In this section I will isolate uses of eventive verbs in which the verbs take a full VP complement from the metonymic uses in which they take an NP complement. By assuming distinct lexical entries and therefore distinct logical forms for the different uses of these verbs, I follow the proposals of Godard & Jayez (1993) and Copestake & Briscoe (1995) rather than the type-shifting proposals of Pustejovsky pustejovsky:95a. This analysis is supported by anaphora, relativisation, and coordination data, outlined by Godard & Jayez (1993:169-170) and Copestake & Briscoe (1995:32-33). For example, the coordination of predicates which require complements of different types is possible, as shown in beg99.
John picked up and finished his beer. John ate and enjoyed the caviar. Sam wrote but later regretted the article.
In these examples, the first verbal predicate requires a physical object complement, while the second requires an eventive complement, yet there is no problem with their coordination. This suggests that the type of the NP complement cannot itself be shifted. A parallel example from the relativisation data appears in beg103.
*John began a book which will take two hours. John began the reading of a book which will take two hours.
Here the relative clauses must be relative to an event. If the type of a book in beg103a were shifted to the eventive interpretation made explicit in beg103b, then this example would be grammatical. It is not, however, further evidence that the type of the NP complement is not shifted.
As Copestake & Briscoe (1995:52) point out, however, there is some data which cannot be accounted for under an analysis in which the particular lexical entry for a verb controls the coercion of complement type, shown in beg101a-b.
I enjoy films and mending antique clocks. We found Sam swimming the channel, which he enjoys more than golf. *John began the beer and watching the television. *John finished composing his symphony and his novel.
It is difficult to explain these examples under the assumption that one lexical entry for enjoy combines with NPs and another with VPs, because there is one instance of the verb which appears to combine happily with both. They argue, however, that these examples occur less frequently and are more marked than the examples in beg99. Furthermore, the data in beg101c-d shows that other eventive verbs do not allow this kind of coordination. I will therefore follow Copestake & Briscoe in preferring the polymorphic analysis of the eventive verbs.
Given the isolation of VP-complement and NP-complement forms of the eventive verbs as a starting point, I will investigate the two uses individually. I will propose simple aspectual constraints on the VP-complement forms of the eventive verbs, but the logical metonymy cases will be shown to be far more complex. No aspectual constraints will suffice for explaining this data; instead I will argue for an analysis which depends on lexical conventions.
My proposals differ from previous proposals in that I do not assume that logical metonymy is a phenomenon which can be fully explained in terms of a generative process governed by general linguistic constraints. My perspective on the logical metonymy data is that if such linguistic constraints exist, they are not straightforwardly identifiable and are much more complex than a simple aspectual or semantic type restriction. My research has not led to the identification of any such constraints. Above we saw that for each of the proposed constraints on logical metonymy, there is a set of examples which violates it. Furthermore, I found that the range of logical metonymies which occurs in a corpus is severely limited, strongly suggesting that this phenomenon is heavily influenced by conventional usage. As a result, I prefer an analysis of the data which derives from the interaction of lexical information about word use with pragmatic reasoning rather than a highly productive process which is constrained by some (arbitrarily complex) set of linguistic restrictions.