Leveraging back-off grammars for authoring context-free grammars

• US 20080133220A1
• Filed: 12/01/2006
• Published: 06/05/2008
• Est. Priority Date: 12/01/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Badger, Eric, Chickering, Max, Paek, Timothy
• Granted Patent: US 8,108,205 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars

Abstract

A system and method of refining context-free grammars (CFGs). The method includes deriving back-off grammar (BOG) rules from an initially developed CFG and utilizing the initial CFG and the derived BOG rules to recognize user utterances. Based on a response of the initial CFG and the derived BOG rules to the user utterances, at least a portion of the derived BOG rules are utilized to modify the initial CFG and thereby produce a refined CFG. The above method can carried out iterativey, with each new iteration utilizing a refined CFG from preceding iterations.

Claims

1. A method of refining context-free grammars (CFGs), comprising:

deriving back-off grammar (BOG) rules from an initial CFG; and

utilizing, based on a response of the initial CFG and the derived BOG rules to user utterances, at least a portion of the derived BOG rules to modify the initial CFG and thereby produce a refined CFG.

2. The method of claim 1 and further comprising repeating the steps of claim 1 iteratively, with each new iteration utilizing a refined CFG from preceding iterations.

3. The method of claim 1 wherein the utilizing step comprises utilizing BOG rule statistics to selectively transcribe the user utterances.

4. The method of claim 3 wherein the BOG rule statistics relate to how frequently different back-off grammar rules were utilized for recognition.

5. The method of claim 3 wherein the BOG rule statistics relate to a degree of success of different BOG rules in handling the user utterances.

6. The method of claim 1 wherein the utilizing step comprises:

selecting, based on a response of the initial CFG and the derived BOG rules to user utterances, a portion of the user utterances for transcription; and

automatically transcribing the user utterances selected for transcription.

7. The method of claim 6 wherein automatically transcribing the user utterances selected for transcription comprises utilizing a wider coverage language model relative to the CFGs to automatically transcribe user utterances selected for transcription.

8. The method of claim 1 wherein the utilizing step comprises:

selectively transcribing the user utterances; and

adjusting the initial CFG manually based on the selectively transcribed user utterances.

9. The method of claim 1 wherein the utilizing step comprises:

selectively transcribing the user utterances; and

adjusting the initial CFG automatically based on the selectively transcribed user utterances.

10. The method of claim 9 wherein adjusting the initial CFG automatically based on the selectively transcribed user utterances is carried out using minimal edit distance-based techniques.

11. An adaptive speech recognition system comprising:

a language model having a CFG and a BOG, which is derived from the CFG; and

a grammar refining component configured to iteratively modify the language model based on a response of the BOG to user utterances.

12. The system of claim 11 wherein the grammar refining component comprises:

a data gathering component configured to log data relating to a response of rules of the BOG to out-of-grammar (OOG) utterances; and

a CFG modification component configured to selective utilize the logged data to modify the CFG.

13. The system of claim 12 wherein the CFG modification component is configured to selectively transcribe the logged data using a wider coverage language model relative to the CFG.

14. The system of claim 13 wherein the CFG modification component is further configured to utilize the selectively transcribed logged data to modify the CFG using minimum edit distance-based techniques.

15. A computer-implemented speech recognition method comprising:

recognizing a user utterance using a CFG as a language model;

identifying an OOG utterance;

processing the OOG utterance through a BOG; and

assigning the OOG a back-off processing result category, the back-off processing result category being one of a plurality of different predefined back-off processing result categories.

16. The method of claim 15 and further comprising updating the CFG based on the categorized OOG utterance.

17. The method of claim 16 and further comprising repeating the steps of claim 16 iteratively until an acceptable speech recognition efficiency level is reached.

18. The method of claim 15 and further comprising saving the OOG utterance as a file copy of the user Utterance.

19. The method of claim 16 wherein updating the CFG based on the categorized OOG utterance is carried out manually.

20. The method of claim 16 wherein updating the CFG based on the categorized OOG utterance takes place automatically.