Automatic Editing Using Probabilistic Word Substitution Models

• US 20080243500A1
• Filed: 03/30/2007
• Published: 10/02/2008
• Est. Priority Date: 03/30/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Divay, Olivier, Bisani, Maximilian, Vozila, Paul
• Granted Patent: US 7,813,929 B2
• Time in Patent Office: Days
• US Class Current: 704/235
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

An input sequence of unstructured speech recognition text is transformed into output structured document text. A probabilistic word substitution model is provided which establishes association probabilities indicative of target structured document text correlating with source unstructured speech recognition text. The input sequence of unstructured speech recognition text is looked up in the word substitution model to determine likelihoods of the represented structured document text corresponding to the text in the input sequence. Then, a most likely sequence of structured document text is generated as an output.

Claims

1. A method of transforming an input sequence of unstructured speech recognition text into output structured document text, the method comprising:

providing a probabilistic word substitution model to establish association probabilities indicative of target structured document text correlating with source unstructured speech recognition text;

considering a set of candidate sequences of structured document text based on the word substitution model with respect to an input sequence of unstructured speech recognition text;

evaluating the likelihood of candidates corresponding to the input sequence of unstructured speech recognition text;

determining as an output a most likely sequence of structured document text.

2. A method according to claim 1, wherein the most likely sequence of structured document text is derived using a dynamic programming beam search.

3. A method according to claim 1, wherein the most likely sequence of structured document text is derived from local posterior probabilities determined using a forward-backward algorithm.

4. A method according to claim 1, wherein the word substitution model derives probabilities by a log-linear combination of one or more feature functions.

5. A method according to claim 4, wherein the word substitution model includes interpolation parameters.

6. A method according to claim 4, wherein the feature functions are defined in terms of associations between a source word and a sequence of target words.

7. A method according to claim 1, where the probabilistic word substitution model is derived based on a set of archived dictations and corresponding text documents.

8. A method according to claim 1, wherein the word substitution model uses speaker dependent models.

9. A method according to claim 1, wherein the word substitution model uses speaker independent models.

10. A method according to claim 1, wherein the word substitution model uses a combination of speaker dependent models and speaker independent models.

11. A method according to claim 1, wherein the structured document text includes medical report document text.

12. A method according to claim 1, wherein the structured document text includes legal document text.

13. A computer program product in a computer readable storage medium for transforming an input sequence of unstructured speech recognition text into output structured document text, the product comprising:

program code for providing a probabilistic word substitution model to establish association probabilities indicative of target structured document text correlating with source unstructured speech recognition text;

program code for considering a set of candidate sequences of structured document text based on the word substitution model with respect to an input sequence of unstructured speech recognition text;

program code for evaluating the likelihood of candidates corresponding to the input sequence of unstructured speech recognition text;

program code for determining as an output a most likely sequence of structured document text.

14. A computer program product according to claim 13, wherein the most likely sequence of structured document text is derived using a dynamic programming beam search.

15. A computer program product according to claim 13, wherein the most likely sequence of structured document text is derived from local posterior probabilities determined using a forward-backward search.

16. A computer program product according to claim 13, wherein the word substitution model derives probabilities by a log-linear combination of one or more feature functions.

17. A computer program product according to claim 16, wherein the word substitution model includes interpolation parameters.

18. A computer program product according to claim 16, wherein the feature functions are defined in terms of associations between a source word and a sequence of target words.

19. A computer program product according to claim 13, where the probabilistic word substitution model is derived based on a set of archived dictations and corresponding text documents.

20. A computer program product according to claim 13, wherein the word substitution model uses speaker dependent models.

21. A computer program product according to claim 13, wherein the word substitution model uses speaker independent models.

22. A computer program product according to claim 13, wherein the word substitution model uses a combination of speaker dependent models and speaker independent models.

23. A computer program product according to claim 13, wherein the structured document text includes medical report document text.

24. A computer program product according to claim 13, wherein the structured document text includes legal document text.