As discussed in the introduction to this thesis, the pragmatic
reasoning framework and discourse model to which I appeal in this
thesis is a highly intuitive one. I have not attempted to propose a
concrete model of contextually-based interpretation, but have instead
concentrated on what kind of lexical semantic structure and linguistic
processing might underpin such interpretation, and what kind of
representations might interact with pragmatic reasoning. The
development of a full-scale model of language understanding, will
however, require a formal model of conceptual structure, world
knowledge, and how this knowledge interacts with the result of
linguistic processing and therefore lexical knowledge. This is a
daunting task, given the huge variations in experience (linguistic and
otherwise) among speakers, but there do seem to be certain conceptual
relations which are primary and consistent (at the word level, these
might include synonymy, antomyny, hyponymy, and semantic similarity;
at the discourse level, coherence constraints should be represented;
world knowledge might include script-like descriptions of common
situations). Interpretation can clearly be influenced by all such
knowledge, even at the level of individual sentences, but formal
models of precisely how this knowledge is structured are the subject
of on-going research (e.g. the WordNet project (Miller
miller:90) and Cycorp's attempts at developing an
ontology of world knowledge
). Further research is also
necessary into precisely how world knowledge and discourse processing
interacts with the results of linguistic processing.
In this work, I have not fully explored the exploitation of the entailments associated with the attributes in the lexical semantic structures for complex semantic reasoning, although this is clearly an important function served by a consistent representation. These entailments would form part of the pragmatic reasoning system, and could provide a basis for establishing the inference mechanisms which interact with world knowledge.
I also have not fully explored the role of selectional restrictions. There are clearly certain combinations of words which are semantically anomolous due to conflicts in the expectations of semantic type which a verb places on its arguments, or more generally due to our expectations of how certain entities can behave (e.g.\ ideas are not alive and so we do not expect them to be able to sleep: #The ideas are sleeping). But many of these apparent restrictions can be violated in certain contexts, like metaphors, poetry or animated films (where, for example, goats might read and the interpretation of The goat began the book would be entirely felicitous). Investigating the nature of these restrictions and the constraints under which they can be violated will be important for complete models of language understanding, and will again involve examining the interaction between lexical semantic constraints and reasoning at the pragmatic level.
I have considered a few constructions in this thesis whose interpretation depended on lexical semantic structure and the linguistic and discourse context. There are clearly many more which need to be investigated, and one of the most active areas of lexical semantic research is in the analysis of verbal alternations. Many verbs have alternate syntactic forms (in addition to the dative and causative alternations, there are alternations such as the spray/load alternation: John loaded books onto the truck and John loaded the truck with books), and corresponding slightly varying interpretations or usage constraints. I have not considered exactly how the lexical semantic representations I have utilised in this thesis might be applied to the problem of efficient representation of these alternations, other than for the dative alternation, or the constraints in place which prevent certain verbs from alternating. Given the frequency of these alternations and the amount of discussion of them existing in the literature, this would be a natural next step to take with this framework in place.