The architecture developed here already appeared in its essentials in my II}R7 book Consciousness and the Computational Mind, where I situated IIIL’ language capacity in a more general theory of mental representations, l·onnccted this view with processing, and stated the version of the modu larity hypo thesis here called Representational Modularity. At that time I assumed that the architecture could be connected with just about any theory of syntax, and left it at that. However, two particular issues led to the real i zati on that my assumption was too optimistic. The (irst concerns the nature of lex ical insertion. In 1991, Joe Emonds published a critique of my theory of conceptual structure, in which it ‘I\’elllcd to me that much of the disagreement turned on differing presup po s i ti ons about how lexical items enter in to syntactic structure.
Around tin’ same time the very same presup pos itions surfaced as crucial in a IIl1ll1hl’r of inte n se discussions with Daniel Biiring and Katharina Hart11101 nil. C on se q u e n tl y in r e ply in g to Emonds, I found it necessary to un earth these presuppositions and decide what I t h o ught lexical insertion is rt’ally like. A fortuitous i n vi ta tion to the 2nd Tilburg Idioms C on feren ce alive IIIl’ the opportun ity to e x pan d these ideas and w ork out their im . ,11I:al ions for idioms, l’oincidentally bringing me back to questi ons of again. ” , . xiv Preface lexical content that I had thought about in connection with my 1 975 paper on lexical redundancy rules, ” Morphological and Semantic Regu larities in the Lexicon.
The second major impetus behind this book was the appearance of Chomsky’s Minimalist Program, whose goal is to determine how much of the structure of language can be deduced from the need for language to obey boundary conditions on meaning and pronunciation. I found myself agreeing entirely with Chomsky’s goals, but differing radically in what I considered an appropriate exploration and realization. Bringing these concerns together with those of lexical insertion led to a 1 994 paper en titled ” Lexical Insertion in a Post-Minimalist Theory of Grammar. ” This paper was circulated widely, and I received (what was for me at least) a flood of comments, mostly enthusiastic, from linguists and cognitive scientists of all stripes. As a consequence, the paper grew beyond a size appropriate for a journal, hence the present monograph.
Many of the ideas in this study have been floating around in various subcommunities of linguistics, sometimes without contact with each other. The idea that a grammar uses unification rather than substitution as its major operation, which I have adopted here, now appears in practically every approach to generative grammar outside Chomsky’s immediate circle; similarly, most non-Chomskian generative approaches have aban doned transformational derivations (an issue about which I am agnostic here). The idea adopted here of multiple, coconstraining grammatical structures appeared first in autosegmental phonology of the mid- 1 970s and continues into the syntax-phonology relation and the syntax-semantics relation in a wide range of approaches.
What is original here, I think, is not so much the technical devices as the attempt to take a larger perspective than usual on grammatical struc ture and to fit all of these innovations together, picking and choosing variations that best suit the whole. In fact, for reasons of length and readability, I have not gone into a lot of technical detail. Rather, my in tent is to establish reasonable boundary conditions on the architecture and to work out enough consequences to see the fruitfulness of the approach. My hope is to encourage those who know much more than I about millions of different details to explore the possibilities of this architecture for their own concerns.
Acknowledgments xpress my gratitude to those people who offered comments materials from which the present work developed, including Steve Anderson, Mark Aronoff, Hans den Besten, Manfred Bierwisch, Daniel Buring, Patrick Cavanagh, Marcelo Dascal, Martin Everaert, Yehuda J;alk, Bob Freidin, Adele Goldberg, Georgia Green, Jane Grimshaw, Ken Ilale, Morris Halle, Henrietta Hung, Paul Kay, Paul Kiparsky, Ron I a n gac ker, Joan Maling, Alec Marantz, Fritz Newmeyer, Urpo Nikanne, Curios Otero, Janet Randall, Jerry Sadock, Ivan Sag, Andre Schenk, Lindsay Whaley, Edwin Williams, Edgar Zurif, and various anonymous referees. A seminar at Brandeis University in the fall of 1 994 was of great help in solidifying the overall point of view and developing the material.
Iklh J ack en do ff’s energy and enthusiasm in collecting the Wheel of l’iJrlune corpus was an important spur to the development of chapter 7. Katharina Hartmann r ead an early draft of most of the book and sug gl:stcd some important structural changes. Comments on the penultimate draft from Robert Beard, Paul Bloom, Daniel Biiring, Peter Culicover, I >an Dcnnett, Pim Levelt, lda Toivonen, a nd Moira Yip were instru nwntal in tuning it up into its present form. Anne Mark’s editing, as ever, ? Illoothed out much stylistic lumpiness and lent the text a touch of class. My first hook. Semantic Interpretation in Generative Grammar, ac knowledged my debt to a numher of individuals in reaction to whose xvi Acknowledgments work my own had developed.
I am pleased to be able to thank again two of the central figures of that long-ago Generative Semantics tradition, Jim McCawley and Paul Postal, for their many insightful comments on aspects of the work presented here. It is nice to know that collegiality can be cultivated despite major theoretical disagreement. I also want to thank three people with whom I have been collaborating over the past few years, on projects that play important roles in the pres ent study. Work with Barbara Landau has deeply influenced my views on the relationship between language and spatial cognition, and conse quently my views on interfaces and Representational Modularity (chapter 2). Peter Culicover and I have been working together on arguments that binding is a relationship stated over conceptual structure, not syntax; I draw freely on this work in chapter 3.
James Pustejovsky’s work on the Generative Lexicon, as well as unpublished work we have done together, is a major component in the discussion of semantic composition (chapter 3) and lexical redundancy (chapter 5). It is a pleasure to acknowledge financial as well as intellectual support. I am grateful to the John Simon Guggenheim Foundation for a fellowship in 1 993- 1 994 during which several parts of this study were conceived, though not yet molded into a book; this research has also been supported in part by National Science Foundation grant IRI 92- 1 3849 to Brandeis University. Finally, as usual, Amy and Beth are the foundation of my existence.
Much research in linguistics during the past three decades has taken place in the context of goals, assumptions, and methodological biases laid down in the 1 960s. Over time, certain of these have been changed within various traditions of research, but to my knowledge there has been little thorough examination of the context of the entire theory. The problem of keeping the larger context in mind has been exacerbated by the explosion of research, which, although it is a testament to the flourishing state of the field, makes it difficult for those in one branch or technological frame work to relate to work in another.
The present study is an attempt to renovate the foundations of linguis tic theory. This first chapter articulates some of the options available for pursuing linguistic investigation and integrating it with the other cognitive sciences. In particular, I advocate that the theory be formulated in such a way as to promote the possibilities for integration. Chapter 2 lays out the basic architecture and how it relates to the architecture of the mind more ge n e rally. The remaining chapters work out some consequences of these b oundary conditions for linguistic theory, often rather sketchily, but in en ough detail to see which options for further elaboration are promising and which are not.
One of my methodol ogical goals in the present study is to keep the arguments as framework-free as possible-to see what conditions make the most sense no matter what machinery one chooses for writing gram mars. My hope is that such an app r oach will put us in a better position to evalu ate the degree to which di ffe re nt frameworks reflect similar concerns, and to scc what is essential and what is accidental in each framework’s way of going about formulating linguistic insights. 2 Chapter 1 1. 1 Universal Grammar The arguments for Universal Grammar have by now become almost a mantra, a sort of preliminary ritual to be performed before plunging into the technical detail at hand.
Yet these arguments are the reason for the existence of generative grammar, and therefore the reason why most of today’s linguists are in the profession rather than in computer science, literature, or car repair. They are also the reason why linguistics belongs in the cognitive sciences, and more generally why linguistics concerns people who have no interest in the arcana of syllable weight or excep tional case marking. These arguments are due, of course, to Chomsky’s work of the late 1950s and early 1960s. Far more than anyone else, Chomsky is responsi ble for articulating an overall vision of linguistic inquiry and its place in larger intellectual traditions.
By taking Universal Grammar as my start ing point, I intend to reaffirm that, whatever differences surface as we go on, the work presented here is down to its deepest core a part of the Chomskian tradition. 1. 1. 1 The Mentalist Stance The basic stance of generative linguistics is that we are studying “the nature of language,” not as some sort of abstract phenomenon or social artifact, but as the way a human being understands and uses language. In other words, we are interested ultimately in the manner in which language ability is embodied in the human brain. Chomsky makes this distinction nowadays by saying we are studying “internalized” language (I-language) rather than “externalized” language (E-language).
Generative grammar is not the only theory of language adopting this stance. The tradition of Cognitive Grammar adopts it as well, Lakoff (1990), for instance, calling it the “cognitive commitment. ” On the other hand, a great deal of work in formal semantics does not stem from this assumption. For instance, Bach (1989) asserts Chomsky’s major insight to be that language is a formal system-disregarding what I take to be the still more basic insight that language is a psychological phenomenon; and Lewis (1972), following Frege, explicitly disavows psychological concerns. and social aspects of language? One can maintain a mentalist stance without simply dismissing them, as Chomsky sometimes seems to.
It might be, for instance, that there are purely ab stract prope rties that any system must have in order to serve the expres- What about the abstract Questions, Goals, Assumptions sive purposes that language serves; and there might 3 be properties that language has because of the social context in which it is embedded. The mentalist stance would say, though, that we eventually need to investigate how such properties are spelled out in the brains of language users, so that people can use language. It then becomes a matter of where you want to place your bets methodologically: life is short, you have to decide what to spend your time studying.
The bet made by generative linguistics is that here are some important properties of human language that can be effectively studied without taking account of social factors. Similar remarks pertain to those aspects of language that go beyond the scale of the single sentence to discourse and narrative. Generative gram mar for the most part has ignored such aspects of language, venturing into them only to the extent that they are useful tools for examining intrasentential phenomena such as anaphora, topic, and focus. Again, I am sure that the construction of discourse and narrative involves a cog nitive competence that must interact to some degree with the competence for constructing and comprehending individual sentences.
My assump tion, perhaps unwarranted, is that the two competences can be treated as relatively independent. Chomsky consistently speaks of I-language as the speaker’s or knowledge linguistic competence, defending this terminology against various alternatives. I would rather not make a fuss over the terminology; ordi nary language basically doesn’t provide us with terms sufficiently differ entiated for theoretical purposes. Where choice of terminology makes a difference, I’ll try to be explicit; otherwise, I’ll use Chomsky’s terms for convenience.
The Notion of Mental Grammar The phenomenon that motivated Chomsky’s Syntactic Structures was the nlimited possibility of expression in human language, what Chomsky now calls the “discrete infinity” of language. In order for speakers of a language to create and understand sentences they have never heard be fore, there must be a way to combine some finite number of memorized units-the words or morphemes of the language-into phrases and sen tences of arbitrary length. The only way this is possible is for the speaker’s knowledge of the language to include a set of principles of combina tion that determine which combinations are well formed and what they mean. Such principles are a conceptually necessary part of a theory of language. 4 Chapter 1 The finite set of memorized units is traditionally called the lexicon.
The set of principles of combination has been traditionally called the grammar (or better, mental grammar) of the language; in recent work Chomsky has called this set the computational system. Alternatively, the term grammar has been applied more broadly to the entire I-language, including both lexicon and computational system. Given that many lexical items have internal morphological structure of interest, and that morphology has traditionally been called part of grammar, I will tend to use the term mental grammar in this broader sense. How the lexicon and its grammat ical principles are related to the extralexical (or phrasal) grammatical principles will be one of the topics of the present study. Now we come to Bach’s point.
The major technical innovation of early generative grammar was to state the combinatorial properties of lan guage in terms of a formal system. This confers many advantages. At the deepest level, formalization permits one to use mathematical techniques to study the consequences of one’s hypotheses, for example the expressive power (strong or weak generative capacity) of alternative hypothesized combinatorial systems (e. g. Chomsky 1957; Chomsky and Miller 1963) or the leamability of such systems (e. g. Wexler and Culicover 1980 ). At a more methodological level, formalization permits one to be more ab stract, rigorous, and compact in stating and examining one’s claims and assumptions.
And, as Chomsky stressed in a much-quoted passage from the preface to Syntactic Structures, a formalization uncovers conse quences, good or bad, that one might not otherwise have noticed. But formalization is not an unmitigated blessing. In my experience, an excessive preoccupation with formal technology can overwhelm the search for genuine insight into language; and a theory’s choice of formalism can set up sociological barriers to communication with re searchers in other frameworks. For these reasons, I personally find the proper formalization of a theory a delicate balance between rigor and lucidity: enough to spell out carefully what the theory claims, but not too much to become forbidding.