We can get so comfortable with certain ideas that we forget why we hold them, until someone makes a proposal that turns things upside down: there is no D-structure, or control is raising, to mention two such proposals. John Bowers's proposal in this book fits into this category, literally turning some of our long-held views upside down. In sum, he argues that agents are merged very low, near the verb, with themes merged above them, and he explores the consequences of this idea.
B's proposal appears to run counter to the Aristotelian view of sentences as consisting of a subject and a predicate [VO] (cf. Baker's 2001 verb object constraint), yet predication remains at the core of his work (Bowers 1993, 2001). The locus of predication has moved around over time.
Since the VP-internal subject hypothesis (VPISH), there have been two potential sites for predication, one involving merge positions, with agent as the subject of a transitive verb, and the other involving grammatical positions, with an EPP-determined subject for the sentence. Some have suggested that the subject-predicate relation might exist only within vP in some languages (e.g. Massam 2001a), whereas B here is suggesting that what remains of D-structure is a verbal root with an upwardly extending ordered string of uniformly introduced arguments, so predication takes place only at the higher level through Agree and/or EPP. His view of argument structure evokes nonconfigurationality, in which there is also no VP, yet unlike such analyses (e.g. Jelinek 1984), for B, phrasal arguments constitute the true arguments of the clause and they are strictly ordered according to grammatical principles.
B's work rests on two key points: first, that all argument structure is built through the ordered merging of functional heads, each taking a thematically specific argument in its specifier; and second, that the order of argument merge is universally fixed, with agents merging below themes. B's analysis of a basic transitive clause depends on his claim that the higher merged argument (theme) is local for the lower case relation (accusative from Voi (= Voice)), leaving the lower merged argument (agent) free to raise via EPP to PrP (Pr = Pred), and then undergo Agree with T (T = Tense), thus surfacing as the subject of the clause. His book is a set of arguments for this point of view, examining a range of constructions such as the passive (Ch. 2), affectee constructions (Ch. 3), applicative constructions (Ch. 4), and derived nominals (Ch. 5). The book also contains a brief appendix (Appendix A) that provides a compositional semantics for his analysis and another (Appendix B) that discusses the formal aspects of labeling and selection. In the rest of this review I outline each chapter of the book in turn, ending with some potential problems for B's view of argument structure.
In his introductory chapter, B presents an overview of his 'radically different idea' (1) in which all arguments and modifiers are introduced uniformly by functional heads in accordance with a UNIVERSAL ORDER OF MERGE (UOM). Primary arguments are Agent, Theme, and Affectee, which are merged in this order, opposite to the norm. In addition to these arguments, there are secondary arguments (e.g. Instruments), and modifiers (e.g. Manner), also merged in accordance with the UOM. B outlines and counters the reasons why agents are traditionally merged high. His view is post-government and binding, in that syntactic structures provide the lexical semantics of the sentence, rather than being projected from it (as in Borer 2005). His approach here brings to mind construction grammar, where a given meaning is rigidly associated with a particular syntactic configuration. In the final section of this chapter B argues, against Marantz (1984), that subject idioms do exist (e.g. the lovebug bit NP, cf. Postal 2004). This chapter ends with a brief overview of the UOM and works through sample derivations of transitive, intransitive, passive, locative, [End Page 354] and expletive sentences, using concepts of ordered merge, phases, linear...