As Time Goes By: Tense and Universal Grammar

This book consists of an analysis, solidly within the framework of Government and Binding, of tense and its interaction with adverbials, temporal connectives, and complementation. There is both good news and bad news about the book; parts of it are carefully worked out and are therefore a valuable resource for the computational linguist, while other parts are less clearly argued. Like much current GB literature, Hornstein takes the goal of linguistic research to be the discovery of the nature of the innate language capacity; this is, of course, a laudable goal. Unfortunately, however, the argument that is presented takes a form that will no doubt prove bewildering to many. The argument, which pervades the book and forms the backbone of its first few chapters, resembles other arguments that have been presented within the same framework: a list of sentences is presented, some ungrammatical and starred; a formal model (one of many possible formal models) of some aspect of the structure of these sentences is presented, with rules for determining ill-formedness in the model and predicting ungrammaticality in the corresponding string; an argument is made that the nature of the formal model is such that it could not be acquired on the basis of primary data available to the child; and the claim is then made that the formal model that has been presented is (in fact, must be) a part of the innate linguistic competence that humans possess. There are several problems with this sort of argument. First, there are often a number of nonequivalent formal models that are consonant with the same set of data. Some of these models are more appealing than others; coming up with one model that covers the data does not in itself constitute proof that that particular model is the one that is innate and should be adopted. Indeed, an appeal to innateness has an insidious consequence: it essentially obviates the need for criteria for distinguishing between possible formal models of the same data. If any formal model can be claimed to be part of the innate language apparatus, there is no need to search for a simpler or cleaner or more parsimonious model than the one that has been proposed; any model is as good as any other, since all models can be innate. Second, the assumptions that are necessary to the claim that the model is unlearnable are not always thoroughly substantiated. In particular, Hornstein claims that "of the data theoretically available to the child, it is likely that only the simple sentences can be absorbed . . . . The child.., is limited to an informationally restricted subset of the potentially relevant data" (p. 2). This claim is made without citation of substantiating psycholinguistic or language acquisition literature; Hornstein stipulates, in effect, that the system he proposes is innate since it could not be learned on the basis of the subset of data he admits as relevant. Hornstein presents a 'neo-Reichenbachian' analysis of tense, by which he means past, present, and future tense as well as perfect aspect; other questions of the treatment