Finite State Grammar inference for connected word recognition

Abstract

An algorithm that infers a finite state grammar from a language specified by simple erumeration of legal phrases , is introduced. The algorithm processes every phrase as a string of phonetic units (words, syllables, demisyllables, etc.), associates one or more states to each phonetic unit and establishes the set of predecessors (and the set of successors) for every state. As a fundamental feature , the algorithm includes an optimization step in order to minimize the number of states. Additionally, the algorittan candeal with languages specified by a combination of other languages or, even, by a context-free grammar.