It is clear from the previous discussion that context can sometimes influence the interpretation of metonymic constructions. Although this does not occur frequently for begin and finish, it does for begin on and also enjoy. The interaction between logical metonymy and pragmatics must therefore be considered.
Lascarides & Copestake (1995) (L&C) extend the lexical approaches to logical metonymy to develop a system which takes into account the influence of context on the reconstruction of an event in a coercive environment. They utilise the idea that lexical defaults -- defaults specified in qualia structure -- persist as default beyond the lexicon into the pragmatic component, and are therefore used in pragmatic reasoning,
L&C formalise their approach in a unification-based framework, with a theory of lexical structure in which the lexical entries are built up from TDFSs so that non-default and default information is demarcated, and which utilises persistent default unification (PDU) (Lascarides et al lascarides_etal:96) in order to allow defaults to persist into logical form. The lexicon is hierarchically ordered in the manner adopted in Chapter 2 (as described in e.g. Copestake (1992)). They adopt Pustejovsky's notion of qualia structure, but augment it with the notion of defeasibility. So the values in qualia structure provide lexical information which is treated as default. For example, the telic role of book is read by default. Other aspects of lexical representation follow the lexical representation language (LRL) (Copestake copestake:93b).
In the pragmatic component DICE (Lascarides and Asher 1991), L&C propose two axioms, i) Defaults Survive: lexical generalisations normally apply in a discourse context and ii) Discourse Wins: conflicting discourse information wins over lexical defaults. The lexicon links to a pragmatic component via these axioms to interact with discourse context. The axioms together can be used to explain when (5.61a) entails (5.61b) rather than (5.61c).
My goat eats anything. He really enjoyed your book. The goat enjoyed eating your book. The goat enjoyed reading your book.
Since L&C build on the Pustejovsky approach to logical metonymy, they adopt the assumption of full representation of qualia structure and rely on constraints applied prior to pragmatic processing to rule out implausible metonymies. The approach therefore suffers from the same over- and under-generation as purely lexical approaches under this assumption (see Section 5.4.2). However, it is fully compatible with the view on what information is lexically specified as presented previously in this chapter. I will outline how integration of the underspecified qualia structure with the L&C model of pragmatic reasoning with defaults can account for constraints on the interaction of logical metonymy with discourse context in the discussion which follows.
It is sometimes assumed that with a sufficiently strong context any interpretation of a semantically underspecified sentence can be coerced, as is suggested by the example beg29. The L&C work, for example, starts from the assumption that contextual information can always override lexical defaults in a logical metonymy. Even when context is added to aid in the interpretation of metonymic constructions, however, there are cases in which the constructions seem impossible despite the availability of an eventive interpretation of the noun phrase from context, as suggested Section 5.2. The example beg30 presented there is repeated here as beg95. beg95b is infelicitous as a follow-up to beg95a, despite the strong pragmatic clues to interpret begin as begin destroying. A construction like beg95c, however, is grammatical and begin on is interpreted as begin destroying. A similar pattern holds in cog3.
John will be audited by the tax service, so he has been destroying things which might incriminate him. He has destroyed the files and the computer disks. *He will begin the books tomorrow. (destroying) He will begin on the books tomorrow. (destroying)
My goat went nuts last night. He ate everything in his cage.
*He began your book at 9pm. He began on your book at 9pm. He particularly enjoyed your book.
The discourse in cog3 exemplifies a contrast between the behaviour of begin (or any aspectual verb which may be substituted) and other eventive verbs. The sentence cog3c(i) is infelicitous as a continuation of the discourse cog3a,b, while the sentences cog3c(ii-iii) are not. This parallels the distinction in beg95. Assuming that default interpretations for begin (on) your book and enjoy your book are predicted from the lexicon, the examples suggest that the default interpretation of begin+NP cannot be overridden by contextual cues for its interpretation, while the default interpretations of begin on+NP and enjoy+NP can be. In formal terms, begin can be considered to force the conversion of lexical defaults into indefeasible information at the interface with logical form, which cannot be overridden regardless of strong contextual clues against the default interpretation. In this situation, a conflict in pragmatics between the interpretation proposed on the basis of lexical information and that suggested by context can only serve to judge the sentence as infelicitous, because it won't connect coherently to the discourse content. I will show how the interaction of lexical information with the pragmatic component can be used to account for incoherent discourses such as that in beg95a,b below.
An additional description must therefore be added to the four provided at the start of Section 5.5.3:
Figure 5.6: Modified coercing forms of begin: One entry picks out agentive
role, the other telic role
So I must also somehow ensure that defaults specified in the qualia structure of a begin complement become indefeasible information in the syntagmatic representation of the sentence. To achieve this, I will modify slightly the lexical entries for the coercing forms of begin presented in Figure 5.3, by forcing unification between the event argument position in the logical form of begin and the default event represented in the telic or agentive role of the complement noun's qualia structure. The modified lexical entries are shown in Figure 5.6. They differ from the original lexical entries in that only the defeasible portion of the TDFS in the complement's qualia structure is carried into the logical form for the sentence as a whole, which means that the event argument of begin will no longer reflect any defaults -- it will simply be the non-default feature structure associated with the defeasible portion of the source TDFS. In the original lexical entries, the full TDFS in the complement's qualia structure was carried into the logical form, so both the default interpretation and its defeasible nature was explicitly represented. For the revised lexical entries the default interpretation is represented, but it is no longer treated as defeasible and so cannot be overridden through pragmatic reasoning. The pragmatic reasoning process from which the infelicity of sentence beg95b in context is concluded, on the basis of the lexical entries proposed here, is outlined below.