next up previous contents
Next: Word knowledge represented in Up: Logical Metonymy Previous: Analysis of the meaning

The range of the data

  I have introduced various grammatical and ungrammatical examples of logical metonymy above. A closer look at the details of the proposed co-composition analyses of logical metonymy will reveal more data which must be taken into consideration.

Much discussion in recent lexical semantic literature has focused on logical metonymy (Pustejovsky 1991, pustejovsky:95a), linguistic constraints on the phenomenon (Godard & Jayez 1993, Pustejovsky & Bouillon 1995), and the influence of discourse context on the interpretation of these metonymies (Briscoe 1990, Lascarides & Copestake 1995). All of this work uses the co-composition analysis as its starting point. The systematic syntactic ambiguity of the eventive verbs is handled on one version by postulating a single logical form for the verb and triggering the operation of type coercion to shift the syntactic type of the complement to a type which is compatible with the verb (Pustejovsky pustejovsky:95a, Pustejovsky & Bouillon pust_bouillon:95). This process is needed because aspectual verbs require eventive complements. Noun phrase complements which refer to objects (like book in the above examples) cannot be construed directly as events and must be converted somehow into an event-referring phrase in order to be compatible with the aspectual verbs. Another approach is to postulate multiple logical forms for the eventive verbs (Godard & Jayez 1993, Copestake & Briscoe 1995), each of which is responsible for appropriately combining with a distinct type of complement. In the latter case, type coercion is internal to the verb semantics, in that the lexical entry for a verb specifies precisely how the interpretation of a full VP is to be established on the basis of the particular complement type specified in the lexical entry. The result under both of these approaches is that the logical form for each combination of the verb plus complements will be identical.

Within this work, there are differences in the status of lexical information. Analyses following Pustejovsky pustejovsky:95a do not address the issue of contextual influence on logical metonymy interpretation and therefore cannot account for the acceptability of the interpretation of beg28aa as beg28d. The lexicon is the sole source of information for the interpretation of logical metonymies under these accounts. Briscoe (1990) and subsequent work (e.g. Lascarides & Copestake lasc_copestake:95) argue for the influence of pragmatics on logical metonymy interpretation and treat lexical information as default information about the interpretation. A lexical default is lexical information which appears as default in the logical form corresponding to the semantics of a sentence or phrase (this notion is formalised in Lascarides 1996). This distinction in the status of lexical information is data-driven, deriving from examples such as the ``goat'' example. Under this perspective, the lexicon can suggest an interpretation, but that interpretation can be overridden by reasoning derived from the discourse context. These approaches formally address the interaction between lexical semantics and discourse context, but do not account for cases in which context cannot override defaults (e.g. beg30) or the full range of constraints on logical metonymy.

The coercion which must occur to get the appropriate readings of the logical metonymies clearly requires more than a simple conversion of the type of the NP argument. The missing element of meaning, the event which serves as an argument to the eventive verb, must be introduced. In the approaches under consideration, this eventive element is suggested by lexical information associated with the object denoted by the complement noun phrase of the eventive verb. Pustejovsky (1991) proposes that the element comes from one of the roles in the noun's lexical semantic structure, the qualia structure, which represents the defining attributes of an object. Type coercion looks (in the first instance, for the accounts which also take into account pragmatic influence) to the qualia structure for something of the type required by the verb.

The existing co-composition analyses of logical metonymy assume that a full qualia structure is represented in each noun's lexical semantics, that is a completely specified representation of the core components of the noun's meaning. These components serve as a source for eventualities involving the noun which can be accessed in metonymic interpretation.




next up previous contents
Next: Word knowledge represented in Up: Logical Metonymy Previous: Analysis of the meaning