Edge-based clausal syntax: A study of (mostly) English object structure (review)

In this book Paul M. Postal argues that the single object in [V + DP] structures does not invariably instantiate a direct object grammatical relation. On the contrary, it can instantiate at least three distinct relations: direct object (a type 2 object), indirect object (a type 3 object), and sub-object (a type 4 object).¹ This three-way classification is claimed to be crucial for characterizing the distribution of English objects in a variety of different constructions: 'Hence what appear to be from one point of view anomalies in the behavior of direct objects will emerge as nonanomalous, as regularities stated on the distinct relational types' (xvi). Whereas type 2 objects have maximal freedom in a wide range of phenomena, type 4 objects are somewhat restricted (e.g. they do not allow middle formation, object raising, nominalization with of objects, nominalization of incorporated forms, -able suffixation, re- prefixation, and object deletion) and type 3 objects are more restricted still (in addition to the constraints that apply to type 4 objects, type 3 objects do not allow left extraction, complex DP shift, right node raising, gapping, stranding, subdeletion, or in-situ human relative pronoun realization). P's argument is that although English [V + DP] sequences show no constituency differences—as far as VP fronting, VP ellipsis, and other traditional constituency tests go—they cannot be modeled by a uniform phrase structure that involves direct objects in all cases.

Ch. 1, 'Introduction', argues that almost none of the facts about English object syntax dealt with in this book have received an adequate account in mainstream syntax (i.e. any of the various generative, transformational grammar frameworks), even though many linguists have focused on this topic for over half a century. This lack of success is taken to be evidence against the correctness of the transformational research program, and 'the plausibility of such a negative conclusion increases along with the time involved' (2). Perhaps Chs. 2 and 7 are the most critical of past research. P criticizes the lack of theoretical consensus about how to analyze double object constructions and the lack of accounts for a number of idiosyncrasies in passives. In particular, accounts that boast of being able to predict that second objects in double object constructions never passivize are dismissed by P on the grounds that such passivizations have long been known to be possible in some dialects of English.

The theory proposed in this book is framed within METAGRAPH GRAMMAR (Postal 1992, 2004), which, like its relational predecessors, diverges from mainstream syntax in a number of ways. I single out two major differences here. First, it aims to characterize the sentences of a natural language, not the knowledge of that natural language nor the psychological and biological mechanisms that yield such knowledge. A natural language is seen as a class of sets, and each sentence is modeled as a certain kind of set (a ten-ary set called a metagraph, formally equivalent to a multigraph (10, n. 4, and 25-29)). The second point of divergence is that a grammar is taken to be a nonordered set of second-order logic statements about sentences, which every sentence must satisfy. Thus, the set of metagraphs that are grammatical sentences in a given language L is the subset of all possible metagraphs whose members satisfy the constraints imposed by a grammar of L. P rejects the need for metaprinciples (higher-order statements about grammar rules or about derivations) that are common to mainstream syntax on the grounds that these 'can only yield testable linguistic evidence indirectly at best, by virtue of complicated chains of reasoning' (6).

The remainder of Ch. 1 formalizes a number of foundational metagraph concepts. Most is borrowed from previous work, but some is novel (e.g. the notion of complex labels is used to formulate the notion of 'quace' in a novel way). The amount of formal concepts introduced in Ch. 1 is arguably...