Computerizing a machine readable dictionary

Abstract

Current research in natural language processing is characterized by the development of theories of grammar which strongly depend on the lexicon to drive parsing systems (e.g. Lexical Function Grammar, General Phrase Structured Grammar, Functional Unification Grammar). These requirements go far beyond the typical small, hand-coded vocabularies developed for theoretical or demonstration purposes. Many researchers have independently discovered the rich, though unstructured, knowledge sources that machine readable dictionaries offer. This paper reports on an attempt to impose structure to the Funk and Wagnalls Dictionary, by means of a parser written in Turbo Pascal, using a mixed approach of pattern matching and transition networks. The resulting computerized dictionary is 95 % accurate, but correcting the final 5% incorrectly parsed involves painstakingly scrutinizing the output and modifying the parser to handle exceptional cases that occur only once or twice in the entire MRD, or editing the machine readable dictionary to remove errors introduced by the OCR process.