A Tabular Method of Finding the Optimal Word String together with its Dependency Structure

This paper describes a tabular method of nding the word string, together with its dependency structure, that has the maximum preference score in a given series of word sets. The technique, which originated in the problem of constraining the process of Japanese speech recognition by linguistic knowledge, is potentially applicable to parsing and disambiguation of various natural languages. The preference score of a word string is de ned by taking syntax, semantics, and reliability of each word into consideration. A set of recurrence equations is derived based on the principle of Dynamic Programming, which leads to an algorithm to solve the combinatorial optimization problem. In the algorithm, searching for the optimal word string proceeds concurrently with parsing of the string, looking up and lling in two triangular tables.