| 摘要 |
PROBLEM TO BE SOLVED: To improve the accuracy of unknown word information without complicating a processing by calculating likelihood based on appearing frequency information on the respective character positions for respective generated candidates of the unknown word information and outputting a part of the candidates of the unknown word information together with word information based on the calculated likelihood. SOLUTION: A retrieved result synthesis part 303 gathers retrieved results of a word storage part 305 and the candidate of a name-unknown word prepared by a name candidate preparation part 304. The name candidate preparation part 304 prepares the candidates of the name-unknown word. In the work storage part 305, words and single KANJI (Chinese character) are stored and the frequency of the case of being used as a name is added. In a reading storage part 306, the reading of a name is stored. In a character string preparation rule storage part 307, a rule for segmenting the name-unknown word is stored and a character string is prepared based on the rule. Thus, since the likelihood of respective reading candidates is calculated and a higher order reading candidate is outputted, the same effect as the case of additionally registering the name-unknown word is obtained. |
