Word grammar By Richard Hudson (review)

REVIEWS157 Chap. 7, 'The characterization of transfinite sentences', argues that 'the exclusion of transfinite sentences from the proper domain of linguistics would actually be quite absurd' (157). This is because, granting megasentences at least a 'platonist' existence, they must be part of the domain of some field of inquiry; and since 'the laws/principles governing transfinite sentences are the same as those governing finite sentences' (162), linguistics is the obvious field in which to put them. The final section takes a look at such pressing issues as what to do about rules (e.g. that governing the choice between English a and an) which depend on words having immediate neighbors, in densely packed sentences. REFERENCES Chomsky, Noam. 1980. Rules and representations. New York: Columbia University Press. ------. 1981a. Lectures on government and binding. Dordrecht: Foris. ------. 1981b. Published remarks. Philosophical Transactions of the Royal Society of London 295.277-81. Johnson, David E., and Paul M. Postal. 1980. Arc Pair Grammar. Princeton: University Press. Katz, Jerrold J. 1972. Semantic theory. New York: Harper & Row. ------. 1981. Language and other abstract objects. Totowa, NJ: Rowman & Littlefield. ------. 1984. An outline of Platonist grammar. Talking minds: The study of language in cognitive science, ed. by Thomas Bever et al., 17-48. Cambridge, MA: MIT Press. [Received 9 July 1985.] Word grammar. By Richard Hudson. Oxford & New York: Blackwell, 1984. Pp. 267. $34.95. Reviewed by Ronald W. Langacker, University of California, San Diego Despite the limitations suggested by its name, 'word grammar' is offered as a novel and comprehensive theory of linguistic structure. Hudson, in adopting it to replace his earlier (1976) and very different theory ofdaughter-dependency grammar, laments the difficulty of being right the first time; but he is confident that word grammar represents a great improvement on the former model. Perhaps because his current theoretical outlook has affinities to my own, I certainly agree that he is right about many more things this second time around. Though numerous points of H's conception and analysis are open to serious question, I can recommend his monograph as a thoughtful and detailed attempt to formulate a radical alternative to theories in the generative tradition. It is not without flaws, and is hardly likely to stop the generativejuggernaut in its tracks; but for those who are so inclined, it provides a useful exercise in employing very different modes of thought for conceptualizing the problems of linguistic description. Word grammar is one of several current approaches that make a serious attempt to analyze language and linguistic semantics as integral facets of cog- 158LANGUAGE, VOLUME 62, NUMBER 1 (1986) nitive structure (cf. Jackendoff 1983, Langacker 1986). A language, according to H, is a network of 'entities' related by 'propositions', and need not be different in kind from the more inclusive network representing knowledge in general . The types of entities to which a grammar can refer are: single words (characterized at different levels of abstraction), parts of words, strings of words involved in coordinate structures, word meanings, and elements of the utterance event. The permissible types of propositions (i.e. relationships between nodes in a network) are: composition, the relation between a 'model' and an 'instance' of the class it defines, the 'companion' relationship between words that co-occur, the connection between a word and its 'referent' (i.e. its semantic structure), and the link between a word and the event of its utterance. Both diagrammatic and formulaic representations are employed for networks. If, as H claims, a word grammar refers only to words (and parts of words, etc.), how can it account for syntactic constructions and novel sentences? It does in fact refer to grammatical constructions: they are simply represented as network fragments rather than single nodes. The companion relationship between two nodes indicates their syntagmatic co-occurrence; moreover, through labeling of the connecting arcs, numerous more specific relationships are identified (e.g. subject, dependency, semantic roles). The structure of the stored network therefore embodies information about constructions; and H speaks of a particular fragment as being, for instance, the 'entry' for extraposition . Given the additional devices which he invokes (including Boolean conditions, variables, indexing, and quantification over sets), the descriptive system is clearly a powerful one. H rejects the...