System and Method for Generating Data for Complex Statistical Modeling for use in Dialog Systems

• US 20100204982A1
• Filed: 02/06/2009
• Published: 08/12/2010
• Est. Priority Date: 02/06/2009
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Weng, Fuliang, Feng, Zhe, Li, Katrina
• Granted Patent: US 8,401,855 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

Embodiments of a dialog system that utilizes grammar-based labeling scheme to generate labeled sentences for use in training statistical models. During the process of training data development, a grammar is constructed manually based on the application domain or adapted from a general grammar rule. An annotation schema is created accordingly based on the application requirements, such as syntactic and semantic information. Such information is then included in the grammar specification. After the labeled grammar is constructed, a generation algorithm is then used to generate sentences for training various statistical models.

Claims

1. A computer-implemented method in a dialog system, comprising:

defining a set of one or more grammar rules;

labeling each grammar rule of the set of grammar rules with semantic or syntactic characteristics to produce labeled grammar rules; and

generating labeled sentences from the set of labeled grammar rules.

2. The method of claim 1 wherein each grammar rule of the set of grammar rules comprises a context-free grammar rule.

3. The method of claim 2 wherein each context-free grammar rule is a 4-tuple that consists of a start point, a non-terminal element, a terminal element, and a rule.

4. The method of claim 1 wherein the labeled sentences each comprise a plurality of words, wherein each word of the plurality of words is annotated with one or more characteristics.

5. The method of claim 4 wherein each characteristic of the one or more characteristic represents at least one of the semantic or syntactic characteristics.

6. The method of claim 1 further comprising using the labeled sentences to train one or more statistical models of the dialog system.

7. The method of claim 6 wherein one or more models are used by a speech recognition stage of the dialog system.

8. The method of claim 6 wherein the one or more models are utilized by a spoken language understanding unit of the dialog system.

9. The method of claim 6 wherein the one or more models are utilized by a dialog manager unit of the dialog system.

10. A dialog system, comprising:

a speech recognition unit receiving a user utterance including one or more sentences;

a spoken language understanding unit receiving the user utterance, and utilizing one or more models to derive a hypothesis for each word of the one or more sentences;

a labeled sentence generator module providing labeled sentences to train the one or more models, the labeled sentence generator labeling each grammar rule of a set of grammar rules with semantic or syntactic characteristics to produce labeled grammar rules and generating labeled sentences from the set of labeled grammar rules.

11. The system of claim 10 wherein each grammar rule of the set of grammar rules comprises a context-free grammar rule.

12. The system of claim 11 wherein each context-free grammar rule is 4-tuple that consists of a start point, a non-terminal element, a terminal element, and a rule.

13. The system of claim 10 wherein the input source comprises a user utterance input to a speech recognition stage of the dialog system.

14. The system of claim 13 further comprising using the labeled sentences to train one or more statistical models of the dialog system.

15. The method of claim 14 wherein the one or more models are utilized by a spoken language understanding unit of the dialog system.

16. The method of claim 15 wherein the one or more models are utilized by a dialog manager unit of the dialog system.

17. A computer readable medium containing a plurality of program instructions, which when executed by a processor, cause the processor to perform the steps of:

processing the rules of a defined set of grammar rules;

labeling each grammar rule of the set of grammar rules with semantic or syntactic characteristics to produce labeled grammar rules; and

generating labeled sentences from the set of labeled grammar rules.

18. The medium of claim 17 wherein each grammar rule of the set of grammar rules comprises a context-free grammar rule.

19. The medium of claim 18 wherein each context-free grammar rule is a 4-tuple that consists of a start point, a non-terminal element, a terminal element, and a rule.

20. The medium of claim 19 wherein the labeled sentences each comprise a plurality of words, wherein each word of the plurality of words is annotated with one or more characteristics.

21. The medium of claim 20 wherein each characteristic of the one or more characteristic represents at least one of the semantic or syntactic characteristics.