This book "supplies a vocabulary of English words and idiomatic phrases 'arranged … according to the ideas which they express'. The thesaurus, continually expanded and updated, has always remained in print, but this reissued first edition shows the impressive breadth of Roget's own knowledge and interests."
Cormack, Annabel (1998) Definitions: Implications for Semantics, Syntax and the Language of Thought. Garland Publishing, New York and London.
Reviewed by Heidi Harley, University of Arizona
This book, one of the useful series of Outstanding Dissertations published by Garland, is an unmodified version of Cormack's 1989 University College London doctoral dissertation. Some dissertations lose their relevance if not read in the context of the theoretical trend du jour; this is not one of them. Indeed, in some respects, it anticipates discussion that followed in the subsequent decade; in others, it addresses foundational questions that will remain relevant to any approach to theoretical semantics and syntax. Before summarizing the material, let me comment on two particularly surprising aspects of Cormack's general approach. Cormack's primary source of data, dictionary definitions, is, to say the least, unusual if not unprecedented in a theoretically-oriented dissertation. Upon initial reaction, dictionary definitions qua semantic data felt tainted by my intuition that dictionary definitions tell you only what is least theoretically interesting about word meaning. Children, after all, learn the meanings of words without ever consulting a dictionary, and I assume that in fact by far the bulk of adult word-learning goes on without the help of a dictionary either. From the point of view of theoretical linguistics, then, it had seemed to me that there were more important things to worry about than what a dictionary might say about a word. On the other hand, I can see that there's a crucial flaw in that stance: I use dictionaries all the time, and I'm not the only one. There are even fields, including theoretical syntax and semantics, where precise definition of terms is crucial for theory-creation. Cormack (perhaps because of her background in mathematics, where definitions are even more central than linguistics) adopts the radical position that definitions actually do what they're supposed to: allow you to understand the meaning of a new term. Since the sound-meaning connection is the essence of language, dictionary definitions then ought to shed light on some questions that are pretty central to linguistic investigation. Cormack was insightful enough to recognize this, and gets a lot of mileage out of her otherwise theoretically virginal data set. The second way in which Cormack's work is surprising is that she explicitly addresses the question of the construction of representations in Fodor's Language of Thought (LoT). That is, she argues that the properties of definitions can provide insight into the actual syntax and semantics of the LoT. That in itself is perhaps not remarkable. What is surprising is that she couches her semantic discussion in model-theoretic terms; most model-theoretic semanticists avoid explicitly addressing the question of whether and how their proposed analyses are mentally represented, perhaps for philosophical reasons, perhaps for practical ones. Fodor himself (I think wrongly) asserts that the necessary externalness of model-theoretic representations means that they don't stand a chance of being appropriate tools for analyzing the LoT. Cormack, dealing with the supremely external dictionary definition, does a convincing job of using model-theoretic tools to make assertions and predictions about the nature of the supremely internal LoT. Cormack's dissertation is written in five fairly long chapters. Chapter 1 introduces the theoretical syntactic and semantic frameworks she adopts as a starting position: essentially Government-Binding theory and model-theoretic semantics. She also outlines the Relevance Theory of Sperber and Wilson and later uses it to address some pragmatic issues; it seemed, however, that there were fewer predictive consequences of Cormack's investigation for Relevance Theory than for the other two (and, to boot, I'm not particularly qualified to comment on it anyway). Chapter 2 begins the investigation with a consideration of off-the-rack dictionary definitions: she assumes that they have a coherent syntax and a coherent semantics, and that they can serve to eliminate the unknown term at some level of representation: perhaps in a natural-language level which Cormack terms LF', or perhaps at LoT itself. Here she addresses questions of type/category mismatch, attributive vs. predicative uses of adjectives, and the problem that non-intersective adjectives pose for the distinction between internal and external arguments. Chapter 3 involves definitions found in running text, like the following: "a 'convex' deltahedron is one in which all the vertices point outwards," or "a mammal suckles its young." Here Cormack faces the problem of identifying the elements which constitute the definitional material. She notes that there are different types of textual definitions: ones which make use of the word "call" (most like dictionary definitions), ones which use "say" (with consequences for the treatment of genericity), and complex definitions with "if" and "when" (with consequences for donkey sentences). Chapter 4 treats syntactic issues raised in the preceding two chapters, in particular, the question of what types of mismatch between c-selection and s-selection may arise. Here, the main subject is the different types of complements which adjectives permit, and the fundamental issue at stake is the question of learnability: how can we constrain the set of possibilities such that all the different possible configurations are learnable? In passing, she treats control constructions extensively , as well as ECM, Raising, tough-movement and expletives. Finally, in Chapter 5, she discusses the implications of the previous four chapters for the LoT. Given her approach to elimination, it is essential that the type system is common to both natural language and the LoT. Further, as for Fodor, NL words correspond to LoT 'words', more-or-less one-to-one. Elimination of an undefined term probably proceeds after all natural language processing is over -- that is, a definition is an instruction to create a new LoT concept, and fix its position via biconditionals in relation to other already understood concepts. (Here, Cormack has concluded that Fodor's long battle for strict nativism (babies born with the concept CARBURETOR) has essentially been misguided. In fact, even Fodor has recently concluded the same thing: concepts, he has recently conceded, are now in fact learnable. What Cormack does is provide a sketch of how such learning could take place through definitions, rather than the canonical ostensive learning.) Cormack's thesis is overall a remarkably finished piece of work, especially given that she wasn't able to revise it before its publication by Garland. As I said earlier, much of her discussion is as relevant now as it was in 1989. On some points, however, it would have been interesting to see how she would have incorporated later developments. A central theoretical device she proposes is a distinction within type theory between external and internal arguments: rather than simply the canonical type for predicates <e,t>, predicates which must take an external argument have type <e*,t>, where the * is intended to indicate the external nature of the argument's syntactic position. Given the explosion of research on the 'split-VP' hypothesis, according to which external arguments are not the arguments of the root V but rather of a light verb, 'little v' or PredP or some such, and the neo-Davidsonian semantics such an approach necessitates, Cormack's notational device has most likely been supplanted. Another question which the advent of the Minimalist program raises for Cormack's discussion is the status of the ECP. She points out, for example, that ECM seems strangely relevant to the construction of legitimate definitions. There are definitions of verbs like 'to launch' of the form "to cause to slide into water," where the object of 'launch' is the subject of the embedded infinitival clause in the definition. She points out that while "to believe to catch mice" seems intuitively like a possible definition, "to believe catches mice", with the same corresponding semantic gap, does not. This corresponds to the status of heavy-NP shift out of these constructions: "*I believe exists more than one solution to the problem." The heavy-NP shift, being rightward movement, produces a violation of the ECP in this configuration, and, Cormack argues, the operator movement which is attempting to produce the subject gap in "to believe catches mice" is also to the right, and subject to the ECP. Hence the ill-formedness of "to believe catches mice" as a definition. (She argued earlier for operator movemnt to create an object gap in a definition of 'putative', "supposed to be".) This account, relying as it does on both the notion of linearity at LF (rather than purely structural relations) and the ECP, would need to be reworked if the proposal were to be recast in a Minimalist frame, without government. In addition to addressing a number of major issues in syntactic and semantic theory, Cormack provides interesting discussion of some relatively minor points. One such interesting point for me was the question of why despite being predicative in nature, don't Ns or NPs turn up as definitions for Adj? All the other predicative categories do: Adj-AdjP ('deaf' = "inattentive", 'dainty' = "hard to please"), Adj-PP ('dark' = "with little or no light"), Adj-Rel ('decisive' = "that decides an issue"), Adj-Ppl ('mouldy' = "covered with mold"). The obvious thing to say is that any phrase of type <e,t> may appropriately define an (intersective) adjective; why should this generalization break down in the case of N's? The answer to the small question of why no N's as definitions of Adjs is clearly that N's may not act as modifiers. This question, Cormack asserts, is part of a bigger one: the question of why the head-modifier relation exists in the first place. The representation of some property as an N vs. its representation as an Adj has to do with its perceived utility in identifying individuals: i.e. its stability with respect to a particular individual. In that respect, "appleness" is a much more stable, hence identificational, property than "redness" is. Heads and adjoined modifiers get their combined interpretation essentially through semantic conjunction: a red apple is red and an apple. So: why isn't a lawn mower both a lawn and a mower? Cormack speculates that it is because when one wants to communicate notion of the set formed by the conjunction of two N predicates, syntactic conjunction is available, so the interpretively trickier process of adjunction, which is used for modifiers, is not necessary. This type of explanation is interestingly reminiscent of more recent Economy-motivated explanations in syntax, but crucially rests on a distinction between adjunction and conjunction with respect to semantic transparency; again, it would be very interesting to see this pursued in a Minimalist framework. My main beef with Cormack's dissertation is its sheer density. It's well-written, with astonishingly few typographical errors, but her expository style makes things very hard on the reader. For instance, in Chapter 4, she introduces a table of adjectives their c-selection and s-selection types, numbered 1 through 16, and then refers to them pretty much solely by number through the rest of the 70-page chapter; that is symptomatic of the overall style. It would be fair to say that in many places the prose is so cryptic that it requires a vigorous conscious effort to follow the argument. That's not to say that the argument isn't there: it (nearly) always is. But you have to be a very dedicated reader to follow every twist and turn. It's not designed for causal flipping. Further, the reader should bring to the table a reasonably well-grounded knowledge of the mechanics of model-theoretic semantics; Cormack's exposition of it in Chapter 1 is unproblematic, but a quick reading of the outlines of the theory will not equip one with the necessary tools to understand all the proposals. A reader who satisfies the determinedness and model-theoretic prerequisites, however, will find Cormack's thesis methodologically innovative, theoretically interesting, and intellectually challenging.
Heidi Harley is an assistant professor at the University of Arizona. Her research interests include argument structure, Case theory, lexical semantics and morphology. For more info, visit http://w3.arizona.edu/~ling/hh/
- ------------------------------------------------------------------- Heidi Harley (520) 626-3554 Department of Linguistics firstname.lastname@example.org Douglass 200E Fax: (520) 626-9014 University of Arizona Tucson, AZ 85721