There exist data which pose a challenge to the co-composition, type-coercive, accounts outlined above. The interpretation of the logical metonymy construction seems to be severely limited in context-neutral environments (i.e. in isolation of a context which might suggest an event in which the entity denoted in the complement NP is participating), as shown by the questionable interpretations of beg5-beg8.
John began the telephone. (making?? calling with??) John began the violin. (making?? studying?) John enjoyed the bicycle. (riding?? having?? receiving the gift of??) John enjoyed the piano. (playing?? its beauty??)
On the assumption that all artifacts have eventualities associated with them through the qualia structure, the logical metonymy construction should be possible with every artifact and should have clear interpretations. The data above suggests that this is not true, indicating that the assumption that qualia structure can always supply an eventuality for the interpretation of a logical metonymy construction is too strong.
Furthermore, there are interpretations of metonymies predicted on the basis of the eventualities which seem to satisfy roles in qualia structure which are actually ungrammatical, such as those in cog1.
John began the highway (*driving on) / the dictionary (*consulting)
As a result of similar observations, it has been argued by Godard and Jayez (1993) and Pustejovsky and Bouillon (1995) that the process of type coercion must be constrained. Their proposals will be discussed below in Section 5.4.1. These approaches, and the approaches deriving from Briscoe et al (1990), seem largely to ignore the role of conventionality in interpreting logical metonymies. Analysis of a wider range of data than previously considered, derived from corpora, reveals more limited possibilities for the application of qualia structure-driven type coercion in explaining logical metonymy than predicted by the highly productive generative accounts proposed, particularly for the aspectual verbs. The existing proposed constraints on metonymy, within approaches that assume full specification of qualia structure for all nouns, don't rule out all implausible metonymies while ruling out some plausible ones. In response to this, in this chapter I will reject many of the proposed constraints on type coercion and offer an account in which logical metonymy depends on lexically-specified conventional information: i.e. properties specified in the semantic component of lexical items will determine whether and how those items can participate in metonymic constructions. In particular, I will show on the basis of the corpus data that not every noun has a telic role specified in its qualia structure, and that metonymies involving the two verbs begin and enjoy display differing behaviour with respect to the ability of context to influence their interpretation. The more general implication of this analysis is that while understanding of language may proceed in general according to certain rules or processes of interpretation, it is ultimately governed by lexical idiosyncrasies and conventionalised linguistic knowledge about individual words.
On its own, a qualia structure-driven approach to non-eventive NP type coercion is inadequate for explaining the data previously introduced in this chapter. It cannot explain variances in the acceptability of sentences and in the availability of default interpretations in context-neutral environments. Nor can it account for the ability of pragmatics to influence interpretation. We will see that it also cannot account for differences in acceptability which derive from changes in the type of the NP complement, such as those caused by pluralisation. Type coercion in this framework must therefore be constrained in some way.
On the other hand, it is clear that some aspects of logical metonymy should be conventionalised via some mechanism such as qualia structure, rather than relegated to pragmatic control. Qualia structure is a good candidate for capturing the information relevant to metonymy, but it must be utilised within a more constrained lexical system than in current approaches.
A purely pragmatic approach would fail to constrain the possible interpretations of logical metonymies, while a purely lexical approach fails to accommodate the potential contextual influence on these interpretations (as required to explain the possibility of beg28aa above being interpreted as beg28ad). A combination of the two approaches is necessary to explain the range of logical metonymy data, a point which echoes observations made by Briscoe (1990) and Lascarides & Copestake (1995). In this chapter, I will argue that there exist lexical constraints on the potential contextual influence which have not previously been formalised. I will show that information derived in the lexicon can be used to constrain the possible interpretations of a phrase in such a way that even a strong context cannot override the lexical specifications. Furthermore, I will argue that these specifications could not be relegated to the pragmatic component without a reduction in the generality of the treatment of logical metonymy phenomena. Thus, lexical specification of conventions are necessary, and the pragmatic component must be able to utilise the information coming from the lexicon in the appropriate ways.