Word clustering for input data

• US 20070118376A1
• Filed: 11/18/2005
• Published: 05/24/2007
• Est. Priority Date: 11/18/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Mukerjee, Kunal
• Granted Patent: US 8,249,871 B2
• Time in Patent Office: Days
• US Class Current: 704/245
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

A clustering tool to generate word clusters. In embodiments described, the clustering tool includes a clustering component that generates word clusters for words or word combinations in input data. In illustrated embodiments, the word clusters are used to modify or update a grammar for a closed vocabulary speech recognition application.

Claims

1. A computer implementable clustering tool comprising:

a clustering component (244) configured to receive input data (240) indicative of a plurality of utterances from a speech recognition component (224) and generate word clusters indicative of words co-occurring in utterances in the input data (240).

2. The clustering tool of claim 1 wherein the word clusters include word combinations of two or more words.

3. The clustering tool of claim 1 wherein the clustering component provides the word clusters and frequency of the word clusters.

4. The clustering tool of claim 1 and further comprising a merge component to merge clusters based upon a similarity metric.

5. The clustering tool of claim 1 and further comprising a pruning component to prune clusters based upon a pruning criteria.

6. The clustering tool of claim 1 and further including a reporting component which labels a word cluster utilizing word position information.

7. The clustering tool of claim 6 wherein the reporting component is configured to use a word cluster label which includes a wildcard to represent intermediate words relative to the co-occurring words.

8. The clustering tool of claim 1 wherein the clustering component includes a word occurrence vector generator and a vector dot product component to compute vector dot products.

9. The clustering tool of claim 8 and further comprising a word co-occurrence vector generator.

10. A computer implementable clustering tool comprising:

a word clustering component (244) configured to receive input data (240) and to generate at least one of word occurrence vectors or word co-occurrence vectors; and

a vector dot product component (294) configured to compute a vector dot product to obtain word clusters (246) in the input data (240) from the word occurrence vectors or the word co-occurrence vectors.

11. The clustering tool of claim 10 wherein the vector dot product component generates a word co-occurrence matrix including word clusters and frequency.

12. The clustering tool of claim 10 wherein the clustering component is configured to compute a pi-product to generate the word co-occurrence vectors.

13. A method comprising the steps of:

providing input data (240);

generating word occurrence vectors for words in the input data (240); and

computing a vector dot product between the word occurrence vectors to generate word clusters being indicative of words that co-occur in clusters in the input data (240).

14. The method of claim 13 and further comprising the steps of:

generating word co-occurrence vectors for the word clusters

computing a vector dot product for the word co-occurrence vectors.

15. The method of claim 13 and comprising the step of:

computing a pi-product to generate word co-occurrence vectors for the word clusters.

16. The method of claim 14 and comprising the step of:

repeating the steps of generating the word co-occurrence vectors and computing the vector dot product for additional word co-occurrence vectors.

17. The method of claim 13 and further comprising the steps of:

logging unrecognized speech data from a closed vocabulary speech recognition component; and

recognizing the logged speech data using a free-form speech recognition system to provide the input data.

18. The method of claim 13 and further comprising the step of:

outputting word clusters including a cluster label and frequency.

19. The method of claim 18 and further comprising the step of:

pruning cluster data prior to outputting the word clusters.

20. The method of claim 13 and further comprising the step of:

merging similar word clusters.