Conversation System and Conversation Software

• US 20080319748A1
• Filed: 01/31/2007
• Published: 12/25/2008
• Est. Priority Date: 01/31/2006
• Status: Active Grant

• Current Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Original Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Inventors: Tsujino, Hiroshi, Okuno, Hiroshi, Nakano, Mikio, Takeuchi, Yohane, Komatani, Kazunori
• Granted Patent: US 8,996,385 B2
• Time in Patent Office: Days
• US Class Current: 704/251
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A first domain satisfying a first condition concerning a current utterance understanding result and a second domain satisfying a second condition concerning a selection history are specified. For each of the first and second domains, indices representing reliability in consideration of the utterance understanding history, selection history, and utterance generation history are evaluated. Based on the evaluation results, one of the first, second, and third domains is selected as a current domain according to a selection rule.

Claims

1. A conversation system comprising a first processing unit which understands an utterance input to a speech input device, a second processing unit which selects one of a plurality of domains based on the utterance understanding result by the first processing unit, and a third processing unit which generates an utterance to be output from a speech output device based on the domain selected by the second processing unit, wherein

the second processing unit specifies a first domain satisfying a first condition concerning a current utterance understanding result by the first processing unit and a second domain satisfying a second condition concerning a selection history by the second processing unit from among the plurality of domains, evaluates a part or all of first, second, and third indices for each of the first and second domains, the first index representing reliability in consideration of an utterance understanding history by the first processing unit, the second index representing reliability in consideration of the selection history by the second processing unit, and the third index representing reliability in consideration of an utterance generation history by the third processing unit, and the second processing unit selects one of the first domain, the second domain, and a third domain different from the first and second domains, as a current domain, based on the evaluation results and according to a selection rule.

2. The conversation system according to claim 1, wherein the second processing unit evaluates likelihood of each domain based on the utterance understanding result by the first processing unit, and specifies the domain satisfying the first condition that the likelihood of the domain is ranked within n₁ (n₁=1, 2, . . . ), as the first domain.

3. The conversation system according to claim 1, wherein in selecting a domain at the k+1-th time (k=1, 2, . . . ), the second processing unit specifies the domain satisfying the second condition that the domain was selected as the current domain up to i times before (i=1, 2, . . . ), as the second domain.

4. The conversation system according to claim 1, wherein in selecting a domain at the k+1-th time (k=1, 2, . . . ), the second processing unit specifies the domain satisfying the second condition that the accumulated number of times that the domain was selected as the current domain is ranked within n₂ (n₂=1, 2, . . . ) among the domains that were selected as the current domain up to i times before, as the second domain.

5. The conversation system according to claim 1, wherein the second processing unit evaluates the first index for each of the first and second domains, based on at least one of:

an index representing presence/absence of a positive response;

an index representing presence/absence of a negative response;

the number of times of slot updating;

the accumulated number of positive responses;

the accumulated number of negative responses;

the accumulated number of times of slot updating;

the number of times of utterance understanding;

an acoustic score of the utterance understanding result;

posterior probability of the utterance understanding result; and

an arithmetic mean of language component reliability in the utterance understanding result.

6. The conversation system according to claim 1, wherein the second processing unit evaluates the second index for each of the first and second domains, based on at least one of an index representing presence/absence of selection and the number of times of selection.

7. The conversation system according to claim 1, wherein the second processing unit evaluates the third index for each of the first and second domains, based on at least one of an index representing presence/absence of task accomplishment and the accumulated number of times of task accomplishment.

8. The conversation system according to claim 1, wherein the second processing unit selects a current domain according to a decision tree serving as the selection rule.

9. A conversation program causing a computer to function as a conversation system including a first processing unit which understands an utterance input to a speech input device, a second processing unit which selects one of a plurality of domains based on the utterance understanding result by the first processing unit, and a third processing unit which generates an utterance to be output from a speech output device based on the domain selected by the second processing unit, wherein the program causes the computer to function such that the second processing unit specifies a first domain satisfying a first condition concerning a current utterance understanding result by the first processing unit and a second domain satisfying a second condition concerning a selection history by the second processing unit from among the plurality of domains, evaluates a part or all of first, second, and third indices for each of the first and second domains, the first index representing reliability in consideration of an utterance understanding history by the first processing unit, the second index representing reliability in consideration of the selection history by the second processing unit, and the third index representing reliability in consideration of an utterance generation history by the third processing unit, and the second processing unit selects one of the first domain, the second domain, and a third domain different from the first and second domains, as a current domain, based on the evaluation results and according to a selection rule.