SEMANTIC CONSISTENCY IN TEXT--AN EXPERIMENT AND SOME SUGGESTIONS,
Abstract
The author attacks the problem of the resolution of word-sense ambiguity in texts by a method that can be extended so as to allow a program to construct a word-sense that was not present in the dictionary at the beginning. The resolution proceeds by means of rules of sequence operating on coded message forms or templates. These rules can be thought of as an explication of 'Semantic consistency' between the fragments of natural language that match with the templates. The rules of sequence select those templates that satisfy the rules of sequence, and thus select among the coded word-senses that the templates express. (The templates were previously assigned to the natural language fragments by a matching routine operating on a semantic coding of the senses of each word in the dictionary.) The experiment reported is a small computation over a set of questions and answers to determine which sequences of question-answer were well formed on the basis of the rules of sequence; a proper sequence being intended to define an answer rather than a right answer. It is hoped to extend this method to a chain of such consistencies, or derivation, running through longer texts so that, in certain cases where the resolution rules failed, a possible sense for a recalcitrant word could be constructed. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 20, 1967
- Accession Number
- AD0648770
Entities
People
- Yorick Wilkes
Organizations
- System Development Corporation