A study of the capability of grammatical analysis to improve accuracy in continuous speech recognition for command and control

The program is directed toward the development of a theory of speech understanding systems which explicates the role to be played by linguistic constraints (syntax and semantics) in correcting the errors resulting from word recognition at the acoustic-phonetic-lexical level alone. The general model of a speech understanding system which provides the framework for this study is that which views speech understanding as basically an acoustic-phonetic word recognition process, whose decisions are questioned, and sometimes overruled, by criteria at the syntactic and semantic levels. The principal components of this model are (1) an acoustic to phonetic transformation, (2) a lexical process (the lexical process along with the acoustic to phonetic transformation constitute the basic word recognition system whose output is judged by higher level linguistic constraints), (3) a model of standard English syntax, (4) a model of standard English semantics, and (5) an integrating process which determines the strategies to be used in backtracking over previous decisions. (Modified author abstract)