The Lexical, Syntactic and Semantic Processing in a Speech Recognition System

This work describes the lexical, syntactic and semantic processing of a recognition system of meaningful sentences spoken in the Italian language. A Transition Network grammar models, in an integrated way, the syntactic and semantic knowledge sources of the robot command protocol used in the experiments. The mechanisms of emission, prediction and verification of lexical hypotheses are coordinated by a parsing strategy according to the hypothesize-and-test paradigm. The most promising alternative interpretations of an input sentence are developed in parallel by the parser which has the capability of recovering errors due to misclassifications and extra or missing words.