Preferences in natural language processing
Jr. James Harmon Barnett · 1993
This dissertation attempts to develop a general model of disambiguation in natural language processing. Disambiguation heuristics, or preferences, can be viewed as preorders on the set of linguistic structures created during processing. Given suitable restrictions on the monotonicity of the preferences, a best-first algorithm can be used to generate the preferred structures first. Simple preferences may be combined into complex ones or iterated to apply more than once in a single structure. The task of acquiring unrestricted preferences is intractable at present, but more restricted problems, in particular the estimation of combination functions, are manageable.