Learning domain theories
Stephen Pulman, Maria Liakata · Amsterdam studies in the theory and history of linguistic science. Series 4, Current issues in linguistic theory · 2004
By a ‘domain theory ’ we mean a collection of facts and generalisations or rules which capture what commonly happens (or does not happen) in some domain of interest. As language users, we implicitly draw on such theories in various disambiguation tasks, such as anaphora resolution and prepositional phrase attachment, and formal encodings of domain theories can be used for this purpose in natural language processing. They may also be objects of interest in their own right, that is, as the output of a knowledge discovery process. We describe a method of automatically learning domain theories from parsed corpora of sentences from the relevant domain. 1 Domain theories The resolution of ambiguity has been a perennial topic of interest among linguists and computational linguists. In the early days of generative linguistics, the existence of various types of ambiguity was used to justify abstract levels of linguistic structure: (1) Flying planes can be dangerous. (2) He saw the man with the telescope. These examples show, respectively, that the same sequence of words can have different assignments of parts of speech (lexical categories), and that the same sequence of parts of speech can have differing constituent structures associated with them. However, after some initial flirtations with ‘semantic features’, the problem of resolving such ambiguities in a given context was not seriously pursued, since it was regarded as what we would now call ‘AI-complete’, requiring the encoding of salient features of the context as well as vast amounts of nonlinguistic knowledge: For practically any item of information about the world, the reader will find it a relatively easy matter to construct an ambiguous sentence whose resolution requires the representation of that item (Katz & Fodor, 1964:489). Early work in computational linguistics reinforced this conclusion, pointing out that it is not just lists of facts that are involved in disambiguation, but reasoning based on those facts. Furthermore, the kind of contextual reasoning required to disambiguate some sentences can draw on facts which require a subtle understanding of social or political structures, rather than facts about the physical world: