Multitask learning for spoken language understanding

Patent metadata

Field	Value
Publication number	US-9406292-B2
Application number	US-201414287892-A
Country	US
Kind code	B2
Filing date	May 27, 2014
Priority date	Jun 9, 2006
Publication date	Aug 2, 2016
Grant date	Aug 2, 2016

Abstract

Systems for improving or generating a spoken language understanding system using a multitask learning method for intent or call-type classification. The multitask learning method aims at training tasks in parallel while using a shared representation. A computing device automatically re-uses the existing labeled data from various applications, which are similar but may have different call-types, intents or intent distributions to improve the performance. An automated intent mapping algorithm operates across applications. In one aspect, active learning is employed to selectively sample the data to be re-used.

First claim

I claim:

1. A method comprising:
mapping call-types between a first spoken dialog system and a second spoken dialog system using individual training models for each spoken dialog system, to yield mapped call-types; and
retraining a model of the individual training models using information based on the mapped call-types.

2. The method of claim 1 , wherein the mapping of the call-types comprises performing on of splitting the call-types, merging the call-types, and renaming the call-types.

3. The method of claim 2 , wherein the merging of the call-types comprises cross-labeling utterances from a dialog using the individual training models.

4. The method of claim 3 , wherein the utterances which are cross-labeled have a confidence score above a threshold.

5. The method of claim 1 , further comprising labeling, as a new call-type, a call-type of the first spoken dialog system when the call-type has more than a specified ratio among the call-types.

6. The method of claim 1 , wherein the retraining of the model comprises active learning to selectively sample data used for the retraining.

7. The method of claim 6 , wherein selectively sampled data is reused during retraining.

8. A system comprising:
a processor; and
a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising:
mapping call-types between a first spoken dialog system and a second spoken dialog system using individual training models for each spoken dialog system, to yield mapped call-types; and
retraining a model of the individual training models using information based on the mapped call-types.

9. The system of claim 8 , wherein the mapping of the call-types comprises performing on of splitting the call-types, merging the call-types, and renaming the call-types.

10. The system of claim 9 , wherein the merging of the call-types comprises cross-labeling utterances from a dialog using the individual training models.

11. The system of claim 10 , wherein the utterances which are cross-labeled have a confidence score above a threshold.

12. The system of claim 8 , the computer-readable storage medium having additional instructions stored which result in operations comprising labeling, as a new call-type, a call-type of the first spoken dialog system when the call-type has more than a specified ratio among the call-types.

13. The system of claim 8 , wherein the retraining of the model comprises active learning to selectively sample data used for the retraining.

14. The system of claim 13 , wherein selectively sampled data is reused during retraining.

15. A computer-readable storage device having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:
mapping call-types between a first spoken dialog system and a second spoken dialog system using individual training models for each spoken dialog system, to yield mapped call-types; and
retraining a model of the individual training models using information based on the mapped call-types.

16. The computer-readable storage device of claim 15 , wherein the mapping of the call-types comprises performing on of splitting the call-types, merging the call-types, and renaming the call-types.

17. The computer-readable storage device of claim 16 , wherein the merging of the call-types comprises cross-labeling utterances from a dialog using the individual training models.

18. The computer-readable storage device of claim 17 , wherein the utterances which are cross-labeled have a confidence score above a threshold.

19. The computer-readable storage device of claim 15 , having additional instructions stored which result in operations comprising labeling, as a new call-type, a call-type of the first spoken dialog system when the call-type has more than a specified ratio among the call-types.

20. The computer-readable storage device of claim 15 , wherein the retraining of the model comprises active learning to selectively sample data used for the retraining.

Assignees

• At & T Ip Ii Lp

Inventors

• Tur Gokhan

Classifications

• G10L15/183

  using context dependencies, e.g. language models · CPC title

• G10L15/1822 Primary

  Parsing for meaning understanding · CPC title

• G10L15/26

  Speech to text systems (G10L15/08 takes precedence) · CPC title

• H04M3/4936

  Speech interaction details (speech recognition per se G10L15/00) · CPC title

• G10L13/08

  Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination · CPC title