LANGUAGE MODEL LEARNING DEVICE, LANGUAGE MODEL LEARNING METHOD, LANGUAGE ANALYSIS DEVICE, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To divide a new character string or symbol string which does not comply with a previously given standard while maintaining the standard.

SOLUTION: An identification model parameter update part 25 converts an NPYLM into a CRF, generates a first integrated model using weights of respective edges of the converted CRF and weights of corresponding edges of the CRF, and learns the first integrated model based upon supervised data. A generated model parameter update part 26 converts the CRF into a Semi-Markov CRF, generates a second integrated model using weights of respective edges of the Semi-Markov CRF and weights of corresponding edges of the NPYLM, and learns the second integrated model based upon unsupervised data. The update by the identification model parameter update part 25 and the update of the generated model parameter update part 26 are alternately repeated until a convergence determination part 27 determines that a specific convergence condition is met.

COPYRIGHT: (C)2012,JPO&INPIT