摘要 |
<P>PROBLEM TO BE SOLVED: To easily register readings of a new coinage and a slang word in an analytic dictionary for speech recognition without any human operation. <P>SOLUTION: A text analysis section 12 analyzes character information that a text input section 11 inputs from the Internet etc., by referring to a current analytic dictionary 15 and extracts an unregistered word from an analyzed document having been shaped by a rule shaping part 13. A speech recognition section 17 recognizes a speech of speech information that a speech input part 16 inputs from a radio etc., to obtain a recognized document. A collation part 19 compares the recognized document with the analyzed document stored in a database for dictionary registration and when retrieval character strings before and after the unregistered word in the analyzed document are present in the recognized document, registers a character string sandwiched between the retrieval character strings in the recognized document to the analytic dictionary as a reading of an unregistered word. <P>COPYRIGHT: (C)2005,JPO&NCIPI |