Apparatus and method for post-processing dialogue error in speech dialogue system using multilevel verification

• US 20090076798A1
• Filed: 05/27/2008
• Published: 03/19/2009
• Est. Priority Date: 09/19/2007
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Wang, Ji Hyun, Lee, Chung Hee, Oh, Hyo Jung, Lee, Chang Ki, Jang, Myung Gil, Hwang, Yi Gyu
• Granted Patent: US 8,050,909 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

Provided are an apparatus and method for post-processing a dialogue error in a speech dialogue system using multilevel verification, in which both of a user'"'"'s current utterance and a whole dialogue flow are taken into account through the multilevel verification including speech recognition results analysis, linguistic analysis, discourse analysis and dialogue analysis. As a result, various errors that may occur in the speech dialogue system are detected, and error post-processing appropriate to a detected error type is performed, so that speech recognition errors may be reduced.

Claims

1. A method of post-processing a dialogue error in a speech dialogue system using multilevel verification, comprising:

recognizing user speech to detect a speech recognition error in a sentence uttered by the user, which is obtained by the speech recognition results;

linguistically analyzing the sentence uttered by the user to detect an error in the linguistic analysis results;

analyzing a meaning of a discourse in the sentence uttered by the user based on a context to detect an error in the discourse analysis results;

analyzing a meaning of a dialogue in the sentence uttered by the user based on a whole dialogue flow to detect an error in the dialogue analysis results; and

analyzing types of the detected errors to perform error post-processing for error correction according to the corresponding error type.

2. The method of claim 1, wherein when the error type of the detected error corresponds to a speech recognition error, further comprising:

correcting an incorrect word or character sequence in the sentence uttered by the user to transfer the linguistically analyzing the sentence uttered by the user to detect an error in the linguistic analysis results.

3. The method of claim 1, wherein when the error type of the detected error corresponds to a linguistic analysis error, further comprising:

correcting the linguistic analysis error according to a noise channel calculated based on a linguistic model to transfer the analyzing a meaning of a discourse in the sentence uttered by the user based on a context to detect an error in the discourse analysis results.

4. The method of claim 1, wherein when the error type of the detected error corresponds to a discourse analysis error, further comprising:

correcting the discourse analysis error based on a context of a previous discourse to transfer the analyzing a meaning of a dialogue in the sentence uttered by the user based on a whole dialogue flow to detect an error in the dialogue analysis results.

5. The method of claim 1, wherein when the error type of the detected error corresponds to a dialogue analysis error, further comprising:

correcting the dialogue analysis error based on a whole dialogue flow to perform a system response process according to the final dialogue analysis results, in which the error is corrected,

6. The method of claim 1, wherein when no error is detected in the speech recognition results, the linguistic analysis results, the discourse analysis results, and the dialogue analysis results, further comprising:

confirming whether or not there is an exceptional condition, and performing a corresponding exceptional process when it is confirmed that there is the exceptional condition.

7. An apparatus for post-processing a dialogue error in a speech dialogue system using multilevel verification, comprising:

a speech recognition unit for recognizing user speech;

a language analyzer for linguistically analyzing a sentence uttered by the user, which is obtained from the speech recognition results;

a discourse analyzer for analyzing a meaning of a discourse in the sentence uttered by the user based on a context;

a dialogue analysis and management unit for analyzing a meaning of a dialogue in the sentence uttered by the user based on a whole dialogue flow to control a system response process according to the dialogue analysis results;

an error detection and analysis unit for detecting a speech recognition error, a linguistic analysis error, a discourse analysis error and a dialogue analysis error from the speech recognition results, the linguistic analysis results, the discourse analysis results and the dialogue analysis results, respectively, to analyze types of the detected errors;

an error post-processing unit for performing post-processing for error correction according to the error types of the detected errors; and

a system response unit for performing a system response process on the sentence uttered by the user according to the control of the dialogue analysis and management unit.

8. The apparatus of claim 7, wherein when the error type of the detected error corresponds to a speech recognition error, the error post-processing unit corrects an incorrect word or character sequence in the sentence uttered by the user so that the language analyzer additionally performs speech recognition according to the error correction results.

9. The apparatus of claim 7, wherein when the error type of the detected error corresponds to a linguistic analysis error, the error post-processing unit corrects the language analysis error according to a noise channel calculated based on a linguistic model so that the discourse analyzer additionally performs discourse analysis according to the error correction results.

10. The apparatus of claim 7, wherein when the error type of the detected error corresponds to a discourse analysis error, the error post-processing unit corrects the discourse analysis error based on a context of a previous discourse so that the dialogue analysis and management unit additionally performs meaning analysis of a dialogue in the sentence uttered by the user according to the error correction results.

11. The apparatus of claim 7, wherein when the error type of the detected error corresponds to a dialogue analysis error, the error post-processing unit corrects the dialogue analysis error based on a whole dialogue flow so that the dialogue analysis and management unit performs a system response process according to the final dialogue analysis results, in which the error is corrected.

12. The apparatus of claim 7, wherein when no error is detected from the error detection and analysis unit, the dialogue analysis and management unit confirms whether or not there is an exceptional condition, and when it is confirmed that there is an exceptional condition, the system response unit performs a corresponding exceptional process.