VOICE RECOGNITION ERROR CORRECTING DEVICE

摘要

PROBLEM TO BE SOLVED: To specify correction places and to perform conversion processings for correcting errors. SOLUTION: A voice recognition error correcting device 10 extracts a correction place which includes a modification larger than a threshold value by generating the dependency structure of a parse tree format while performing a structural analysis with respect to the input sentence of a voice recognition result and by calculating the semantic distance between an example word to the midification and an input word with respect to respective modifications. Then, morpheme strings of correction cadidates are obtained by retrieving example sentences including character strings whose editing distances are close to character strings of the correction place from example sentences and by extracting morpheme partial strings corresponding to the correction place from morpheme strings of the retrieved example sentences and by correcting the morpheme string of the input sentence while replacing the morpheme partial strings with the corresponding morpheme partial strings. Moreover, correction cadidates whose phonemic distances are small are made to remain by converting morpheme strings of recognition results before and after the correction into phoneme strings while using a conversion dictionary and by calculating phonemic distances of phoneme strings of the correction candidates with respect to the phoneme strings of the recognition result before the correction and a correction candidate whose recalculated semantic distance is equal to or smaller than a threshold value is made to be an error correction result.