Method and system for domain-specific noisy channel natural language processing (NLP)

• US 8,688,447 B1
• Filed: 08/21/2013
• Issued: 04/01/2014
• Est. Priority Date: 08/21/2013
• Status: Active Grant

• Current Assignee: Ask Ziggy Inc.
• Original Assignee: Ask Ziggy Inc.
• Inventors: De, Nirmalya Kumar, Grove, Kyle Wade
• Primary Examiner(s): Vo, Huyen X.
• Application Number:
  US13/972,300
• Time in Patent Office: 223 Days
• Field of Search: 704/231, 704/233, 704/235, 704/255, 704/251, 704/257, 704/270, 704/270.1
• US Class Current: 704/235
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

Abstract

A method for processing transcriptions using natural language processing (NLP), the method includes obtaining transcriptions corresponding to an utterance from a user device, where each of the transcriptions is a different speech-to-text conversion of the utterance. The method further includes tagging a first transcription with at least one entity tag and a first transcription-level tag to obtain a first tagged transcription, tagging the second transcription with at least one entity tag and a second transcription-level tag to obtain a second tagged transcription, determining a highest probability transcription-level tag from the first transcription-level tag and second transcription-level tag. The method further includes identifying a subject-matter domain using the highest probability transcription-level tag, retagging the first transcription and the second transcription using entity tags associated with the subject-matter domain to obtain retagged transcriptions, performing, using the retagged transcriptions, an action to obtain a result, and sending the result to the user device.

Claims

1. A non-transitory computer readable medium comprising instructions, which when executed perform a method for processing transcriptions using natural language processing (NLP), the method comprising:

obtaining a plurality of transcriptions corresponding to an utterance, wherein each of the transcriptions is a different speech-to-text conversion of the utterance, wherein the plurality of transcriptions comprises a first transcription and a second transcription, and wherein the utterance was obtained by a user device;

tagging the first transcription with at least one entity tag to obtain a first tagged transcription;

tagging the first transcription with a first transcription-level tag using the at least one entity tag associated with the first tagged transcription;

tagging the second transcription with at least one entity tag to obtain a second tagged transcription;

tagging the second transcription with a second transcription-level tag using the at least one entity tag associated with the second tagged transcription;

determining a highest probability transcription-level tag, wherein the highest probability transcription-level tag is one selected from a group consisting of the first transcription-level tag and the second transcription-level tag;

identifying a subject-matter domain using the highest probability transcription-level tag;

retagging the first transcription and the second transcription using entity tags associated with the subject-matter domain to obtain a retagged first transcription and a retagged second transcription;

performing, using the retagged first transcription and the retagged second transcription, an action to obtain result; and

sending the result to the user device.

2. The non-transitory computer readable medium of claim 1, wherein performing the action comprising:

identifying tagged entities in the retagged first transcription, wherein the tagged entities are used to perform action.

3. The non-transitory computer readable medium of claim 1, wherein performing the action comprising:

identifying a first tagged entity in the retagged first transcription,

identifying a second tagged entity in the retagged second transcription,

wherein the first tagged entity and the second tagged entity are used to perform the action.

4. The non-transitory computer readable medium of claim 1, wherein performing the action comprising:

identifying an entity tag in the subject-matter domain that is not associated with any entity in the first transcription and the second transcription;

generating a dialogue to obtain an entity for the entity tag;

sending the dialogue to the user device;

obtaining the entity from the user device in response to the dialogue, wherein the action is performed using the entity.

5. The non-transitory computer readable medium of claim 1, wherein the first transcription-level tag is different than the second transcription-level tag and wherein the highest probability transcription-level tag is the first transcription-level tag.

6. The non-transitory computer readable medium of claim 5, wherein the second transcription-level tag is associated with a second subject-matter domain, wherein the second subject-matter domain is different than the first subject-matter domain.

7. The non-transitory computer readable medium of claim 6, wherein the subject matter domain is associated with a first set of entity tags and the second subject-matter domain is associated with a second set of entity tags.

8. The non-transitory computer readable medium of claim 7, wherein the first tagged transcription comprises using at least the first set of entity tags and the second set of entity tags.

9. The non-transitory computer readable medium of claim 7, wherein at least one of the entity tags in first set of entity tags is not present in second set of entity tags.

10. The non-transitory computer readable medium of claim 1, wherein tagging the first transcription with the first transcription-level tag comprises using NLP.

11. The non-transitory computer readable medium of claim 10, wherein NLP comprises using a maximum entropy Markov model.

12. The non-transitory computer readable medium of claim 1, wherein determining a highest probability transcription-level tag comprises selecting one from a group consisting of the first transcription-level tag and the second transcription-level tag that has a highest marginal probability.

13. The non-transitory computer readable medium of claim 12, wherein the highest marginal probability for each of the transcription is determined using at least a confidence level associated with the transcription and a probability of a transcription-level tag being associated with the transcription.

14. A Natural Language Processing (NLP) system, comprising:

a processor;

a memory comprising instructions, which when executed by a processor perform a method, the method comprising;

obtaining a plurality of transcriptions corresponding to an utterance, wherein each of the transcriptions is a different speech-to-text conversion of the utterance, wherein the plurality of transcriptions comprises a first transcription and a second transcription, and wherein the utterance was obtained by a user device;

tagging the first transcription with at least one entity tag to obtain a first tagged transcription;

tagging the first transcription with a first transcription-level tag using the at least one entity tag associated with the first tagged transcription;

tagging the second transcription with at least one entity tag to obtain a second tagged transcription;

tagging the second transcription with a second transcription-level tag using the at least one entity tag associated with the second tagged transcription;

determining a highest probability transcription-level tag, wherein the highest probability transcription-level tag is one selected from a group consisting of the first transcription-level tag and the second transcription-level tag;

identifying a subject-matter domain using the highest probability transcription-level tag;

retagging the first transcription and the second transcription using entity tags associated with the subject-matter domain to obtain a retagged first transcription and a retagged second transcription;

performing, using the retagged first transcription and the retagged second transcription, an action to obtain result; and

sending the result to the user device.

15. The system of claim 14, wherein the user device is a mobile device.

16. The system of claim 14, wherein the user device comprising an application that is configured to send the utterance to the NLP system.

17. The system of claim 14, the method further comprising:

receiving the utterance;

providing the utterance to a third-party speech-to-text service; and

receiving the plurality of transcriptions from the third-party speech-to-text service.

18. A method for processing transcriptions using natural language processing (NLP), comprising:

obtaining a plurality of transcriptions corresponding to an utterance, wherein each of the transcriptions is a different speech-to-text conversion of the utterance, wherein the plurality of transcriptions comprises a first transcription and a second transcription, and wherein the utterance was obtained by a user device;

tagging the first transcription with at least one entity tag to obtain a first tagged transcription;

tagging the first transcription with a first transcription-level tag using the at least one entity tag associated with the first tagged transcription;

tagging the second transcription with at least one entity tag to obtain a second tagged transcription;

tagging the second transcription with a second transcription-level tag using the at least one entity tag associated with the second tagged transcription;

determining, by a processor, a highest probability transcription-level tag, wherein the highest probability transcription-level tag is one selected from a group consisting of the first transcription-level tag and the second transcription-level tag;

identifying a subject-matter domain using the highest probability transcription-level tag;

retagging the first transcription and the second transcription using entity tags associated with the subject-matter domain to obtain a retagged first transcription and a retagged second transcription;

performing, using the retagged first transcription and the retagged second transcription, an action to obtain result; and

sending the result to the user device.

19. The method of claim 18, wherein performing the action comprising:

identifying a first tagged entity in the retagged first transcription,

identifying a second tagged entity in the retagged second transcription,

wherein the first tagged entity and the second tagged entity are used to perform the action.

20. The method of claim 18, wherein performing the action comprising:

identifying an entity tag in the subject-matter domain that is not associated with any entity in the first transcription and the second transcription;

generating a dialogue to obtain an entity for the entity tag;

sending the dialogue to the user device;

obtaining the entity from the user device in response to the dialogue, wherein the action is performed using the entity.