System and method for modifying a language model and post-processor information

• US 20050165598A1
• Filed: 09/29/2004
• Published: 07/28/2005
• Est. Priority Date: 10/01/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Dictaphone Corporation (Microsoft Corporation)
• Inventors: Cote, William F., Chadbourne, Jeffrey
• Granted Patent: US 7,774,196 B2
• Time in Patent Office: Days
• US Class Current: 704/1
• 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
• CPC Class Codes: G10L 2015/227 of the speaker; Human-factor methodology
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A system and method for automatically modifying a language model and post-processor information is provided. The system includes the steps of language model identification, site-specific model creation, and language model adaptation.

Claims

1. A system and method for modifying a language model and post-processor information, the method comprising the steps of:

collecting a set of language models;

collecting a set of corrected dictated reports;

collecting site specific data from a user;

matching each dictated report with each language model;

determining a perplexity factor for each language model; and

determining the language model having the lowest perplexity factor.

2. The method according to claim 1, wherein site specific data comprises add word lists and post processing information such as spacing and capitalization.

3. The method according to claim 1, further comprising creating a site-specific language model by adding the site specific data to the language model having the lowest perplexity factor.

4. The method according to claim 3, further comprising adapting the site specific language model to create a task language model, and outputting the task language model.

5. The method according to claim 4, wherein adapting the site specific language model comprises:

updating a set of statistics for the language model; and

adding an accent word list to the language model to create a task language model.

6. The method according to claim 5, wherein the set of statistics includes bigrams or trigrams.

7. The method according to claim 1, further comprising adapting the language model with the lowest perplexity factor to create a task language model, and outputting the task language model.

8. The method according to claim 7, wherein adapting the language model with the lowest perplexity factor comprises:

updating a set of statistics for the language model; and

adding an accent word list to the language model to create the task language model.

9. The method according to claim 8, wherein the set of statistics includes bigrams or trigrams.

10. A system and method for modifying a language model and post-processor information, the method comprising the steps of:

collecting the language model;

collecting a set of corrected dictated reports;

collecting site specific data from a user;

determining whether the user has an accent;

creating an accent specific list based on the user'"'"'s accent; and

creating a site-specific language model by adding the site specific data to the language model.

11. The method according to claim 10, further comprising adapting the site specific language model to create a task language model, and outputting the task language model.

12. The method according to claim 11, wherein adapting the site specific language model comprises:

updating a set of statistics for the language model; and

adding an accent word list to the language model to create the task language model.

13. The method according to claim 12, wherein the set of statistics includes bigrams or trigrams.

14. A system and method for modifying a language model and post-processor information, the method comprising the steps of:

collecting the language model;

collecting a set of corrected dictated reports;

collecting site specific data from a user;

determining whether the user has an accent;

creating an accent specific list based on the user'"'"'s accent; and

adapting the language model to create a task language model, and outputting the task language model.

15. The method according to claim 14, wherein adapting the language model comprises:

updating a set of statistics for the language model; and

adding an accent word list to the language model to create the task language model.

16. The method according to claim 15, wherein the set of statistics includes bigrams or trigrams.