PERSONALIZING A CONTEXT-FREE GRAMMAR USING A DICTATION LANGUAGE MODEL

• US 20070239454A1
• Filed: 04/06/2006
• Published: 10/11/2007
• Est. Priority Date: 04/06/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Wu, Qiang, Badger, Eric, Paek, Timothy, Chickering, David
• Granted Patent: US 7,689,420 B2
• Time in Patent Office: Days
• US Class Current: 704/257
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

Architecture for integrating and generating back-off grammars (BOG) in a speech recognition application for recognizing out-of-grammar (OOG) utterances and updating the context-free grammars (CFG) with the results. A parsing component identifies keywords and/or slots from user utterances and a grammar generation component adds filler tags before and/or after the keywords and slots to create new grammar rules. The BOG can be generated from these new grammar rules and can be used to process the OOG user utterances. By processing the OOG user utterances through the BOG, the architecture can recognize and perform the intended task on behalf of the user.

Claims

1. A system for generating a dictation language model in speech recognition applications, comprising:

a detection component that identifies out-of-grammar (OOG) utterances, the OOG utterances are not recognized by context-free grammar (CFG) rules;

a grammar generation component that generates a dictation language model by creating new rules based in part on the OOG utterances; and

a personalization component that updates old CFG rules with the new rules.

2. The system of claim 1, wherein the new rules are based in part on the OOG utterances and the old CFG rules.

3. The system of claim 2, wherein only a portion of the CFG rules are re-written to generate the new rules.

4. The system of claim 1, wherein the grammar generation component adds filler tags proximate at least one of keywords and slots to generate the new rules.

5. The system of claim 4, wherein the filler tags are dictation tags.

6. The system of claim 4, wherein the filler tags are based on phonetic similarity to keywords.

7. The system of claim 1, further comprising a processing component that analyzes user utterances using the dictation language model after the CFG has failed to recognize the user utterances.

8. The system of claim 1, further comprising a processing component that processes user utterances through the dictation language model before the CFG processes the utterances.

9. A computer-implemented method for using a dictation language model to personalize a context-free grammar (CFG), comprising:

generating a dictation language model based in part on out-of-grammar (OOG) utterances;

identifying frequently used OOG utterances;

determining if the frequently used OOG utterances should be incorporated into the CFG;

updating the CFG with the frequently used OOG utterances; and

eliminating utterances not frequently employed by a user from the CFG.

10. The method of claim 9, further comprising verifying if the OOG utterances contain a keyword or slot that distinctly identifies an intended rule.

11. The method of claim 9, further comprising employing dictation tags to determine which rule is most likely an intended rule.

12. The method of claim 9, further comprising deriving a measure of phonetic similarity between the OOG utterance and keywords.

13. The method of claim 9, further comprising automatically updating the CFG with phrases based in part on the OOG utterances.

14. The method of claim 9, further comprising requesting permission to update the CFG with phrases based in part on the OOG utterances.

15. The method of claim 9, further comprising utilizing a predictive user model to determine the most likely command result based on the OOG utterances.

16. The method of claim 15, further comprising engaging in a confirmation of a predicted command result with the user.

17. The method of claim 16, wherein if the confirmation is correct, then automatically adding the command result to the CFG.

18. The method of claim 9, further comprising augmenting the CFG with utterances that users do utilize.

19. A computer-implemented system for generating a dictation language model in command-and-control speech recognition applications, comprising:

computer-implemented means for identifying out-of-grammar (OOG) utterances that are not recognized by context-free grammar (CFG) rules;

computer-implemented means for generating new rules based in part on the OOG utterances;

computer-implemented means for updating the CFG rules with the new rules; and

computer-implemented means for eliminating old rules from the CFG not frequently used by the user.

20. The system of claim 19, further comprising computer-implemented means for processing user utterances through the dictation language model at least one of before and after the CFG processes the utterances.