The Syntax of Adjuncts (review)

In his 1972 monograph Semantic interpretation in generative grammar, Ray Jackendoff begins the chapter on adverbs saying, ‘the adverb is perhaps the least studied and most maligned part of speech, . . . maltreated beyond the call of duty’ (47). Thirty years later, adverbs and adverbials are finally receiving the needed attention, and Thomas Ernst’s book is an important and ambitious contribution to the recent body of work. Its goal is a comprehensive theory of adverbial distribution: the base positions and licensing of a wide range of adverbs and adverbial expressions.

The book is transparently organized and well written and will be accessible to those with a good understanding of current syntactic and semantic theory. It is data-rich, with most of the evidence coming from English, supported by examples from Italian, French, and Chinese. In many places, E is very careful to formulate descriptive generalizations, which any theory of adverb placement will need to address. It is thus of interest to those with a wide range of theoretical and descriptive concerns. For those who like clearly laid out ideas, interesting predictions, and deft argumentation, this book is also a must. The analytical proposals are consistently argued for, rather than simply asserted, and E attempts to show how his proposals make sense within current syntactic theory. In what follows, I briefly review the contents of the nine chapters in an effort to give a flavor for his overall approach as well as a few of the particulars.

Ch. 1 provides a bird’s eye view of the work and outlines the main theses within the context of earlier work. E’s theory is largely semantic and there are four modules that interact to account for adverbial distribution: (1) the lexicosemantics of individual adverbs—lexical specifications for individual adverbs indicating the type of semantic object(s) (speech act, proposition, fact, event, specified event, etc.) that the adverb takes in its scope; (2) a compositional-semantics rule system—a set of rules ‘which takes the basic event and builds “layers” of event types and proposition types until the representation of the proposition for the whole sentence is completed’ (35); (3) weight theory—preferences or dispreferences for certain constituent orders based on the ‘weight’ of constituents in a sentence; and (4) directionality principles—syntactic principles governing the direction of phrase structure construction, including a head-initial/head-final parameter. In E’s system, adverbials are simply syntactic adjuncts and their distribution is determined largely by their lexical semantics and a compositional semantic interpretation system. Adverbs adjoin to syntactic projections that correspond to the semantic objects specified in their lexical entries. Various illicit adverbial positions are filtered out by the four modules.

Ch. 2 illustrates the system by providing an in-depth semantic analysis of nonquantificational, predicational adverbs (modal adverbs, subject-oriented adverbs, speaker-oriented adverbs, and exocomparatives such as similarly or likewise). Such adverbs have semantic selectional requirements, taking speech acts, facts, propositions, events, or specified events as arguments.

(1)

a. Frankly, [why would he do such a thing]?	(speech act)
b. Unfortunately, [she lay down on a scorpion’s nest].	(fact)
c. Sam probably [has made an appointment].	(proposition)
d. Kim intelligently [bought the tickets].	(event)

A main goal of the chapter is to account for the well-known clausal/manner ambiguity of such adverbs without resorting to homophony.

In Ch. 3, E shows how his semantic theory of adverb licensing accounts for a wide range of traditionally syntactic facts: adverb positioning and ordering restrictions with which most theories are concerned. The crucial ingredients are the lexical semantics and compositional rule system (modules 1 and 2 above). To illustrate, an evaluative adverb like probably must occur above a subject-oriented adverb like unfortunately, because the former takes a fact as its argument while the latter takes an event and an event may be converted into a fact by the rule system (2a). In the reverse ordering, unfortunately selects a fact but the rule system cannot then convert this...