American Scientist

Foundations of Language: Brain, Meaning, Grammar, Evolution. Ray Jackendoff. xx + 477 pp. Oxford University Press, 2002. $40.

What is language for? A nonlinguist would probably reply, "For expressing meaning by means of sound (or gesture)." If this commonsense answer is right, we should expect semantics (the study of meaning) to be at the heart of linguistic theory. It comes as a surprise to most beginners in contemporary mainstream linguistics when they find that, instead, the central component of language is presented as syntax. Semantics is not even in second place; what comes next in respect to time devoted to it in linguistic curricula is phonology (the study of speech sounds). No wonder linguistics can seem such a dry discipline: The aspect of language that to a nonexpert seems most important, namely the substance of what it can convey, is downgraded in favor of the austere technicalities of conveyance. This is true not only of the Chomskyan approach that has been dominant since the 1960s, but also of the structuralist approaches that preceded it.

In Foundations of Language, Ray Jackendoff challenges this dominant, syntactocentric view. Semantics, he maintains, is not just a handmaiden of syntax, humbly interpreting structures that are generated elsewhere. Understanding what John kissed Mary means is not just a matter of slotting the words John, Mary and kiss into a syntactic frame [NounPhrase [Verb NounPhrase]]. Rather, semantics, or conceptual structure, to use Jackendoff's term, has a generative role in its own right. In conceptual structure, KISS combines with two objects, JOHN and MARY, to form the event [KISS (JOHN, MARY)]. Conceptual representations such as [KISS (JOHN, MARY)], syntactic representations such as [NounPhrase [Verb NounPhrase]] and phonological representations such as Jòhn kìssed Máry have what Jackendoff refers to as parallel architecture, in that all are generated by formation rules of their own and are linked by interface rules. Much of the book is devoted to exploring the interface rules that link syntax and semantics.

Jackendoff is not the first linguist to challenge syntactocentrism. For decades, various rival approaches that might be called semantocentric have argued that the syntactic structure of a sentence is in some fashion derivable from its meaning. These approaches challenge, in greater or lesser degree, Noam Chomsky's view of language as essentially separate from the rest of the human cognitive apparatus. But Jackendoff argues against giving primacy to meaning, on several grounds. For example, nothing in the conceptual structure [DEFEAT (CAESAR, GAULS)] explains why it can be linked to two different kinds of syntactic structure—a sentence (Caesar defeated the Gauls) and a noun phrase (Caesar's defeat of the Gauls). What makes Jackendoff's work both interesting and refreshing is that his challenge to syntactocentrism is mounted from a viewpoint fundamentally sympathetic to Chomsky's view of language. Jackendoff agrees that language stands substantially apart from the rest of cognition, even while questioning Chomsky's view of how the language faculty is organized.

Sympathy with Chomsky's approach to language often goes along with a lack of interest in other approaches, such as those explored by psychologists in recent years under labels such as connectionism and parallel distributed processing. But in this respect, too, Jackendoff is not a typical Chomskyan. He is keen to build bridges between research on grammar pure and simple and research that involves modeling or exploring directly what happens in the brain when language is used. He is thus not content with the doctrine that a firm line can be drawn between linguistic competence as an abstract system and the way in which this competence is implemented in human brains, with only the former being of concern to linguists.

In particular, Jackendoff is interested in psychological and neurological questions about the lexicon—that is, about what linguistic items are stored or memorized as units, and about what relationships can exist between one stored item and another, and between them and linguistic expressions that are "constructed online" from their constituent words in working memory. He shows that various seemingly promising answers to these questions are wrong. In particular, stored items are not necessarily words, and words are not necessarily stored. An item bigger than a word that is necessarily stored is an idiom, such as red herring or kick the bucket. A word that is not stored is one that is complex, in that it contains more than one element, such as dogs (made up of dog and -s), but which is formed in a regular fashion and whose meaning is entirely predictable. An example of a complex word whose shape is stored, because it is irregular, is the plural teeth; and one whose meaning is stored because it is unexpected is scissors, which does not mean "more than one scissor." As one might expect, many words are stored for both reasons, such as commitment. Speakers of English just have to learn that commit accepts the suffix -ment whereas admit and submit, for example, do not, and just have to learn also that commitment does not mean "commission" (as in the commission of the crime). But Jackendoff makes the point that the lexicon of stored items may contain many whose formation is regular and whose meaning is predictable, but which the brain nevertheless seems to prefer to access "ready-made," so to speak, because of their frequency of use. In saying this, Jackendoff insists on the theoretical importance of an aspect of language that would be relegated by many linguists to the domain of "performance" rather than "competence," or to the domain of implementation rather than abstract structure.

At first sight, there is no obvious link between Jackendoff's parallel-architecture view of language and his interest in the psychology of lexical storage. A link is established through what is perhaps the most startling proposal in the book. Some lexically stored items have empty slots, namely idioms with gaps such as to take X to task or to be Xed out, meaning "to be weary from too much X" (as in I was conferenced out after four days or Fred was Beethovened out after hearing all nine symphonies in one week). Jackendoff proposes that every syntactic construction, such as the English construction whereby a sentence can consist of a noun phrase followed by a verb phrase, is simply an idiom in which all the slots are empty. Thus, to the question whether linguistic rules exist alongside stored items—a question much disputed among psychologists, neuroscientists and linguists—Jackendoff answers no, but he does so for a novel reason. For connectionists, all regularity is a matter of degree, so a rule is merely a widely instantiated pattern of resemblance between stored items. For Jackendoff, by contrast, the pattern itself is a kind of stored item. The basic clausal structure [NounPhrase [Verb NounPhrase]] is an idiom, just like The cat got his tongue, the only difference being that the former is instantiated in a vast number of versions, and the latter in just one. Jackendoff's proposal thus promotes the lexicon from the periphery of linguistic theory to its very center, despite the fact that lexical knowledge is the aspect of language that is subject to most variation between individuals. Will this proposal prove sustainable? I expect it to provoke lively discussion.

A second novelty is Jackendoff's interest in how language has evolved. Most linguists have refused to discuss language evolution, on the grounds that one can do no more than speculate about it. But Jackendoff is not so pessimistic. If (as he claims) syntax and semantics are structured differently, this cries out for an explanation—it seems more natural that syntax should reflect semantics rather directly. Jackendoff offers a few hints toward an explanation in terms of linguistic prehistory; his proposals in this area are tentative, but he is certainly right in thinking that the question of why language has come to be as it is is one that linguists cannot permanently ignore.

Nearly all of the book should be comprehensible to readers with some background in linguistics, even if they are a bit rusty. Readers with no such training but some knowledge of cognitive science or psychology may find it harder going but still rewarding, because of Jackendoff's keenness to build bridges between disciplines. His breadth of knowledge and soundness of judgment, along with just the right amount of adventurousness, make for a book that deserves to be read and reread by anyone seriously interested in the state of the art of research on language.