the cognitive underpinnings of language processing and change. The sources of data for the development of usage-based grammars include corpus-based studies, experimental studies, and modeling10. As the reader will have noticed, there is a lot in common between the ECG model and the NTL paradigm as a whole and the broader perspective of language as a CAS. First of all, the theoretical underpinnings of the former, related to the world of Embodied Cognition and Cognitive Linguistics, can be seen as representing a subset of those of the latter, whose scope also covers a finer-grained, more detailed investigation of social dynamics and whose assumptions also draw inspiration from other usagebased and functionally-oriented approaches to the study of language, such as Systemic Functional Linguistics (see Matthiessen's 2010 article), and also include insights from Dynamic Systems Theory (see Dörnyei's 2010 contribution), one of the most radical approaches available in the camp of embodied cognitive science. Second, the same can be said with regard to the methodology of investigation adopted: within the NTL research program, scholars make extensive use of psycholinguistic experiments, probabilistic and connectionist computational models, crosslinguistic analyses, and also pay attention to the results obtained in the field of neuroscience. The perspective of language as a CAS also deepens into issues related to anthropological factors, the study of grammaticalization, and language evolution. On the basis of this partial overlapping, in our opinion it is quite reasonable to hypothesize that an integration of ECG and NTL in this broader perspective would be auspicable, since an interaction between the projects carried out within the NTL paradigm and the work developed by the proponents of the CAS approach might be beneficial for both sides. As a matter of fact, at present NTL scholars have been prominently interested in the study of intrapersonal aspects of language, but they have never denied the importance of its sociocultural dimension; on the other hand, academics who adopts the CAS perspective generally focus a bit more on the social function of the language system, but they recognize the all-important role of the embodied and neural nature of intrasubjective factors. As a result, the overall picture appears to show that ECG scholars and advocates of the CAS perspective are pursuing largely overlapping routes whose starting point is roughly the same (i.e. the common views of linguistic knowledge as a network of form-meaning pairings, meaning as the result of conceptualization, the context-dependence of the communicative function, the acquisition of language as driven by statistical patterns of use, etc.), looking to reach the same goal (an explanation of language as a product of the constant interaction between embodied individual agents and the physical as well as sociocultural environment they are situated in). As a consequence, it seems plausible to posit that a close cooperation between ECG/NTL scholars and advocates of the CAS 10 The role of modeling is awarded particular emphasis. Indeed, while the complex interaction between form, user, and use (which are inextricably linked) are often difficult to investigate in vivo, since many properties of language are emergent, models allow one to provide empirical support to the hypothesis that basic mechanisms can combine to produce some observed effect. Moreover, modeling can reveal something about which mechanisms most strongly affect the emergent behavior and which ones play only a marginal role. See Blythe and Croft (2010), Christiansen and MacDonald (2010); Cornish et al. (2010); Ellis and Larsen-Freeman (2010b). 258 4. Conclusion: ECG and beyond approach might cause the progress made in these two enterprises to provide both of them with valuable insights for future work, in order to improve the level of accuracy of their analyses of language phenomena, thus helping the emergentist framework to grow stronger and stronger, gaining more and more reliability as a viable alternative to the dominant nativist paradigm. 259 References References Allen, Joseph, and Mark S. Seidenberg. (1999). The Emergence of Grammaticality in Connectionist Networks. In B. Macwhinney (ed.), Emergentist Approaches to Language. Proceedings of the 28th Carnegie Symposium on Cognition. Mahwah, NJ: Lawrence Erlbaum. 115-152. Andorno, Cecilia. 2003. Linguistica Testuale. Un'Introduzione. Roma: Carocci. Bailey, David R. (1997). When Push Comes to Shove. A Computational Model of the Role of Motor Control in the Acquisition of Action Verbs. Ph.D. dissertation. University of California, Berkeley. Barðdal, Jóhanna. (2006). Construction-Specific Properties of Syntactic Subjects in Icelandic and German. Cognitive Linguistics 17 (1): 39-106. Baroni, Marco, and Adam Kilgariff. (2006). Large Linguistically-Processed Web Corpora for Multiple Languages. In Proceedings of the Eleventh Conference of the European Chapter of the Association for Computational Linguistics. Ben-Gal, Irad. (2008). Bayesian Networks. In F. Ruggeri, R.S. Kennett, and F.W. Faltin (eds.), Encyclopedia of Statistics in Quality and Reliability. Chichester: Wiley. Bergen, Benjamin K. (2003). Introduction to Embodied Construction Grammar. Paper presented at the 8th International Cognitive Linguistics Conference. Logroño, Spain. Bergen, Benjamin K. (2005a). Mental Simulation in Literal and Figurative Language Understanding. In S. Coulson and B. Lewandowska-Tomaszczyk (eds.), The Literal / Non-literal Distinction. Frankfurt am Mein: Peter Lang. 255-278. Bergen, Benjamin K. (2005b). Mental Simulation in Spatial Language Processing. In Proceedings of the 27th Annual Conference of the Cognitive Science Society. Bergen, Benjamin K. (2007). Experimental Methods for Simulation Semantics. In M. González-Márquez, I. Mittelberg, S. Coulson, and M. Spivey (eds.), Methods in Cognitive Linguistics. Amsterdam: John Benjamins. 277-301. Bergen, Benjamin K., Shweta Narayan, and Jerome A. Feldman. (2003). Embodied Verbal Semantics. Evidence from an Image-Verb Matching Task. In Proceedings of the 25th Annual Conference of the Cognitive Science Society. Bergen, Benjamin K., Nancy C. Chang, and Shweta Narayan. (2004). Simulated Action in an Embodied Construction Grammar. In Proceedings of the Twenty-Sixth Annual Conference of the Cognitive Science Society. Bergen, Benjamin K., and Nancy C. Chang. (2005). Embodied Construction Grammar in SimulationBased Language Understanding. In J.-O. Östman and M. Fried (eds.), Construction Grammars. Cognitive Grounding and Theoretical Extensions. Amsterdam and Philadelphia: John Benjamins. 147-190. Bergen, Benjamin K., and Kathryn B. Wheeler. (2005). Sentence Understanding Engages Motor Processes. In Proceedings of the 27th Annual Conference of the Cognitive Science Society Bergen, Benjamin K., and Jerome A. Feldman. (2009). Embodied Concept Learning. In P. Calvo and T. Gomila (eds.), Handbook of Cognitive Science. An Embodied Approach. Amsterdam: Elsevier. 313-331. Bergen, Benjamin K., and Kathryn B. Wheeler. (2010). Grammatical Aspect and Mental Simulation. Brain and Language 112: 150-158. Berk, Lynn M. 1999. English Syntax. From Words to Discourse. New York: Oxford University Press. Bicknell, Klinton, and Roger Levy. (2009). A Model of Local Coherence Effects in Human Sentence Processing As Consequences of Updates from Bottom-Up Prior to Posterior Beliefs. In Proceedings of Human Language Technology. The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics. Blythe, Richard, and William Croft. (2010). The Speech Community in Evolutionary Language Dynamics. In N.C. Ellis and D. Larsen-Freeman (eds.). 47-63. Boas, Hans C. forthcoming. Computational Approaches to Embodied Construction Grammar. Amsterdam and San Diego: John Benjamins. Boroditsky, Lera. (2001). Does Language Shape Thought? English and Mandarin Speakers' Conception of Time. Cognitive Psychology 43: 1-22. Boroditsky, Lera, and Michael Ramscar. (2002). The Roles of Body and Mind in Abstract Thought. Psychological Science 13: 185-189. 261 Grounding Meaning in Everyday Experience in the World Bretones Calleja, Carmen. (2004). Embodied Construction Grammar Approach to Compositionality and Conceptual Representation. Paper presented at the Brain Inspired Cognitive System. Stirling, UK. Bretones Calleja, Carmen, María Cristóbal, and Iraide Ibarretxe-Antuñano. (2001). The Construction Salir-De in Spanish. How Spanish Speakers Conceptualize Existing Events. Paper presented at the 1st International Construction Grammar Conference. Berkeley, CA. Broccias, Cristiano. (2002). Review of Croft, William. 2001. Radical Construction Grammar. Syntactic Theory in Typological Perspective. New York: Oxford University Press. In: LINGUIST List: Vol-13-1516. Fri May 24 2002. ISSN: 1068-4875. Broccias, Cristiano. (2003). Quantity, Causality, and Temporality in Change Constructions. In Proceedings of the 29th Annual Meeting of the Berkeley Linguistics Society. Broccias, Cristiano. (2006). Cognitive Approaches to Grammar. In G. Kristiansen, M. Achard, R. Dirven, and F.J. Ruiz de Mendoza Ibáñez (eds.), Cognitive Linguistics. Current Applications and Future Perspectives. Berlin and New York: Mouton de Gruyter. 81-115. Bryant, John, and Luca Gilardi. (forthcoming). A Cognitive Model of Sentence Interpretation. In H.C. Boas (ed.). Bybee, Joan. 2003. Cognitive Processes in Grammaticalization. In M. Tomasello (ed.), The New Psychology of Language. Cognitive and Functional Approaches to Language Structure, vol. 2. Mahwah, NJ: Lawrence Erlbaum Associates. 145-168. Chan, Ting Ting, and Benjamin K. Bergen. (2005). Writing Direction Influences Spatial Cognition. In Proceedings of the 27th Annual Conference of the Cognitive Science Society. Chang, Nancy C. (2004). A Computational Model of Comprehension-Based Construction Acquisition. Paper presented at the Child Language Research Forum. Stanford, CA. Chang, Nancy C. (2008). Constructing Grammar. A Computational Model of the Emergence of Early Constructions. Ph.D. dissertation. University of California, Berkeley. Chang, Nancy C., Jerome A. Feldman, Robert Porzel, and Keith Sanders. (2002). Scaling Cognitive Linguistics. Formalisms for Language Understanding. In Proceedings of the 1st International Workshop on Scalable Natural Language Understanding. Chang, Nancy C., Jerome A. Feldman, and Srinivas Narayanan. (2004). Structured Connectionist Models of Language, Cognition, and Action. Paper presented at the 9 th Neural Computation and Psychology Workshop. Plymouth, UK. Chang, Nancy C., and Olya Gurevich. (2004). Context-Driven Construction Learning. In Proceedings of the 26th Annual Meeting of the Cognitive Science Society. Chang, Nancy C., and Eva H. Mork. (2006). A Structured Context Model of Grammar Learning. In Proceedings of the 2006 International Joint Conference on Neural Networks. Chater, Nick, and Christopher D. Manning. (2006). Probabilistic Models of Language Processing and Acquisition. Trends in Cognitive Sciences 10 (7): 335-344. Chater, Nick, and Morten H. Christiansen. (2010). Language Evolution as Cultural Evolution. How Language Is Shaped by the Brain. WIRES Cognition 1 (5): 623-628. Chen, Chao-Jan. (2008). Gestaltist Semantic Compositionality in Chinese V-V Compounds. Language and Linguistics 9 (2): 325-340. Chomsky, Noam A. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Christiansen, Morten H., Joseph Allen, and Mark S. Seidenberg. (1998). Learning to Segment Speech Using Multiple Cues. A Connectionist Model. Language and Cognitive Processes 13 (2): 221268. Christiansen, Morten H., and Nick Chater. (2001). Connectionist Psycholinguistics. Capturing the Empirical Data. Trends in Cognitive Sciences 5 (2): 82-88. Christiansen, Morten H., and Maryellen C. MacDonald. (2010). A Usage-Based Approach to Recursion in Sentence Processing. In N.C. Ellis and D. Larsen-Freeman (eds.). Clark, Andy. (2006). Language, Embodiment, and the Cognitive Niche. Trends in Cognitive Sciences 10 (8): 370-374. Clark, Andy. 2008. Supersizing the Mind. Embodiment, Action, and Cognitive Extension. Oxford: Oxford University Press. Corbett, Greville G. 1991. Gender. Cambridge, UK: Cambridge University Press. Cornish, Hannah, Monica Tamariz, and Simon Kirby. (2010). Complex Adaptive Systems and the Origins of Adaptive Structure. What Experiments Tell Us. In N.C. Ellis and D. Larsen-Freeman (eds.). 187-205. Cristofaro, Sonia. (2008). A Constructionist Approach to Complementation. Evidence from Ancient 262 References Greek. Linguistics 46 (3): 571-606. Croft, William. 2001. Radical Construction Grammar. Syntactic Theory in Typological Perspective. New York: Oxford University Press. Croft, William. 2003 (2nd ed.). Typology and Universals. Cambridge, UK: Cambridge University Press. Croft, William. (2007). Construction Grammar. In D. Geeraerts and H. Cuyckens (eds.). 463-508. Croft, William. (2009). Methods for Finding Language Universals in Syntax. In S. Scalise, E. Magni, and A. Bisetto (eds.), Universals of Language Today. Berlin: Springer. 145-164. Croft, William, and Alan D. Cruse. 2004. Cognitive Linguistics. Cambridge, UK: Cambridge University Press. Dąbrowska, Ewa. 2004. Language, Mind and Brain. Some Psychological and Neurological Constraints on Theories of Grammar. Edinburgh: Edinburgh University Press. De Beule, Joachim, and Bart De Vylder. (2005). Does Language Shape the Way in which We Conceptualize the World?. In Proceedings of the 27th Annual Conference of the Cognitive Science Society. De Beule, Joachim, and Benjamin K. Bergen. (2006). On the Emergence of Compositionality. In A. Cangelosi, A.D.M. Smith, and K. Smith (eds.), The Evolution of Language. Proceedings of the 6th International Conference. Singapore: World Scientific Publishing. 35-42. Dennison, Heeyeon Y., and Benjamin K. Bergen. (2010). Language-Driven Motor Simulation Is Sensitive to Social Context. In Proceedings of the 32nd Annual Conference of the Cognitive Science Society. De Mulder, Walter. (2007). Force Dynamics. In D. Geeraerts and H. Cuyckens (eds.). 294-317. Dodge, Ellen K. (2010a). Embodied Compositional Semantics. Unpublished material. University of California, Berkeley. Dodge, Ellen K. (2010b). Variations on a Theme. Patterns of Constructional and Conceptual Composition. Unpublished material. University of California, Berkeley. Dodge, Ellen K., and Abby Wright. (2002). Herds of Wildebeest, Flasks of Vodka, Heaps of Trouble. An Embodied Construction Grammar Approach to English Measure Phrases. In Proceedings of the 28th Annual Meeting of the Berkeley Linguistics Society. Dodge, Ellen K., and John Bryant. (forthcoming). Computational Cognitive Linguistics. An Embodied Construction Grammar Analysis of Transitive Constructions. In H.C. Boas (ed.). Doleschal, Ursula. (2006). La Relazione Semiotica fra la Categoria Grammaticale del Genere e la Sua Denotazione. In S. Luraghi and A. Olita (eds.), Linguaggio e Genere. Grammatica e Usi. Roma: Carocci. 42-53. Dörnyei, Zoltán. (2010). Individual Differences. Interplay of Learner Characteristics and Learning Environment. In N.C. Ellis and D. Larsen-Freeman (eds.). 230-248. Du Bois, John W. (2003). Argument Structure. Grammar in Use. In J.W. Du Bois, L.E. Kumpf and W.J. Ashby (eds.), Preferred Argument Structure. Grammar as Architecture for Function. Amsterdam and Philadelphia: John Benjamins. Ellis, Nick C., and Diane Larsen-Freeman (eds.). 2010a. Language as a Complex Adaptive System. Chichester: Wiley. Ellis, Nick C., and Diane Larsen-Freeman. (2010b). Constructing a Second Language. Analyses and Computational Simulations of the Emergence of Linguistic Constructions From Usage. In N.C. Ellis and D. Larsen-Freeman (eds.). 90-125. Elman, Jeffrey L. (1993). Learning and Development in Neural Networks. The Importance of Starting Small. Cognition 48: 71-99. Ettlinger, Marc. (2005). Constructional Compositionality and the English Resultative. Paper presented at the 41st Meeting of the Chicago Linguistic Society. Chicago, IL. Evans, Vyvyan. 2007. A Glossary of Cognitive Linguistics. Edinburgh: Edinburgh University Press. Evans, Vyvyan, and Melanie Green. 2006. Cognitive Linguistics. An Introduction. Edinburgh: Edinburgh University Press. Fauconnier, Gilles. [1985] 1994. Mental Spaces. Aspects of Meaning Construction in Natural Language. Cambridge, UK: Cambridge University Press. Fauconnier, Gilles, and Eve E. Sweetser (eds.). 1996. Spaces, Worlds, and Grammar. Chicago: The University of Chicago Press. Fauconnier, Gilles, and Mark Turner. 2002. The Way We Think. Conceptual Blending and the Mind's Hidden Complexities. New York: Basic Books. Feldman, Jerome A. (2002). The Meaning of Reference in Embodied Construction Grammar. Technical Report, International Computer Science Institute, ICSI TR 02-011, Berkeley. 263 Grounding Meaning in Everyday Experience in the World Feldman, Jerome A. (2004). Computational Cognitive Linguistics. In Proceedings of the 20th International Conference of Computational Linguistics. Feldman, Jerome A. 2006. From Molecule to Metaphor. A Neural Theory of Language. Cambridge, MA: MIT Press. Feldman, Jerome A. (2010). Embodied Language, Best-Fit Analysis, and Formal Compositionality. Physics of Life Review, doi:10.1016/j.plrev.2010.06.006. Feldman, Jerome A., and Srinivas Narayanan. (2004). Embodied Meaning in a Neural Theory of Language. Brain and Language 89: 385-392. Feldman, Jerome A., Ellen K. Dodge, and John Bryant. (2009). Embodied Construction Grammar. In B. Heine and H. Narrog (eds.), The Oxford Handbook of Linguistic Analysis. New York: Oxford University Press. 111-138. Feldman, Jerome A. , and Luca Gilardi. (forthcoming). Extending ECG to Communities, Mental Spaces, and Maps. In H.C. Boas (ed.). Ferreira, Fernanda, Kiel Christianson, and Andrew Hollingworth. (2001). Misinterpretations of GardenPath Sentences. Implications for Models of Sentence Processing and Reanalysis. Journal of Psycholinguistic Research 30 (1): 3-20. Fillmore, Charles J. (1982). Frame Semantics. In Linguistic Society of Korea (ed.), Linguistics in the Morning Calm. Seoul: Hanshin Publishing. 111-137. Fillmore, Charles J., Paul Kay, and Mary C. O'Connor. (1988). Regularity and Idiomaticity in Grammatical Constructions. The Case of Let Alone. Language 64: 501-538. Fodor, Jerry A. 1975. The Language of Thought. Cambridge, MA: Harvard University Press. Fodor, Jerry A. 1983. The Modularity of Mind. An Essay on Faculty Psychology. Cambridge, MA: MIT Press. Gallese, Vittorio. (2006). Embodied Simulation. From Mirror Neuron System to Interpersonal Relations. In Novartis Foundation (eds.), Empathy and Fairness. Novartis Foundation Symposium 278. Chichester: Wiley. Gallese, Vittorio. (2008). Mirror Neurons and the Social Nature of Language. The Neural Exploitation Hypothesis. Social Neuroscience 3 (3-4): 317-333. Gallese, Vittorio, Luciano Fadiga, Leonardo Fogassi, and Giacomo Rizzolatti (1996). Action Recognition in the Premotor Cortex. Brain 119: 593-609. Gallese, Vittorio, Christian Keysers, and Giacomo Rizzolatti. (2004). A Unifying View of the Basis of Social Cognition. Trends in Cognitive Sciences 8 (9): 396-403. Gallese, Vittorio, and George Lakoff. (2005). The Brain's Concepts. The Role of the Sensory-Motor System in Conceptual Knowledge. Cognitive Neuropsychology 22 (1): 455-479. Gaskell, M. Gareth (ed.). 2007a. The Oxford Handbook of Psycholinguistics. New York: Oxford University Press. Gaskell, M. Gareth. (2007b). Statistical and Connectionist Models of Speech Perception and Word Recognition. In M.G. Gaskell (ed.). 55-70. Geeraerts, Dirk, and Hubert Cuyckens (eds.). 2007a. The Oxford Handbook of Cognitive Linguistics. Oxford and New York: Oxford University Press. Geeraerts, Dirk, and Hubert Cuyckens. (2007b). Introducing Cognitive Linguistics. In D. Geeraerts and H. Cuyckens (eds.). 3-21. Gentilucci, Maurizio. (2003). Object Motor Representation and Language. Experimental Brain Research 153: 260-265. Gibbs, Raymond W. Jr. 2005. Embodiment and Cognitive Science. New York: Cambridge University Press. Glenberg, Arthur M. (2007). Language and Action. Creating Sensible Combinations of Ideas. In M.G. Gaskell (ed.). 361-370. Goldberg, Adele E. 1995. Constructions. A Construction Grammar Approach to Argument Structure. Chicago: The University of Chicago Press. Goldberg, Adele E. (2003). Constructions. A New Theoretical Approach to Language. Trends in Cognitive Sciences 7 (5): 219-224. Goldberg, Adele E. 2006. Constructions at Work. The Nature of Generalization in Language. New York: Oxford University Press. Goldberg, Adele E. (2009a). The Nature of Generalization in Language. Cognitive Linguistics 20 (1): 93127. Goldberg, Adele E. (2009b). Constructions Work. Cognitive Linguistics 20 (1): 201-224. Goldberg, Adele E., and Devin Casenhiser. (2006). English Constructions. In B. Aarts and A. MacMahon 264 References (eds.), Handbook of English Linguistics. Oxford: Blackwell. 343-355. Gómez Torrego, Leonardo. 2002 (8th ed.). Gramática Didáctica del Español. Madrid: Ediciones SM. Gonzálvez García, Francisco. (2009a). An Overview of Functional-Cognitive Space in the 21st Century. Seminar held at Università degli Studi di Pavia. Gonzálvez García, Francisco. (2009b). Cognition and Language Learning in Functional-Cognitive Models. Seminar held at Collegio Ghislieri, Università degli Studi di Pavia. Graffi, Giorgio. 1994. Sintassi. Bologna: Il Mulino. Gurevich, Olya. (2006). Constructional Morphology. The Georgian Version. Ph.D. dissertation. University of California, Berkeley. Haegeman, Lilian. 1991. Introduction to Government and Binding Theory. Oxford: Blackwell. Haspelmath, Martin. (2003). The Geometry of Grammatical Meaning. Semantic Maps and Crosslinguistic Comparison. In M. Tomasello (ed.), The New Psychology of Language. Cognitive and Functional Approaches to Language Structure, vol. 2. Mahwah, NJ: Lawrence Erlbaum Associates. 211-242. Hilpert, Martin. (2008). New Evidence Against the Modularity of Grammar. Constructions, Collocations, and Speech Perception. Cognitive Linguistics 19 (3): 491-511. Hopper, Paul J., and Elizabeth C. Traugott. 2003 (2a ed.). Grammaticalization. Cambridge, UK: Cambridge University Press. Huang, Yan. 2007. Pragmatics. Cambridge, UK: Cambridge University Press. Itkonen, Esa. 2003. What Is a Language? A Study in the Philosophy of Linguistics. Turku: Ǻbo Akademis Tryckeri. Janda, Laura A. (2006). Cognitive Linguistics. Glossos 8 (http://www.seelrc.org/glossos/issues/8/janda.pdf) Johnson, Mark. 1987. The Body in the Mind. The Bodily Basis of Meaning, Imagination, and Reason. Chicago: The University of Chicago Press. Johnson, Mark, and George Lakoff. (2002). Why Cognitive Linguistics Requires Embodied Realism. Cognitive Linguistics 13 (3): 245-263. Klein, Devorah E., and Gregory L. Murphy. (2001). The Representation of Polysemous Words. Journal of Memory and Language 45: 259-282. Lakoff, George. 1987. Women, Fire, and Dangerous Things. What Categories Reveal about the Mind. Chicago: The University of Chicago Press. Lakoff, George, and Mark Johnson. 1980. Metaphors We Live By. Chicago: The University of Chicago Press. Lakoff, George, and Mark Johnson. 1999. Philosophy in the Flesh. The Embodied Mind and Its Challenge to Western Thought. New York: Basic Books. Lambrecht, Knud. 1994. Information Structure and Sentence Form. Topic, Focus and the Mental Representations of Discourse Referents. Cambridge, UK: Cambridge University Press. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, vol. 1. Theoretical Prerequisites. Stanford, CA: Stanford University Press. Langacker, Ronald W. 1991. Concept, Image, and Symbol. The Cognitive Basis of Grammar. Berlin: Mouton de Gruyter. Langacker, Ronald W. (2007). Cognitive Grammar. In D. Geeraerts and H. Cuyckens (eds.). Oxford and New York: Oxford University Press. 421-462. Langacker, Ronald W. 2008. Cognitive Grammar. A Basic Introduction. New York: Oxford University Press. Langacker, Ronald W. (2009). Cognitive (Construction) Grammar. Cognitive Linguistics 20 (1): 167-176. Lewandowska-Tomaszczyk, Barbara. (2007). Polysemy, Prototypes, and Radial Categories. In D. Geeraerts and H. Cuyckens (eds.). 139-169. Lombardi Vallauri, Edoardo. (2009). In che Modo il Linguaggio Non E' nel Cervello. Seminar held at Collegio Ghislieri, Università degli Studi di Pavia. Luraghi, Silvia. (2009). A Model for Representing Polysemy. The Italian Preposition Da. In J. François, E. Gilbert, C. Guinier, and M. Krause (eds.), Actes du Colloque “Autor de la Préposition”. Caen: Press Universitaire de Caen. 167-178. Maalej, Zouhair. (2008). The Heart and Cultural Embodiment in Tunisian Arabic. In R. Dirven, Y. Ning, S. Niemeier (eds.), Culture, Body, and Language. Conceptualization of Internal Body Organs across Cultures and Languages. Berlin and New York: Mouton de Gruyter. 395-428. Masini, Francesca, and Paola Pietrandrea. (2010). Magari. Cognitive Linguistics 21 (1): 75-122. Massaro, Dominic W., and Alexandra Jesse. (2007). Audiovisual Speech Perception and Word 265 Grounding Meaning in Everyday Experience in the World Recognition. In M.G. Gaskell (ed.). 19-36. Mathiessen, Christian M.I.M. (2010). Meaning in the Making. Meaning Potential Emerging from Acts of Meaning. In N.C. Ellis and D. Larsen-Freeman (eds.). 206-229. Matlock, Teenie, Michael Ramscar, and Lera Boroditsky. (2003). The Experiential Basis of Meaning. In Proceedings of the 25th Annual Conference of the Cognitive Science Society. Maybin, Janet, and Joan Swann (eds.). 2006. The Art of English. Everyday Creativity. Houndmills: Palgrave Macmillan. Mazzone, Marco. (2004). Linguistica Cognitiva. In A. Pennisi and P. Perconti (eds.), Le Scienze Cognitive del Linguaggio. Bologna: Il Mulino. 93-134. McClelland, James L., and Jeffrey L. Elman. (1986). The TRACE Model of Speech Perception. Cognition 18: 1-86. McClelland, James L., and Axel Cleeremans. (2009). Connectionist Models. In T. Byrne, A. Cleeremans, and P. Wilken (eds.), Oxford Companion to Consciousness. New York: Oxford University Press. 177-181. Mok, Eva H., John Bryant, and Jerome A. Feldman. (2004). Scaling Understanding Up to Mental Spaces. In Proceedings of the 3rd Workshop on Scalable Natural Language Understanding. Mok, Eva H., and John Bryant. (2006). A Best-Fit Approach to Productive Omission of Arguments. In Proceeding of the 32nd Annual Meeting of the Berkeley Linguistics Society. Moneti, Annamaria, and Graziella Lazzarino. 2007 (6th ed.). Da Capo. Boston: Thomson Learning. Narayan, Shweta, Benjamin K. Bergen, and Zachary Weinberg. (2004). Embodied Verbal Semantics. Evidence from a Lexical Matching Task. In Proceedings of the 30th Annual Meeting of the Berkeley Linguistic Society. Narayanan, Srinivas. (1996). Embodiment in Language Understanding. Modeling the Semantics of Causal Narratives. Paper presented at the AAAI Fall-Symposium on Embodied Cognition and Action. Cambridge, MA. Narayanan, Srinivas. (1997). KARMA: Knowledge-based Action Representation for Metaphor and Aspect. Ph.D. dissertation. University of California, Berkeley. Narayanan, Srinivas. (1999). Moving Right Along. A Computational Model of Metaphoric Reasoning about Events. In Proceedings of the National Conference on Artificial Intelligence. Narayanan, Srinivas, and Daniel Jurafsky. (1998). Bayesian Models of Human Sentence Processing. In Proceedings of the 20th Annual Conference of the Cognitive Science Society. Narayanan, Srinivas, and Daniel Jurafsky. (2001). A Bayesian Model Predicts Human Parse Preferences and Reading Times in Sentence Processing. In Proceedings of the 14th Neural Information Processing Systems Conference. Næss, Åshield. (2007). Defining Transitivity. Markedness vs. Prototypicality. In M. Miestamo and B. Wălchli (eds.), New Challenges in Typology. Broadening the Horizon and Redefining the Foundations. Berlin: Mouton de Gruyter. 179-198. Nuyts, Jan. (2007). Cognitive Linguistics and Functional Linguistics. In D. Geeraerts and H. Cuyckens (eds.). 543-565. Oakley, Todd. (2007). Image Schemas. In D. Geeraerts and H. Cuyckens (eds.). 214-235. Ochsenbauer, Anne-Katharina, and Maya Hickmann. (2010). Children's Verbalizations of Motion Events in German. Cognitive Linguistics 21 (2): 217-236. Onnis, Luca, Morten H. Christiansen, and Nick Chater. (2006). Human Language Processing. Connectionist Models. In K. Brown (ed.), The Elsevier Encyclopedia of Language and Linguistics. Amsterdam: Elsevier. 401-409. Pedersen, Johan. (2005). The Spanish Impersonal Se-Construction. Constructional Variation and Change. Constructions 1 (www.constructions-online.de, urn:nbn:de: 0009-4-1457, ISSN 1860-2010). Perissutti, Anna Maria. (2008). L'Alternanza Locativa Argomentale in Ceco. Un Approccio Costruzionista. Linguistica e Filologia 26: 141-168. Petruck, Miriam R.L. (1996). Frame Semantics. In J. Verschueren, J-O. Östman, J. Blommaert, and C. Bulcaen (eds.), Handbook of Pragmatics. Philadelphia: John Benjamins. 1-13. Pisoni, David B., and Susannah V. Levi. (2007). Representations and Representational Specificity in Speech Perception and Spoken Word Recognition. In M.G. Gaskell (ed.). 3-18. Plebe, Alessio. (2004). Modelli Neurocomputazionali del Linguaggio. In A. Pennisi and P. Perconti (eds.), Le Scienze Cognitive del Linguaggio. Bologna: Il Mulino. 163-192. Polesini Karumanchiri, Luisa, and Raffaella Uslenghi Maiguashca. 1988. L'Italiano d'Oggi. Note di Grammatica per Corsi Universitari. Toronto: University of Toronto Press. Proudfoot, Anna, and Francesco Cardo. 2005 (2nd ed.). Modern Italian Grammar. A Practical Guide. New 266 References York: Routledge. Pulvermüller, Friedemann. (2007). Word Processing in the Brain As Revealed by Neurophysiological Imaging. In M.G. Gaskell (ed.). 119-140. Quochi, Valeria. (2007). A Usage-Based Approach to Light Verb Constructions in Italian. Development and Use. Ph.D. dissertation. Università degli Studi di Pisa. Regier, Terry. 1996. The Human Semantic Potential. Spatial Language and Constrained Connectionism. Cambridge, MA: MIT Press. Regier, Terry. (2003a). Constraining Computational Models of Cognition. In L. Nadel (ed.), Encyclopedia of Cognitive Science. London: Macmillan. 611-615. Regier, Terry. (2003b). Emergent Constraints on Word Learning. A Computational Perspective. Trends in Cognitive Sciences 7 (6): 263-268. Rizzolatti, Giacomo, and Michael A. Arbib. (1998). Language within Our Grasp. Trends in Neuroscience 21 (5): 188-194. Rizzolatti, Giacomo, and Laila Craighero. (2007). Language and Mirror Neurons. In M.G. Gaskell (ed.). 771-786. Rohde, Douglas L.T., and David C. Plaut. (2003). Connectionist Models of Language Processing. Cognitive Studies 10 (1): 10-28. Rohrer, Tim. (2007). Embodiment and Experientialism. In D. Geeraerts and H. Cuyckens (eds.). 25-47. Ruiz de Mendoza Ibáñez, Francisco J. (2010). Levels of Meaning Construction. An Introduction to the Lexical Constructional Model. Seminar held at Università degli Studi di Pavia. Schneider, Nathan. (2010). Computational Cognitive Morphosemantics. Modeling Morphological Compositionality in Hebrew Verbs with Embodied Construction Grammar. Paper presented at the 36th Annual Meeting of the Berkeley Linguistic Society. Berkeley, CA. Schoenemann, P. Thomas. (2010). Evolution of Brain and Language. In N.C. Ellis and D. LarsenFreeman (eds.). 162-186. Seidenberg, Mark S. (2007). Connectionist Models of Reading. In M.G. Gaskell (ed.). 235-250. Shapiro, Lawrence A. 2010. Embodied Cognition. New York: Routledge. Shastri, Lokendra, and Dean J. Grannes. (1996). A Connectionist Treatment of Negation and Inconsistency. In Proceedings of the 18th Annual Conference of the Cognitive Science Society. Shastri, Lokendra, and Carter Wendelken. (2000). Seeking Coherent Explanations – a Fusion of Structured Connectionism, Temporal Synchrony, and Evidential Reasoning. In Proceedings of the 22nd Annual Conference of the Cognitive Science Society. Sinha, Chris. (2002). The Cost of Renovating the Property. A Reply to Marina Rakova. Cognitive Linguistics 13 (3): 271-276. Sinha, Chris. (2007). Cognitive Linguistics, Psychology, and Cognitive Science. In D. Geeraerts and H. Cuyckens (eds.). 1266-1294. Søgaard, Anders. (2006). Embodied Construction Grammar As Layered Modal Languages. In Proceedings of the 3rd Workshop on Scalable Natural Language Understanding. Steels, Luc, and Joachim de Beule. (2006). A (Very) Brief Introduction to Fluid Construction Grammar. In Proceedings of the 3rd Workshop on Scalable Natural Language Understanding. Sweetser, Eve E. 1990. From Etymology to Pragmatics. Metaphorical and Cultural Aspects of Semantic Structure. Cambridge, UK: Cambridge University Press. Talmy, Leonard. 2000. Toward a Cognitive Semantics. 2 Vols. Cambridge, MA: MIT Press. Taylor, John R. 2002. Cognitive Grammar. New York: Oxford University Press. Taylor, John R. 2003 (3rd ed.). Linguistic Categorization. Prototypes in Linguistic Theory. New York: Oxford University Press. Taylor, John R. (2007). Cognitive Linguistics and Autonomous Linguistics. In D. Geeraerts and H. Cuyckens (eds.), The Oxford Handbook of Cognitive Linguistics. Oxford and New York: Oxford University Press. 566-588. The Five Graces Group. (2010). Language Is a Complex Adaptive System. Position Paper. In N.C. Ellis and D. Larsen-Freeman (eds.). 1-26. Tognozzi, Elissa, and Giuseppe Cavatorta. 2009 (2nd ed.). Ponti. Italiano Terzo Millennio. Boston: Cengage Learning. Tomasello, Michael. 1999. The Cultural Origins of Human Cognition. Cambridge, MA: Harvard University Press. Tomasello, Michael. 2003. Constructing a Language. A Usage-Based Theory of Language Acquisition. Cambridge, MA: Harvard University Press. Tomasello, Michael. 2008. Origins of Human Communication. Cambridge, MA: MIT Press. 267 Grounding Meaning in Everyday Experience in the World Traugott, Elizabeth C., and Richard Dasher. 2002. Regularities in Semantic Change. Cambridge, UK: Cambridge University Press. Tseng, Meylysa J., and Benjamin K. Bergen. (2005). Lexical Processing Drives Motor Simulation. In Proceedings of the 27th Annual Conference of the Cognitive Science Society. Tuggy, David. (2007). Schematicity. In D. Geeraerts and H. Cuyckens (eds.). 82-116. Tyler, Andrea, and Vyvyan Evans. 2003. The Semantics of English Prepositions. Spatial Scenes, Embodied Meaning and Cognition. Cambridge, UK: Cambridge University Press. Ungerer, Friedrich, and Hans-Jörg Schmid. 2006 (2nd ed). An Introduction to Cognitive Linguistics. Harlow: Longman. Varela, Francisco J., Evan Thompson, and Eleanor Rosch. 1991. The Embodied Mind. Cognitive Science and Human Experience. Cambridge, MA: MIT Press. Vigliocco, Gabriella, and David P. Vinson. (2007). Semantic Representation. In M.G. Gaskell (ed.). 195216. Wendelken, Carter, and Lokendra Shastri. (2005). Connectionist Mechanisms for Cognitive Control. Neurocomputing 65-66: 663-672. Wheeler, Kathryn B., and Benjamin K. Bergen. (2010). Meaning in the Palm of Your Hand. In S. Rice and J. Newman (eds.), Empirical and Experimental Methods in Cognitive / Functional Research. Stanford, CA: CSLI. Wilson, Nicole L., and Raymond W. Gibbs Jr. (2007). Real and Imagined Body Movement Primes Metaphor Comprehension. Cognitive Science 31: 721-731. Zelinsky-Wibbelt, Cornelia (ed.). 1993. The Semantics of Prepositions. From Mental Processing to Natural Language Processing. Berlin: Mouton de Gruyter. Zlatev, Jordan. (2005). What's in a Schema? Bodily Mimesis and the Grounding of Language. In B. Hampe (ed.), From Perception to Meaning. Image Schemas in Cognitive Linguistics. Berlin: Mouton de Gruyter. 313-342. Zlatev, Jordan. (2007a). Embodiment, Language, and Mimesis. In T. Ziemke, J. Zlatev, and R. Frank (eds.), Body, Language and Mind, vol. 1. Embodiment. Berlin: Mouton de Gruyter. 297-337. Zlatev, Jordan. (2007b). Spatial Semantics. In D. Geeraerts and H. Cuyckens (eds.). 318-350. 268