On the nature of grammatical relations (review)

674LANGUAGE, VOLUME 62, NUMBER 3 (1986) On the nature of grammatical relations. By Alec P. Marantz. (Linguistic Inquiry monograph 10. ) Cambridge, MA: MIT Press, 1984. Pp. xi, 339. Cloth 25. 00, paper 15. 00. Reviewed by Robin Clark, UCLA Here Marantz attempts to explicate the role of grammatical relations (e. g. 'subject' or 'object') in linguistic theory. As he notes in Chap. 1, there are only two ways to approach the topic, given its pervasiveness in various linguistic theories. One may either take grammatical relations as derived notions, to be explicated in terms of primitives of the theory; or one may take them as primitives that are simply not definable. The question oftheir definability is therefore a matter of great interest, since the form and functioning of linguistic theory are at stake. M places grammatical relations at the interface between lexical semantics and the expression of semantic relations in the syntax. Thus one may think of such relations as determining the syntactic realization of argument structure. This places M's theory in a class with theories like Government/Binding and Relational Grammar, both of which assume that the proper treatment of linguistic competence involves multiple levels of representation. In fact, M assumes four distinct levels: (1) logico-semantic structure surface structure—syntactic structure phonological structure Logico-semantic (1-s) structure is a representation of the compositional semantics of a sentence; syntactic (s) structure is a representation of the grammatical relations among constituents. In M's terms, this means that s-structure is the syntactic realization of the semantic roles of various constituents. The relationship between the two levels is mediated by 'Principle M', which matches semantic and syntactic dependencies, and can be paraphrased as follows: If an element bears a relation with respect to some other element at 1-s structure, then those two elements must stand in some grammatical relation at s-structure. One may detect here a certain affinity between M's 'Principle M' and the assumption, found (e. g. ) in Montague Grammar, that the syntactic category of an element reflects its semantic type. Finally, surface structure is a representation of the sentence in terms. of constituent government, string adjacency, case-marking, and agreement. The formal machinery necessary to account for the mapping between the various levels of representation is developed in Chap. 2. Particular attention is paid to the relationship between semantic dependencies, reflected at 1-s structure, and syntactic dependencies at s-structure. The correspondence relies heavily on a set of primitive relations (e. g. , X is an argument of Y), a set of features (such as ± logical subject), a form of X-bar theory, government, and Case. Finally, a number ofprinciples guarantee correspondence between the expression of core structural relations at each level of representation. While much of this theoretical apparatus bears a relationship to that of GB theory, there are some important differences. For example, the head of VP will govern the subject position in M's theory—but not, crucially, in GB theory. To his credit, M pays particularly close attention to the formation of complex predicates; and he makes a number of inter- REVIEWS675 esting observations about the assignment ofcomplex semantic roles to the subject of a predication. The principles outlined in Chap. 2 are illuminated by their application to a number of examples drawn from English. A brief analysis of case-marking in Japanese and Icelandic concludes the chapter. Chap. 3 presents an analysis of raising, small clauses, Exceptional Case Marking, and control structures. In M's theory, these constructions involve arguments which appear to bear multiple grammatical relations. Although the chapter presents a system for generating such constructions, a number of questions remain; e. g. , M does not discuss left-branch effects for subjects of small clauses, or explain why such subjects appear to create opaque domains for the purposes ofbinding. The chapter concludes with a discussion of the empty categories assumed by this theory—essentially, pro for control structures, and variables for constructions involving wH-movement. The material presented in Chaps. 4-7 will be ofparticular interest to those involved with research on the interaction of morphology and argument structure, and on the expression of argument structure in the. . .