The Semantic Architecture of the Faculty of Language: Compositional Operations and Complex Lexical Representations Guillermo Del Pinal Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Graduate School of Arts and Sciences COLUMBIA UNIVERSITY 2015 ©2015 Guillermo Del Pinal All rights reserved ABSTRACT The Semantic Architecture of the Faculty of Language: Compositional Operations and Complex Lexical Representations Guillermo Del Pinal This dissertation defends two controversial claims about the types of opera- tions and representations used by the Faculty of Language (FL) to compute the meaning of expressions. (I) FL is compositional in the sense that it computes the meaning of complex expressions from their structure and the meanings of their immediate constituents. Compositionality, I argue, is an innate constraint on the meaning-determining operations of FL. (II) FL uses lexical items which, in most cases, have internal semantic structure. This internal structure has extension-determining and non-extension determining components. I argue that the non-extension components are part of linguistic meaning because they play a key role in certain compositional linguistic constructions. These claims lie at the center of important debates in Philosophy of Mind, Language, and Cognitive Science. Theorists who approach the study of language using formal tools tend to be sympathetic to the view that FL is compositional. However, this position has recently come under serious attack from many directions. In contrast, the- sis (II) regarding the internal complexity of lexical items has been increasingly supported by Psychologists and Cognitive Scientists, but tends to be rejected by formal Philosophers and Formal Semanticists. In this dissertation, I argue that both claims are independently plausible and, despite the consensus to the contrary, mutually reinforcing. In addition, I present an implementation of a compositional type-driven formal semantic theory that has access to the non- extension determining components of lexical items, and discuss some crucial advantages of this approach over more traditional formal semantic theories. Contents 1 General Introduction 1 I On the Semantic Operations of FL 4 2 Why is FL Compositional? 5 2.1 Introduction . .6 2.2 FL as a Cognitive Computational System . .8 2.3 Compositionality as a functional constraint of FL ......... 12 2.4 Compositionality as a MDC .................... 16 2.5 CO vs Non-compositional MDCs.................. 23 2.5.1 Context-Sensitive Non-compositional MDCs . 24 2.5.2 General Beliefs Sensitive Non-compositional MDCs . 32 2.6 Objections and Open Issues . 41 2.6.1 Against the Chomskian FL ................. 41 2.6.2 Descriptive Adequacy . 43 2.6.3 Why doesn't CO over-generate meanings? . 45 2.6.4 Are we ignoring trade-offs? . 47 2.6.5 Why not take CO as a defeasible default . 49 2.6.6 Why can't MDCs be learned . 51 i 2.6.7 Implications for Constructionist Approaches . 55 2.7 Conclusion . 58 3 The Systematic Productivity of FL 60 3.1 Introduction . 61 3.2 PRODUCTIVITY . 63 3.3 SYSTEMATICITY . 72 3.4 Conclusion . 88 II On the Semantic Representations of FL 92 4 Privatives and Dual Content Semantics 93 4.1 Introduction . 94 4.2 Philosophical Motivations . 97 4.3 Implementing DUAL CONTENT . 103 4.4 Contextualist account of privative NPs . 121 4.5 DUAL CONTENT vs SHIFTING HEADS . 125 4.6 An attempt to save SHIFTING HEADS . 130 4.7 Objections and Clarifications . 134 4.7.1 DUAL CONTENT and Contextualism . 134 4.7.2 Why not enhance E-structure? . 138 4.7.3 Why not drop E-structure? . 141 4.7.4 The relation between E-structure and C-structure . 144 4.8 Conclusion . 147 5 Prototypes and Compositionality 149 5.1 Introduction . 150 5.2 Compositional prototype combinatorics . 153 5.3 The problem: emergent features . 157 ii 5.4 The usual response: abandon C ................... 159 5.5 Compositionality revisited . 164 5.6 Lexical prototypes revisited . 171 5.7 Pet fish and beach bicycle reconsidered . 174 5.8 Experimental findings reconsidered . 178 5.8.1 Time-sensitive compositionality experiment . 178 5.8.2 The modifier effect . 181 5.9 Objections and clarifications . 185 5.10 Conclusion . 190 Bibliography 192 iii Acknowledgements This dissertation benefited enormously from the help and guidance of the mem- bers of my committee, Akeel Bilgrami, Daniel Rothschild and Achille Varzi. Akeel Bilgrami brilliant knack for coming up with clear and succinct versions of my arguments lead to various substantial improvements. Achille Varzi forced me to address important foundational issues that I sometimes, unwisely, tried to avoid. Daniel Rothschild has helped me in more ways that I can single out: he helped me frame the main problems, steered me clear of countless fruitless paths, raised general and specific challenges and helped me develop potential so- lutions. I also discussed various parts of this dissertation with various colleagues at Columbia University. I am especially grateful to Michael Brent, Sebastian Watzl, Anubav Vaseduvian, Brian H. Kim, Marco Nathan, Nathan Brie, and Matt Moss. I would also like to thank Zoltan Szabo for some very interesting and helpful discussions of compositionality, which lead to substantial improve- ments to Chapters 2 and 3. Finally, I would also like to thank Louise McNally and Barbara Partee for their sharp and constructive comments in a version of Chapter 4 which was submitted for publication. iv for Corina Grethel Saenz Lehnhoff v Chapter 1 General Introduction The Faculty of Language enables us to learn languages that we can use to express and understand an unbounded number of thoughts. What is the basic structure of this faculty of the mind such that it can perform that task? According to a distinguished tradition that runs through psychology, linguistics and philosophy, the first step to answer this question is to assume that the Faculty of Language is a cognitive computational system. This computational approach is generally adopted in investigations of our phonetic and syntactic competence. But some prominent theorists have argued that our semantic competence, our capacity to understand the meaning of expressions, is not sufficiently systematic to call for a computational explanation. This dissertation begins with a defense of the view that our semantic compe- tence is systematic and productive in just the way we would expect if the Faculty of Language were a recursive computational system, part of whose function is to compute not only the phonetic and syntactic structure of expressions but also their meaning (Chapters 1-2). Once this move is accepted, we can ask a more specific question about the basic structure of the Faculty of Language: 1 What are the basic operations and representations which it uses to compute the meaning of expressions? The main goal of this dissertation is to defend two claims about the kinds of operations and representations used by the Faculty of Language to compute the meaning of expressions: (I) The Faculty of Language is compositional in the sense that it computes the meaning of complex expressions from their structure and the meanings of their immediate constituents. Compositionality is an innate constraint on the operations of the Faculty of Language (Chapter 3). (II) The Faculty of Language uses lexical items which, in most cases, have in- ternal semantic structure. This internal structure has a part which deter- mines their extension and a part which does not determine their extension but that plays a role in certain compositions (Chapter 4-5). (I) and (II) lie at the center of heated debates about the architecture of the Fac- ulty of Language. My strategy for defending them flows from a general dissatis- faction with many of the current discussions: despite the occasional lip-service, theorists of language too often forget that questions about compositionally and lexical semantics are paradigmatic questions of cognitive science, which require a genuine interdisciplinary approach. For example, many theorists argue, against (I), that even if we assume that the Faculty of Language has access to non- compositional operations to compute the meaning of expressions, we can still account for the systematicity of our semantic competence. While this is true, it is not a good reason to reject (I). For the assumption that the Faculty of Language can use non-compositional operations, while locally acceptable, has unacceptable consequences for language acquisition (Chapter 3). Similarly, many linguists and philosophers try to account for a whole range of problematic expressions without significantly enriching the formal semantic 2 framework which we inherited from Frege and Montague. So they are commit- ted, against (II), to the view that lexical items do not have, or at least that the compositional operations are not sensitive to, non-extension determining internal semantic structure. As a result, some of their accounts look like im- plausible ad hoc attempts to save the original formal semantic framework. In the meantime, cognitive psychologists are actively developing sophisticated the- ories about the complex structure of our lexical concepts. If we reformulate our semantic theories so that they can use the most plausible of these complex structures, we can provide more principled compositional accounts for a whole range of problematic expressions (Chapter 4-5). When tackling questions about the basic architecture of the Faculty of Lan- guage we have to face some classic debates in the philosophy of mind and lan- guage: nativism vs empiricism, descriptivism vs essentialism, and the priority of thought vs that of language. To defend (I) is to defend a form of nativism about our linguistic competence, for (I) says that compositionality is an innate constraint on the basic semantic operations of the Faculty of Language. To defend (II) is to defend a mix between descriptivism and essentialism about lexical semantics, for (II) says that lexical terms have a dual structure with an extension-determining component and a descriptive, non-extension deter- mining component.