| 摘要 |
PURPOSE:To simplify a memory of a dictionary, by uttering and inputting a Japanese sentence as it is in case of a non-inflective word, and by changing to an end form in case of an inflective word, and deciding an inflective form by precedent condition information of the following word, as for the inflective word, at the time of recognition. CONSTITUTION:From a microphone 1, a Japanese sentence is inputted as it is in case of a non-inflective word, and by an end form in case of an inflective word. For instance, ''Watakushi-wa yukanakatta'' (I did not go) is inputted as ''Watakushi/wa/yuku/ nai/ta'', is LPC-analyzed 3, a feature is extracted 4 as to each word, matching 5 is taken with a table of a word dictionary 6, that of the shortest distance is derived, a non-inflective word is inputted to a selector 9, an inflective word is stored 7 temporarily, and also the precedent condition and the inflective information of an inflective table 8, and temporary storage information are selected 9 and are read out as a result of recognition. ''Watakushi-wa'' is inputted as it is, ''yuku'' is stored 7 temporarily, and since the following ''nai'' requires its beforehand form in the inflection, as the precedent condition, ''yuka'' being the beforehand form in the inflection, of ''yuku'' is selected from the inflective table and is inputted. After that, the processing is executed in the same way. In this way, capacity of a dictionary can be simplified. |
