Evaluation of the CMU ATIS system

摘要: The CMU Phoenix system is an experiment in understanding spontaneous speech. It has been implemented for the Air Travel Information Service task. In this task, casual users are asked to obtain information from a database of air travel information. Users not given vocabulary, grammar or set sentences read. They compose queries themselves manner. This task presents speech recognizers with many new problems compared Resource Management Not only fluent, but vocabulary and open. Also, just produce transcription, action, retrieve data database. Taking such actions requires parsing "understanding" utterance. Word error rate as important utterance rate.Phoenix attempts deal phenomena that occur Unknown words, restarts, repeats, poorly formed unusual common very disruptive standard recognizers. These events lead misrecognitions which often cause total parse failure. Our strategy apply grammatical constraints at phrase level use semantic rather than lexical grammars. Semantics provide more constraint parts must ultimately be delt order take actions. Applying flexible recognizing whole while providing much word-spotting. Restarts repeats most between phase occurences, so individual phrases can still recognized correctly. Poorly constructed consists well-formed phrases, semantically well-formed. syntactically incorrect. We associate by frame-based semantics. Phrases represent word strings fill slots frames. frame able act on.The current uses bigram language model Sphinx recognition system. top-scoring string passed parser. parser assigns (word strings) input content needed frame. A beam hypotheses produced best scoring one used SQL query.