Automatic generation of superwords

• US 6,021,384 A
• Filed: 10/29/1997
• Issued: 02/01/2000
• Est. Priority Date: 10/29/1997
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Gorin, Allen Louis, Riccardi, Giuseppe
• Primary Examiner(s): Isen, Forester W.
• Assistant Examiner(s): EDOUARD, PATRICK NESTOR
• Application Number:
  US08/960,291
• Time in Patent Office: 825 Days
• Field of Search: 704/1, 704/9, 704/10, 704/251, 704/255, 704/256, 704/257
• US Class Current: 704/1
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

This invention is directed to the automatic selection and generation of superwords based on a criterion relevant to speech recognition. Superwords are used to refer to those word combinations which are so often spoken that they can be recognized by a recognition device as being a single word.

Claims

1. A method for determining superwords, comprising the steps of:

generating a set of candidate phrases from a database of observed sequences of words, symbols and/or sounds, the database having a perplexity value based on a language model;

incorporating one of the candidate phrases from the database into the language model;

analyzing how the perplexity value of the database, which is based on the language model, is affected by the incorporated candidate phrase; and

determining if the candidate phrase is a superword based on the analyzing step.

2. The method of claim 1, wherein the determining step determines that if the perplexity value for the database decreases in the analyzing step, the candidate phrase is added to the database as a superword.

3. The method of claim 2, wherein the determining step further determines a new perplexity value for the database using the superword.

4. The method of claim 1, wherein the determining step determines that if the perplexity value for the database does not decrease in the analyzing step, the candidate phrase is rejected as a superword.

5. The method of claim 3, wherein the inserting, analyzing, and determining steps are repeated for another one of the candidate phrases using the new perplexity value.

6. The method of claim 1, wherein the superword is a word pairs.

7. The method of claim 1, wherein the superword is a plurality of adjacent words.

8. The method of claim 1, wherein the database is used by a speech recognizer.

9. An apparatus that determines superwords, comprising:

a database of observed sequences of words, symbols and/or sounds, the database having a perplexity value based on a language model;

generating means for generating candidate phrases from the database;

input means for incorporating one of the candidate phrases from the database into the language model;

analyzing means for analyzing how the perplexity value of the database, which is based on the language model, is affected by the incorporated candidate phrase, the analyzing means producing an output; and

determining means for determining if the candidate phrase is a superword from the output of the analyzing means.

10. The apparatus of claim 9, wherein the determining means decides based on the output of the analyzing means that if the perplexity value for the database decreases, the candidate phrase is added to the database as a superword.

11. The apparatus of claim 10, wherein the determining means determines a new perplexity value for the database using the superword.

12. The apparatus of claim 11, wherein the analyzing means uses the new perplexity value determined by the determining means for another candidate phrase.

13. The apparatus of claim 9, wherein the determining means decides based on the output of the analyzing means that if the perplexity value of the database is not decreased, the candidate phrase is rejected.

14. The apparatus of claim 9, wherein the superword is a word pairs.

15. The apparatus of claim 9, wherein the superword is a plurality of adjacent words.

16. The apparatus of claim 9, wherein the database is used by a speech recognizer.

17. An apparatus that generates and selects superwords, comprising:

a database of observed sequences of words, symbols and/or sounds, the database having a perplexity value based on a language model; and

a superword selector that generates candidate phrases from the database, the selector incorporating one of the candidate phrases from the database into the language model, the selector further analyzing how the perplexity value of the database, which is based on the language model, is affected by the incorporated candidate phrase and determining whether the candidate phrase is a superword based on the analysis.

18. The apparatus of claim 17, wherein if the superword selector determines that the perplexity value of the database decreases when the candidate phrase is inserted into the database, the candidate phrase is added to the database as a superword.

19. The apparatus of claim 17, wherein if the superword selector determines that the perplexity value is not decreased when the candidate phrase is inserted into the database, the candidate phrase is rejected as a superword.

20. The apparatus of claim 17, further comprising:

a recognizer that uses the database to recognize any of the superwords contained in a user'"'"'s input.

21. The apparatus of claim 20, wherein the recognizer is a speech recognizer and the user'"'"'s input is speech.