Methods and apparatus for generating dialog state conditioned language models

• US 20030187648A1
• Filed: 03/27/2002
• Published: 10/02/2003
• Est. Priority Date: 03/27/2002
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Visweswariah, Karthik, Printz, Harry W., Monkowski, Michael Daniel, Dharanipragada, Satyanarayana
• Granted Patent: US 7,143,035 B2
• Time in Patent Office: Days
• US Class Current: 704/260
• 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 2015/228 of application context

Abstract

Techniques are provided for generating improved language modeling. Such improved modeling is achieved by conditioning a language model on a state of a dialog for which the language model is employed. For example, the techniques of the invention may improve modeling of language for use in a speech recognizer of an automatic natural language based dialog system. Improved usability of the dialog system arises from better recognition of a user'"'"'s utterances by a speech recognizer, associated with the dialog system, using the dialog state-conditioned language models. By way of example, the state of the dialog may be quantified as: (i) the internal state of the natural language understanding part of the dialog system; or (ii) words in the prompt that the dialog system played to the user.

Claims

1. A method for use in accordance with a dialog system, the method comprising the steps of:

generating at least one language model, the at least one language model being conditioned on a state of dialog associated with the dialog system; and

storing the at least one language model for subsequent use in accordance with a speech recognizer associated with the dialog system.

2. The method of claim 1, wherein the dialog system is a natural language understanding based dialog system.

3. The method of claim 2, wherein the state corresponds to an internal state of the natural language understanding part of the dialog system.

4. The method of claims 1, wherein the state corresponds to a prompt that the dialog system presents to a user.

5. The method of claim 1, wherein the step of generating the at least one language model conditioned on a state of dialog associated with the dialog system further comprises the steps of:

dividing training data which is labeled by state into different state sets depending on the state to which the training data belongs; and

building a separate language model for each of the state sets.

6. The method of claim 5, wherein each separate language model is a trigram language model.

7. The method of claim 5, wherein each separate language model is built using a modified Kneser-Ney smoothing technique.

8. The method of claims 5, wherein each separate language model is interpolated with a base model obtained from available training data for a domain of the dialog system.

9. The method of claim 5, further comprising the step of clustering together two or more state sets to reduce the number of states for which a separate language model is built.

10. The method of claim 9, wherein a decision to cluster two states is based on a distance measure computed between respective word distributions associated with the two states.

11. The method of claim 1, wherein the step of generating the at least one language model conditioned on a state of dialog associated with the dialog system further comprises the step of building a model with one or more feature functions incorporating the state of dialog.

12. The method of claim 11, wherein the feature function model is an exponential model based on a trigram model.

13. The method of claim 11, wherein parameters of the feature function model are chosen to optimize likelihood of training data.

14. The method of claim 13, wherein optimization is based on an iterative scaling algorithm.

15. The method of claim 11, wherein the one or more feature functions are chosen based on one of a number of times that associated features occur in training data and a gain in likelihood realized by using a particular feature.

16. The method of claim 1, wherein the step of generating the at least one language model conditioned on a state of dialog associated with the dialog system further comprises the steps of:

caching words in a prompt presented to a user by the dialog system;

building a unigram model on the cached words; and

interpolating the unigram model with a baseline model.

17. The method of claim 16, wherein the baseline model is a trigram language model trained from available training data for a domain of the dialog system.

18. Apparatus for use in accordance with a dialog system, the apparatus comprising:

at least one processor operative to generate at least one language model, the at least one language model being conditioned on a state of dialog associated with the dialog system; and

memory, coupled to the at least one processor, for storing the at least one language model for subsequent use in accordance with a speech recognizer associated with the dialog system.

19. The apparatus of claim 18, wherein the dialog system is a natural language understanding based dialog system.

20. The apparatus of claim 19, wherein the state corresponds to an internal state of the natural language understanding part of the dialog system.

21. The apparatus of claims 18, wherein the state corresponds to a prompt that the dialog system presents to a user.

22. The apparatus of claim 18, wherein the operation of generating the at least one language model conditioned on a state of dialog associated with the dialog system further comprises:

(i) dividing training data which is labeled by state into different state sets depending on the state to which the training data belongs; and

(ii) building a separate language model for each of the state sets.

23. The apparatus of claim 18, wherein the operation of generating the at least one language model conditioned on a state of dialog associated with the dialog system further comprises building a model with one or more feature functions incorporating the state of dialog.

24. The apparatus of claim 18, wherein the operation of generating the at least one language model conditioned on a state of dialog associated with the dialog system further comprises:

(i) caching words in a prompt presented to a user by the dialog system;

(ii) building a unigram model on the cached words; and

(iii) interpolating the unigram model with a baseline model.

25. An article of manufacture for use in accordance with a dialog system, comprising a machine readable medium containing one or more programs which when executed implement the steps of:

generating at least one language model, the at least one language model being conditioned on a state of dialog associated with the dialog system; and

storing the at least one language model for subsequent use in accordance with a speech recognizer associated with the dialog system.

26. A data structure for use in accordance with a dialog system, the data structure comprising:

at least one language model, the at least one language model being conditioned on a state of dialog associated with the dialog system, and the at least one language model being for subsequent use in accordance with a speech recognizer associated with the dialog system.

27. A dialog system, comprising:

at least one language model, the at least one language model being conditioned on a state of dialog associated with the dialog system; and

a speech recognizer for recognizing speech utilizing the at least one language model.