Cat egorial Disambiguat ion

This paper presents an implemented, computationally inexpensive technique for disambiguating categories (parts of speech) by exploiting constraints on possible category combinations. Early resolutions of category ambiguities provide a great deal of leverage, simplifying later resolutions of other types of lexical ambiguity. Prioritized pattern action rules (CONDS) are specified for each known categorial ambiguity. Approximately 40 such rules are currently in place. Each rule is passed lists of the words and categories preceding and succeeding a categorial ambiguity. Other categorial ambiguities are represented as lists embedded within these preceding and succeeding category lists. Unambiguous categories are represented as symbols in those lists. Additionally, each rule knows the current word and maintains a list of its possible category assignments, which had previously been extracted from a lexicon.