PREDICTING SYNTAX FROM SEMANTICS 509semantic representation in its lexical entry together with an independently
motivated set of morphosyntactic principles. Ideally, only the semantic rep
resentation of the verb should be required in its lexical entry. However, we
are well aware of Sapir's dictum, "all grammars leak"(1921:38), and there
fore it is inevitable that some idiosyncratic information will have to be
stipulated for some verbs. Our goal is to uncover independently motivated,
linguistically significant generalizations about the semantic basis of comple
ment systems and thereby to keep such stipulations to an absolute
minimum.
The first step in the analysis is the development of a semantic metalan
guage for the representation of verbs, although in the context of this paper
it will be possible only to sketch the outlines of what such a metalanguage
would look like and to develop detailed representations only for the verbs
in question. It is appropriate to ask, why is such a metalanguage necessary?
In what ways is a Dowty-style minimalist representation scheme
inadequate? The fundamental inadequacy of the minimalist system of
decomposition can be seen most clearly with cognition complement-taking
verbs. In the RRG version of Dowty's system, state and activity predicates
are treated as primitive, and consequently a verb like remember would be
represented simply as remember'. But remember can have three different
interpretations in terms of the Dowty scheme: achievement, signalling an
inchoative activity, as in John suddenly remembered the faucet he left on;
second, activity, as in John consciously remembered the names of all of the
linguists that he met at the party; and third, stative, as in John remembers his
first day at school very vividly. The stative sense of remember is in fact an
entailment of the activity sense that can result from the achievement sense,
i.e. the consequence of the completion of the beginning of an activity is the
activity. This accords with the fact that all stative interpretations of
remember can also have activity interpretations, but not vice versa. The sta
tive readings follow from the fact that in remembering a person starts to
actively think about something, and for the duration of this activity there is
an entailment that the person has this something in mind. This entailment
would be difficult to capture in Dowty's system because he provides no
mechanism for relating non-motion activities to non-locative states. The
achievement sense of remember would be represented as BECOME
remember'; the accomplishment verb based on remember is remind, which
would have the logical structure [...] CAUSE [BECOME remember'...].