Advances in Role and Reference Grammar

502 ROBERT D. VAN VALIN, JR. & DAVID P. WILKINS

clude that while lexical entries must specify s-selection (as part of the

semantic characterization of an item) and transitivity, they need not specify

c[onstituent]-selection"(1986:90). This is an important move, but as long as

the predictions are based on the canonical structural realizations of the ele­

ments in the arguments structure of a verb alone, they will be severely

limited. For example, persuade takes two kinds of complement S, the infin­

itive of persuade Bill to leave and the tensed S oî persuade Bill that the earth

is flat, if these are both simply represented as "proposition" in the verb's

argument structure, then the crucial contrast between these two structures

will have to be stipulated in an ad hoc way, thus drastically reducing the

explanatory value of the approach.

Another attempt to predict syntactic facts from the semantics of predi­

cates is the system of predicate semantics developed in Dowty (1979) as

employed in RRG. The primary work in this area in Foley & Van Valin

1984 (henceforth FVV) concerns predicting the prepositions that cooccur

with particular verbs. It was demonstrated that given the RRG system of

lexical representation and the theory of semantic relations and grammatical

relations, the occurrence of to, from and with with particular verbs could be

predicted, and hence these prepositions need not be listed in the lexical

entry of the verbs in question. (See FVV, section 3.3.1, and also Jolly, this

volume, for an expansion of this analysis.) The possibility of deriving the

types of complements that particular verbs take from their semantic rep­

resentation is raised in FVV but not explored (p. 315).

1.2 Semantic decomposition

As far as semantic decomposition is concerned, there are two classes of syn­

tactic theories: those in which there is no semantic decomposition of predi­

cates and those which have some degree of predicate decomposition. In the

first class lie GPSG, LFG, most work in Montague Grammar [MG] aside

from Dowty (1979) and GB. In GPSG (Gazdar et al. 1985), LFG (Halvor-

sen 1983) and MG (e.g. Bach 1979) predicates are treated as logical con­

stants, and the representation of a predicate such as "kill" would simply be

kill'(x,y). In GB, which does not employ intensional logic in its semantic

representations and therefore does not use this notation for representing

predicates, there does not appear to be as yet any standardized format for

representing predicates semantically.