Method for preserving contextual accuracy in an extendible speech recognition language model

• US 20020116194A1
• Filed: 02/21/2001
• Published: 08/22/2002
• Est. Priority Date: 02/21/2001
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Lewis, James R., Ortega, Kerry A., Smith, Maria E., Rutherfoord, C. Thomas
• Granted Patent: US 7,103,533 B2
• Time in Patent Office: Days
• US Class Current: 704/257
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

A method of generating language model statistics for a new word added to a language model incorporating at least one class file containing contextually related words. The method can include the following steps: First, language model statistics can be computed based on references to at least one incorporated class file. Second, a new word can be substituted for each reference to a selected class file. Additionally, the language model statistics can be re-computed based on the new word having been substituted for the reference. Third, the re-computed language model statistics can be displayed in a user interface and modifications can be accepted to the re-computed language model statistics through the user interface. Fourth, the language model statistics can be further re-computed based on the modifications. In consequence, the language model statistics are re-computed for the new word without introducing contextual inaccuracies in the language model.

Claims

1. A method of generating language model statistics for a new word added to a language model incorporating at least one class file containing contextually related words, the steps of the method comprising:

(a) computing language model statistics based on references to at least one incorporated class file;

(b) substituting a new word for each reference to a selected one of said at least one incorporated class file, and re-computing said language model statistics based on said new word having been substituted for said reference;

(c) displaying said re-computed language model statistics in a user interface and accepting modifications to said re-computed language model statistics through said user interface; and

,

(d) further recomputing said language model statistics based on said modifications,

whereby said language model statistics are re-computed for said new word without introducing contextual inaccuracies in the language model.

2. The method of claim 1, further comprising:

(e) if no modifications are accepted in step (c), adding said new word to said selected one of said at least one incorporated class file, said new word inheriting existing context-based language model statistics associated with said selected one of said at least one incorporated class file.

3. The method of claim 2, further comprising repeating steps (a)-(e) for each class file associated with said new word.

4. The method of claim 1, wherein said displaying step comprises:

combining said computed language model statistics into a single line of source material; and

,

displaying said single line of source material in said user interface.

5. The method of claim 1, wherein said language model statistics comprise m-grams.

6. The method of claim 5, wherein said language model statistics further comprise a frequency corresponding to each m-gram.

7. The method of claim 5, wherein said m-grams are selected from the group consisting of unigrams, bigrams and trigrams.

8. The method of claim 2, further comprising associating a relative frequency with each contextually related word in said selected one of said at least one incorporated class file.

9. The method according to claim 8, wherein said language model statistics are computed and re-computed according to said user-specified relative frequencies.

10. The method according to claim 8, wherein said displaying step comprises:

for each contextually related word in said selected one of said at least one incorporated class file, combining all re-computed m-grams and associated relative frequencies into a line of source material; and

,

displaying said line of source of material in a user interface.

11. A machine readable storage, having stored thereon a computer program generating language model statistics for a new word added to a language model incorporating at least one class file containing contextually related words, said computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:

(a) computing language model statistics based on references to at least one incorporated class file;

(b) substituting a new word for each reference to a selected one of said at least one incorporated class file, and re-computing said language model statistics based on said new word having been substituted for said reference;

(c) displaying said re-computed language model statistics in a user interface and accepting modifications to said re-computed language model statistics through said user interface; and

,

(d) further recomputing said language model statistics based on said modifications,

whereby said language model statistics are re-computed for said new word without introducing contextual inaccuracies in the language model.

12. The machine readable storage of claim 11, further comprising:

(e) if no modifications are accepted in step (c), adding said new word to said selected one of said at least one incorporated class file, said new word inheriting existing context-based language model statistics associated with said selected one of said at least one incorporated class file.

13. The machine readable storage of claim 12, further comprising repeating steps (a)-(e) for each class file associated with said new word.

14. The machine readable storage of claim 11, wherein said displaying step comprises:

combining said computed language model statistics into a single line of source material; and

,

displaying said single line of source material in said user interface.

15. The machine readable storage of claim 11, wherein said language model statistics comprise m-grams.

16. The machine readable storage of claim 15, wherein said language model statistics further comprise a frequency corresponding to each m-gram.

17. The machine readable storage of claim 15, wherein said m-grams are selected from the group consisting of unigrams, bigrams and trigrams.

18. The machine readable storage of claim 12, further comprising associating a relative frequency with each contextually related word in said selected one of said at least one incorporated class file.

19. The machine readable storage of claim 18, wherein said language model statistics are computed and re-computed according to said user-specified relative frequencies.

20. The machine readable storage of claim 18, wherein said displaying step comprises:

for each contextually related word in said selected one of said at least one incorporated class file, combining all re-computed m-grams and associated relative frequencies into a line of source material; and

,

displaying said line of source of material in a user interface.