PREDICTING SYNTAX FROM SEMANTICS 503Theories of the second type include that of Jackendoff (1972,1976), an
elaboration of Gruber (1965), and RRG, which is based on Dowty (1979).
The Gruber-Jackendoff system and the Dowty system are lexicai-decom-
position-based analyses. The former is a system of abstract predicates from
which thematic relations are derived, e.g. GO (x,y,z), with χ = theme, y =
source, and ζ = goal. Dowty's approach is similar but is founded on the
verb classification scheme proposed in Vendler (1967); it has the advantage
that, unlike the Gruber-Jackendoff system, it provides independent tests for
assigning a verb to a particular class, which allows a non-ad hoc decision as
to what basic predicates more complex predicates are composed of. How
ever, Dowty does not attempt to derive thematic relations from the argu
ment positions in the abstract lexical structures. The RRG system of lexical
decomposition represents a unification of aspects of the Gruber-Jackendoff
system with the Dowty system, in that thematic relations in RRG are
defined in terms of argument positions in the decomposed semantic struc
tures, as done in the Gruber-Jackendoff system, but the decomposition is
based on the Dowty decomposition system. Jackendoff s recent semantic
theory (1983, 1985), "Conceptual Semantics", bears many of the features of
his earlier work: it is a decompositional approach, and thematic relations
are still defined in terms of argument positions in semantic structure. He
argues for a notation which represents alternative realizations of a variable
in semantic representation in lexical entries. While the alternatives should
be stated somewhere, they are not part of the semantic decomposition of a
predicate and therefore should not be listed in the lexical entry. With
respect to the relationship between syntactic subcategorization and seman
tic properties of predicates, he lists both separately in the lexical entry and
links the syntactic positions to argument positions in the semantic represen
tation through indexing; no attempt is made to relate them by any general
principles or to predict the linking in any way. As Jackendoff (1985: 294)
notes, "[e]ventually, of course, one would like to adequately constrain the
theory of syntax-semantics linkages in the lexicon." The present paper
takes up this particular challenge.
FVV (p. 35) have noted that theories of lexical decomposition can be
contrasted in terms of how much of the meaning of predicates they attempt
to represent. On the one hand, there are those which aim to account for
only certain aspects of the meaning of predicates; these we will call
"minimalist" approaches. On the other, there is the "rich" semantic
approach which attempts to represent all aspects of the meaning of predi-