摘要 |
PURPOSE:To select a correct tuple with high probability by using a means which measures a physical parameter of the input information for visual recognition to estimate the probability of recognition and using the attribute permutation optimized from the result of said estimation to carry out the retrieval collation. CONSTITUTION:A physical parameter is extracted out of a voice input 'on 20th, at S hall, with Mr. M' and the reliability of words is judged from the difference between said parameter and the reference value. Thus a permutation of attribute is decided in the order of higher reliabilities. If the input of a 1st attribute data is wrong, the words of correct answers are excluded and no correct tuple remains on a data base. For the attribute permutations decided by the reliabilities, a 1st attribute permutation is decided as 'singer concert place date' together with a 2nd attribute permutation defined as 'concert place singer date' respectively. Then the input of a singer 'Mr. M' and a concert place 'S hall' are applied in the order of higher reliabilities together with an attribute date of low reliability supplied last. Thus an accurate word '20th' excluded due to the shortage of input is satisfactorily recognized. |