A lexically-driven approach to logical metonymy allows predictions about the range of interpretations for these constructions and the defeasibility or indefeasibility of those interpretations to be captured in a general way. The definition of a pragmatic component which has access to this lexical information is critical to the modeling of the behaviour of logical metonymies in discourse contexts. I have shown that the infelicity of certain logical metonymy constructions in some discourses depends on the non-default nature of the lexicosyntactically determined interpretation for such constructions. When a non-default interpretation for a sentence cannot be coherently tied in to the discourse in which the sentence appears, discourse information cannot override that interpretation with a more coherent one, and so the sentence is judged infelicitous in that discourse -- the discourse as a whole is weak. This work emphasises the complex nature of the interaction between lexicosyntactic and pragmatic processing; discourse-level analysis is often constrained by lexical properties of the constituents of the discourse.
I have proposed a treatment of logical metonymy which depends on lexical specification of the behaviour of nouns and verbs interacting in this construction; aspectual verbs such as begin treat default information coming from the qualia of noneventive NP complements as indefeasible, other eventive verbs leave lexical defaults in their NP complements as defaults, and noneventive NPs will only have a telic event specified lexically if this event has been conventionalised. This conventionalisation has been supported by corpus data and shown to hold across different metonymic constructions, suggesting that it is reasonable to assume lexical specifications reflecting it. The basis of this conventionalisation, however, remains unclear and must be investigated in future work. More generally, an investigation of the motivations for qualia structure seems necessary at this juncture, including a theory of how qualia structure is acquired when learning a language and what dictates the inclusion of information in qualia structure in the lexicon. The current situation demands analysing corpora for specific phenomena to determine for which verbs the telic role should be specified. Clearly if the proposals here are to be generally useful in a computational framework, such that lexical structures can be built in a principled way, a motivated explanation of the conventions is required. These issues will be taken up in the next chapter.
The general importance of the discussion in this chapter is that it has provided additional evidence of the complex interactivity of syntactic, semantic, and pragmatic knowledge. Certain structures have underspecified meanings which must be resolved through a combination of lexically specified facts, compositionality, and general reasoning processes. None of these knowledge sources can be ignored in modeling the interpretation process.