Method and apparatus for distribution-based language model adaptation

• US 7,254,529 B2
• Filed: 09/13/2005
• Issued: 08/07/2007
• Est. Priority Date: 10/13/2000
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Li, Mingjing, Gao, Jianfeng
• Primary Examiner(s): Hudspeth; David
• Assistant Examiner(s): Serrou; Abselali
• Application Number:
  US11/225,543
• Publication Number:
  US 20060009965A1
• Time in Patent Office: 693 Days
• Field of Search: 704/255, 704/250, 704/9, 704/257, 704/25, 704/10, 704/240, 704/243, 704/244
• US Class Current: 704/9
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

Abstract

A method and apparatus are provided for adapting a language model to a task-specific domain. Under the method and apparatus, the relative frequency of n-grams in a small training set (i.e. task-specific training data set) and the relative frequency of n-grams in a large training set (i.e. out-of-domain training data set) are used to weight a distribution count of n-grams in the large training set. The weighted distributions are then used to form a modified language model by identifying probabilities for n-grams from the weighted distributions.

Claims

1. A method of forming a language model, the method comprising:

selecting out-of-task training data having n-gram counts;

selecting task-specific training data having n-gram counts;

modifying an n-gram count for an n-gram in the out-of-task training data by applying an n-gram-specific weight that is based in part on an n-gram count for the n-gram in the task-specific training data to form modified training data, wherein the n-gram count from the task-specific training data is only used for forming the n-gram-specific weight;

identifying probabilities for the language model based on the modified training data; and

storing the identified probabilities for the language model.

2. The method of claim 1 wherein applying a weight comprises forming a weight based on a relative frequency of an n-gram in the task-specific training data.

3. The method of claim 2 wherein forming a weight further comprises forming a weight based on a relative frequency of the n-gram in the out-of-task data.

4. The method of claim 1 wherein identifying probabilities for the language model comprises using only the modified data.

5. A computer-readable storage medium having computer-executable instructions for forming a language model through steps comprising:

determining a count of an entity in a small set of training data;

changing a count of the entity in a large set of training data based on an entity-specific weight that is formed in part from the count of the entity in the small set of training data to form a modified count of the entity, wherein the count of the entity in the small set of training data is only used for forming the entity-specific weight;

using the modified count of the entity to identify probabilities for the language model; and

storing the probabilities for the language model.

6. The computer-readable storage medium of claim 5 wherein the entities comprise n-grams.

7. The computer-readable storage medium of claim 5 wherein the entity-specific weight is formed based in part on a relative frequency of the entity in the small set of training data.

8. The computer-readable storage medium of claim 7 wherein the entity-specific weight is further formed based on a relative frequency of the entity in the large set of training data.

9. A method of adapting a general language model to cover a task-specific domain, the method comprising:

determining a weight based on the relative frequency of an n-gram in the task-specific domain;

multiplying the weight by a count of the n-gram in a distribution domain associated with the general language model to form a modified count, the modified count formed without adding counts of n-grams in the task-specific domain;

using only the modified counts of n-grams to determine a probability that forms part of an adapted language model; and

storing the probability as part of the adapted language model.

10. The method of claim 9 wherein determining a weight comprises determining a separate weight for each of a plurality of n-grams.

11. A computer-readable storage medium having computer-executable components for performing steps comprising:

selecting out-of-task training data having n-gram counts;

selecting task-specific training data having n-gram counts;

modifying an n-gram count for an n-gram in the out-of-task training data by applying an n-gram-specific weight that is based in part on an n-gram count for the n-gram in the task-specific training data to form modified training data, wherein the n-gram count from the task-specific training data is only used for forming the n-gram-specific weight;

identifying probabilities for the language model based on the modified training data; and

storing the probabilities for the language model.

12. The computer-readable storage medium of claim 11 wherein applying a weight comprises forming a weight based on a relative frequency of an n-gram in the task-specific training data.

13. The computer-readable storage medium of claim 12 wherein forming a weight further comprises forming a weight based on a relative frequency of the n-gram in the out-of-task data.

14. The computer-readable storage medium of claim 11 wherein identifying probabilities for the language model comprises using only the modified data.