STATISTICAL LANGUAGE MODEL GENERATING DEVICE AND VOICE RECOGNITION DEVICE

摘要

PROBLEM TO BE SOLVED: To efficiently generate a statistical language model by taking a domain such as a topic and a sentence pattern into consideration for adaptive data of a language model. SOLUTION: A language model generation part 20 classifies all words into part-of-speech classes of backward classes based upon part-of-speech properties of words connected before a word to be processed and part-of-speech classes of forward classes based upon part-of-speech properties of words connected after the word to be processed. Then connecting words given characteristic readings when words succeed are classified into part-of-speech class of the word at the head in the connecting words as to the backward classes and into the part-of-speech class of the word at the tail in the connecting words as to the forward classes; and then the appearance probability of multiple class N-gram having multiple part-of-speech classes is computed by connections before and after the word to be processed and its statistical language model is generated.