PREDICTING SYNTAX FROM SEMANTICS^529derived from general, independently motivated semantic, lexical, and mor-
phosyntactic principles. It is clear from this analysis that attempts such as
that in Chomsky (1986) to predict syntactic structure solely from a thema
tic-relations based notion of argument structure are doomed to failure.
Nevertheless, this paper shows how Chomsky's (1986) goal of deriving
information about the c-selection properties of lexical items from their s-
selection properties can be realized, albeit in ways that he would not neces
sarily countenance.
In the area of lexical semantics, this paper adds to the theoretical
research on non-Indo-European languages and to the very small body of
comparative lexical semantic studies of markedly different languages in
some well-defined domain, in this case cognition verbs. The semantic
metalanguage we are developing, along with the semantic redundancy rules
which play a major role in it, will be significant in several respects. First,
since they will be developed through comparative studies, they will have
cross-linguistic applicability. Second, the use of a basic natural language
metalanguage should allow intuitive verification of the definitions by native
speakers without the need to learn a highly formal set of symbols and logic-
based representations. This does not, however, negate the possibility of
translating the proposed representations into the notation of formal seman
tic theories. Indeed, since the decomposition program starts from the
Dowty decomposition system, it is directly applicable to any formal seman
tic theory like Dowty (1979). The replacement of remember' (x,y) by the
semantic structure presented in section 2 will have no affect on the higher
level combinations which this logical constant enters into. In addition, the
tests which are being developed and refined for the selection of the basic
semantic components in the metalanguage will be useful to any semantic
theory regardless of how those components and their relations are to be
represented notationally. If, for example, one can show in a non-ad hoc
way that think underlies remember, then all decompositional approaches
must be able to capture this. Third, this analysis is itself an argument for
decompositional approaches to semantic representation. The generaliza
tions proposed regarding remember must be captured by any descriptively
adequate semantic theory, and this analysis stands as a challenge to non-
decompositional theories. Finally, even though the metalanguage is being
constructed to deal with complement-taking predicates in specific semantic
domains, it nevertheless will have varying degrees of applicability to other
predicates and construction types, and future work will be directed toward
extending it.