| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a lexical acquisition apparatus for learning a new name in interaction. <P>SOLUTION: The lexical acquisition apparatus includes a phoneme recognition section 2 for creating a phoneme string candidate from an input voice, a word matching section 3 for creating a plurality of word sequences based on the phoneme sequences candidate, a discrimination section 4 for selecting a word sequences of high likelihood as a recognition result from the plurality of word sequences, an acquisition section 5 for acquiring a new word based on the word sequences selected by the discrimination section 4, a list 4A of teaching words used when a name is taught, and a probability model 4B of the teaching words and unknown words. The discrimination section 4 calculates a first evaluation value indicating a matching degree of the words in each word sequences with the teaching word in the list 4A, and a second evaluation value indicating the connection probability between the words in each word sequences, and selects the word sequences where the sum of the first evaluation value and the second evaluation value is maximum. The acquisition section 5 obtains, as a new word, a word that is not related to the calculation of the first evaluation value, of the word sequences selected by the discrimination section 4. <P>COPYRIGHT: (C)2011,JPO&INPIT</p> |
