Lexicon and Grammar: The English Syntacticon (review)

As expected from Emonds, this book is full of interest and originality. It sets out to develop a more explicit and coherent view of lexical entries than is normally found in the generative literature and arrives at a number of conclusions that are as important as they are controversial.

The first is that lexical subcategorization is very, very simple. The only characteristic of a complement that an individual lexical item can select is its syntactic feature-set—no semantic roles (theta roles or argument structure), no semantic features, no lexical specifics. For example, his lexical entry for transitive disperse is (89):

(1) disperse, V, + 〈plural〉

The feature plural is purely syntactic although, as he notes, it covers such ‘inherent plurals’ as the crowd (47). The verb simply cannot see that its complement also refers to a (semantic) collectivity.

Given that complements have so many other characteristics, this is a surprising conclusion. He presents a certain amount of evidence, but unfortunately he doesn’t consider the most obvious objections, such as the fact that examples like disperse have been used as evidence for semantic selection. If we say that crowd is ‘inherently plural’, what about its plural crowds, or the pronoun it used to refer to a crowd? How is ‘inherently plural’ different from ‘semantically plural’? It is true that there is a discussion of examples like herd of horses compared with ?herd of dogs (44), but it is very briefand dismisses them as examples of knowledge of the world (like the knowledge that men don’t get pregnant). That surely must be wrong—pregnant men are equally odd in any language, but it is very easy to imagine a language in which groups of dogs and horses share the same name. Selection for semantic features survives, I feel.

Similarly, this is true for semantic relations. Again we can illustrate with disperse, and again he mentions the relevant phenomenon: the unaccusative alternative (47), in which the intransitive subject has precisely the same ‘plural’ restriction as the transitive object. He notes that the object can be moved into an unoccupied subject position, but how do we know, from the entry in 1, that this is possible? It’s true that he shows how a verb feature that he calls location can predict whether an object is theme or location (64), but this does not explain why some verbs alternate as disperse does whereas others (e.g. hit) don’t.

Throughout the book he takes a clear and uncompromising position on semantics: the idiosyncratic meaning of an individual lexical item is not, and cannot be, relevant to its syntax (43); and even if semantics were relevant in principle, we couldn’t use such information because semantic analysis is in such a primitive state (369). Personally I take the opposite view, and I would hope to be able to produce a semantic analysis of disperse that would explain why its argument has to be semantically plural: dispersing necessarily involves the break-up of a collectivity. But no doubt other readers will agree with him.

Another weakness of his theory is that (like many other theories) it does not allow verbs to impose any restrictions on their subjects. This is a serious problem for his analysis of disperse because there is a more obvious analysis than the one he suggests: that the transitive is derived (as causative) from the intransitive, rather than vice versa. Here, too, I think a semantic analysis of dispersing would provide an explanation. Leaving this example aside, there are plenty of intransitive verbs that place semantic (or other) restrictions on their subjects; for example, differ is rather like disperse in requiring a plural subject (or a from complement), so John differed is odd for the same reason as John dispersed. If lexical entries could refer to semantic relations, these problems would disappear and semantic relations would be amply justified.

This example leads to another general claim—that lexical entries must not mention phrases since...