System and method for managing conversation

• US 9,514,742 B2
• Filed: 01/18/2013
• Issued: 12/06/2016
• Est. Priority Date: 05/17/2012
• Status: Expired due to Fees

• Current Assignee: Postech Academy-Industry Foundation
• Original Assignee: Postech Academy-Industry Foundation
• Inventors: Lee, Gary Geunbae, Noh, Hyungjong, Lee, Kyusong
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US14/386,927
• Publication Number:
  US 20150058014A1
• Time in Patent Office: 1,418 Days
• Field of Search: 704/270, 704/2, 704/231, 704/236, 704/243, 704/254, 704/257, 707/999.004
• US Class Current: 1/1
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G09B 19/06 Foreign languages with audible presentation of material to be studied G09B5/04
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 2015/0633 using lexical or orthographic knowledge sources

Abstract

A conversation management system includes: a training unit that generates an articulation speech act and an entity name of a training corpus, that generates a lexical syntactic pattern, and that estimates a speech act and an entity name of a training corpus; a database that stores the articulation speech act, the entity name, and the lexical syntactic pattern of the training corpus; an execution unit that generates an articulation speech act and an entity name of a user, that generates a user lexical syntactic pattern, that estimates a speech act and an entity name of a user, that searches for an articulation pair corresponding to a user articulation at the database using a search condition including the estimated user speech act and the generated user lexical syntactic pattern, and that generates a final response by selecting an articulation template using a restriction condition including an estimated entity name among the found articulation pair; and an output unit that outputs a final response that is generated by the execution unit.

Claims

1. A conversation management system, comprising:

a training unit that generates an articulation speech act and an entity name of a training corpus, that generates a lexical syntactic pattern which is generated from the training corpus, and that estimates the articulation speech act and the entity name of the training corpus;

a database that stores the articulation speech act, the entity name, and the lexical syntactic pattern of the training corpus;

an input unit receiving a user articulation when an actual conversation is performed by a user;

an execution unit that generates a user articulation speech act and an entity name of the user articulation, that generates a user lexical syntactic pattern, that estimates the user articulation speech act and the entity name of the user articulation, that searches for an articulation pair corresponding to the user articulation at the database using a search condition comprising the estimated user articulation speech act and the generated user lexical syntactic pattern, and that generates a final response by selecting an articulation template using a restriction condition comprising the estimated entity name of the user articulation among the found articulation pair; and

an output unit that outputs the final response that is generated by the execution unit,

wherein the training unit analyzes a part of speech of the training corpus, defines a priority of parts of speech, and generates the lexical syntactic pattern from the part of speech, and

wherein the execution unit estimates the user speech act and the entity name of the user articulation using a machine learning training model generated by the training unit.

2. The conversation management system of claim 1, wherein the training unit displays entity name information in a word of the training corpus, and attaches a symbol when the entity name cannot be replaced.

3. The conversation management system of claim 1, wherein the execution unit comprises a search relieving unit,

wherein the search relieving unit enlarges a search range to a plurality of speech acts comprising the estimated user articulation speech act, and searches again for an articulation pair when there is no articulation that is selected by the execution unit.

4. The conversation management system of claim 1, wherein the execution unit comprises a search relieving unit,

wherein the search relieving unit enlarges a search range by reducing a range of the user lexical syntactic pattern, and searches again for an articulation pair when no articulation is selected by the execution unit.

5. The conversation management system of claim 1, wherein the execution unit comprises a search relieving unit,

wherein the search relieving unit enlarges a search range by reducing a range of the user lexical syntactic pattern and a plurality of speech acts comprising the estimated user articulation speech act, and searches again for an articulation pair when there is no articulation selected by the execution unit.

6. The conversation management system of claim 1, wherein the training unit comprises:

a machine learning model generator that estimates the articulation speech act and the entity name of the training corpus;

an entity name information display unit that displays an information of the articulation speech act and the entity name of the training corpus, and that displays information between the displayed entity name and articulation;

an articulation part of speech analysis unit that analyzes an articulation part of speech; and

a lexical syntactic pattern generator that generates the lexical syntactic pattern from the analyzed part of speech.

7. The conversation management system of claim 1, wherein the execution unit comprises:

a speech act analysis unit that estimates the user articulation speech act;

an entity name recognition unit that recognizes the entity name of the user articulation;

an articulation pair search unit that searches for an articulation pairs corresponding to the user articulation using the user articulation speech act, the user lexical syntactic pattern, and the entity name of the user articulation; and

a restriction condition applying unit that searches for an articulation pair satisfying an entity name restriction among the articulation pairs that are found in the articulation pair search unit.

8. The conversation management system of claim 1, wherein the database indexes and stores the articulation speech act, the entity name, and the lexical syntactic pattern.

9. The conversation management system of claim 1, wherein the training unit comprises a machine learning model generator,

wherein the machine learning model generator estimates the articulation speech act of the training corpus through a classifier that is generated according to a machine learning algorithm.

10. A method of managing a conversation, the method comprising:

a training step of generating an articulation speech act and an entity name of a training corpus, generating a lexical syntactic pattern which is generated from the training corpus, and estimating the articulation speech act and the entity name of the training corpus;

storing the articulation speech act, the entity name of the training corpus, and the lexical syntactic pattern of the training corpus at to a database;

receiving a user articulation when an actual conversation is performed by a user;

an execution step of generating a user articulation speech act and an entity name of the user articulation, generating a user lexical syntactic pattern, estimating the user articulation speech act and the entity name of the user articulation, searching for an articulation pair corresponding to the user articulation at the database using a search condition comprising the estimated user articulation speech act and the generated user lexical syntactic pattern, and selecting an articulation template and generating a final response using a restriction condition comprising the estimated entity name of the user articulation among the found articulation pairs; and

outputting the final response that is generated at the execution step,

wherein the training step comprises;

analyzing a part of speech of the training corpus;

defining a priority of the parts of speech; and

generating the lexical syntactic pattern from the part of speech, and

wherein the user articulation speech act and the entity name of the user articulation is estimated by using a machine learning training model generated in the training step.

11. The method of claim 10, wherein the training step comprises:

displaying entity name information in a word of the training corpus;

displaying information between an entity name and an articulation in the training corpus; and

training a classifier that estimates the articulation speech act and the entity name of the training corpus.

12. The method of claim 10, wherein the execution step comprises relieving a search,

wherein the relieving of a search comprises searching again for, if there is no articulation selected at the execution step, an articulation pair by enlarging a search range to a plurality of speech acts comprising the estimated user articulation speech act.

13. The method of claim 10, wherein the execution step comprises relieving a search,

wherein the relieving of a search comprises searching again for, if there is no articulation selected at the execution step, an articulation pair by enlarging a search range by reducing a range of the user lexical syntactic pattern.

14. The method of claim 10, wherein the execution step comprises relieving a search,

wherein the relieving of a search comprises searching again for, if there is no articulation selected at the execution step, an articulation pair by enlarging a search range by reducing a range of the user lexical syntactic pattern and a plurality of speech acts comprising the estimated user articulation speech act.

15. The method of claim 10, wherein the training step comprises:

estimating the articulation speech act and the entity name of the training corpus;

displaying an information of the articulation speech act and the entity name of the training corpus and displaying information between the displayed entity name and articulation;

analyzing an articulation part of speech; and

generating the lexical syntactic pattern from the analyzed part of speech.

16. The method of claim 10, wherein the execution step comprises:

estimating the user articulation speech act;

recognizing the entity name of the user articulation;

searching for articulation pairs corresponding to the user articulation using the user articulation speech act, the user lexical syntactic pattern, and the entity name of the user articulation; and

searching for an articulation pair satisfying an entity name restriction among the found articulation pairs.

17. The method of claim 10, wherein the storing of the articulation speech act comprises indexing and storing the articulation speech act, the entity name, and the lexical syntactic pattern.

18. The method of claim 10, wherein the training step comprises generating the machine learning training model,

wherein the generating of the machine learning training model comprises estimating the articulation speech act of the training corpus through a classifier that is generated according to a machine learning algorithm.