摘要 |
PROBLEM TO BE SOLVED: To efficiently conduct conversation voice recognition with a high degree of freedom while accepting freer conversation than the conversation based on a fixed and formalized grammar. SOLUTION: A key phrase detector 11 detects plural key phrases based on the set of the phrase part grammar of the intrinsic condition of the conversation. Then, a key phrase verifier 12 assigns reliability scales to these key phrases, the scales are compared with threshold values to verify the key phrases and a set of verified key phrase candidates is obtained. Then, a sentence hypothesis generator 13 couples the candidates based on a prescribed (task intrinsic, for example) meaning information 25 and a sentence hypothesis is obtained. Lastly, a sentence hypothesis verifier 14 verifies the sentence hypothesis, verified sentence hypothesis is generated and a final recognition result is obtained. In the conversation base system, individual phrase part grammar is used in accordance with the specific state of the conversation in a large scale task (i.e., against 'subtasks'). |