| 摘要 |
PURPOSE:To inprove the propriety for secection of candidate words as well as the fixing process speed of these words and at the same time to set a proper back track range, by taking the position of each word in a sentence into consideration when the candidate words for conversion are selected. CONSTITUTION:An extracting part 21 for the maximum liklihood candidate word consults a dictionary at and after the head of an input sentence to perform recognition of the appearing probability and sends the extracted words to a word buffer. Then the words of the word buffer are extracted one by one to give the integrated evaluation for appearing probability, and a word having the highest evaluation value is delivered. An evaluation totalizing part 22 obtains the totalized value of evaluation of each word every time a new maximum liklihood candidate word is extracted and sends it to a comparison part 24. The part 24 compares the obtained totalized value of evaluation with the prescribed threshold value. Then it is decided that the fixing conditions are satisfied when the totalized evaluation value exceeds the threshold value. Then a fixing processing part 27 fixes the candidate words.
|
