Conclusions

In this chapter I have discussed the need for lexical semantic representation, and the particular form which that representation should take. The conclusions to be drawn from this discussion can be summarised as follows:

• Lexical decomposition is important for capturing the interaction between syntactic structure and meaning, and for capturing generalisations over the syntactic behaviour and interpretation of related words. The identification of lexical semantic components should derive from examination of the interaction between syntactic structure and meaning as reflected in sentences exhibiting minimal variations.
• The examination of linguistic data can lead to development of a highly constrained framework
• An ontological structure is required which reflects linguistic (as opposed to world knowledge) generalisations, laying the foundation for a modular treatment of language processing in which language-specific knowledge is isolated from general reasoning.
• There must exist a logical interpretation of the representation which allows for certain inferences to be drawn on the basis of a consistent structure.

I reviewed two proposals for lexical semantic representation, Jackendoff (1983, 1990) and Davis (1995). I argued that Jackendoff's system captures the appropriate level of decomposition but that Davis' architecture captures relationships between representations in a much more natural and efficient manner, particularly in light of current developments of constraint-based grammars. I therefore introduced a representation which is based on Jackendoff's Conceptual Semantics in spirit, but on Davis' architecture in the formal details. This representation is fully compatible with the details of HPSG, overriding only the kind of relations which are assumed to reflect the semantics of words and phrases. I expanded Davis' representation on the basis of Jackendoff's proposals to allow for the representation of situation-level interpretations of full sentences, and to accommodate default information associated with individual words. The subsequent chapters of this thesis will build on this representation to address the problems of sense shifts, lexical semantic interactions, and polysemy.