DICTIONARY-LANGUAGE MODEL COMPRESSION METHOD, DEVICE THEREFOR, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To provide a dictionary-language model compression method which can effectively reduce size of a dictionary by deleting words having little influence.;SOLUTION: A dictionary-language model compression method includes an entropy variation calculation step S10, a language model compression processing step S20, and a dictionary compression processing step S30. The entropy variation calculation step S10 calculates entropy variations when language entries for a language model are deleted by expanding into a 1-gram probability, and when entropy variations of words composing a dictionary in the 1-gram probability are equal to or less than a threshold, defines the corresponding language entries as deletion target language entries and corresponding words as deletion target words. The language model compression processing step S20 deletes the deletion target language entries from the language model. The dictionary compression processing step S30 deletes the deletion target words from the dictionary.;COPYRIGHT: (C)2014,JPO&INPIT