Question answering system and method for structured knowledgebase using deep natural language question analysis

• US 9,633,006 B2
• Filed: 11/04/2015
• Issued: 04/25/2017
• Est. Priority Date: 11/20/2014
• Status: Active Grant

• Current Assignee: Hyundai Motor Company (Hyundai Motor Group), Kia Corp.
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Ryu, Pum Mo, Kim, Hyun Ki, Bae, Yong Jin, Oh, Hyo Jung, Lee, Chung Hee, Lim, Soo Jong, Lim, Joon Ho, Jang, Myung Gil, Choi, Mi Ran, Heo, Jeong
• Primary Examiner(s): YEN, ERIC L
• Application Number:
  US14/932,604
• Publication Number:
  US 20160147737A1
• Time in Patent Office: 538 Days
• CPC Class Codes: G06F 16/243 Natural language query formulation
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Disclosed are a question answering system for structured knowledgebase using deep natural language question analysis, and a method thereof, the question answering system for structured knowledgebase using deep natural language question analysis includes a deep natural language question analysis unit configured to create a structure of a semantic frame by analyzing a natural language question that is input, a question-intermediate expression creation unit configured to create a question-intermediate expression of a lexicon level based on the semantic frame, a knowledgebase-specialized query creation unit configured to create a query used to search in knowledgebase that is a subject of search, based on the question-intermediate expression, and a knowledgebase search unit configured to find a correct answer in the knowledgebase that is subject of search based on the query, to provide an accuracy of the correct answer, a confidence of the correct answer and an evidence for the correct answer.

Claims

1. A question answering system for structured knowledgebase using deep natural language question analysis, the question answering system comprising:

one or more processors that process computer executable program code embodied in computer readable storage media, the computer executable program code comprising;

deep natural language question analysis program code that creates a structure of a semantic frame by analyzing a natural language question that is input;

question-intermediate expression creation program code that creates a question-intermediate expression of a lexicon level based on the semantic frame;

knowledgebase-specialized query creation program code that creates a query used to search in a knowledgebase that is a subject of search, based on the question-intermediate expression; and

knowledgebase search program code that finds a correct answer in the knowledgebase that is the subject of search based on the query, and that provides an accuracy of the correct answer, a confidence of the correct answer and an evidence for the correct answer.

2. The question answering system of claim 1, wherein the deep natural language question analysis program code creates the semantic frame by analyzing the natural language question in terms of morpheme, named entity, syntax and semantic role.

3. The question answering system of claim 2, wherein the deep natural language question analysis program code divides the natural language question in units of morphemes, and assigns a morpheme tag to each of the morphemes.

4. The question answering system of claim 2, wherein the deep natural language question analysis program code recognizes an entity of the natural language question based on a result of a morpheme analysis of the natural language question, and assigns a type to the entity.

5. The question answering system of claim 2, wherein the deep natural language question analysis program code assigns a dependency relation between word-phrases by analyzing a dependency syntax structure of the natural language question based on a result of a morpheme analysis of the natural language question.

6. The question answering system of claim 2, wherein the deep natural language question analysis program code determines a semantic role of an argument for a predicate of the natural language question based on results of a morpheme analysis and a syntax analysis of the natural language question.

7. The question answering system of claim 6, wherein the semantic role is defined as a key semantic case and an additional semantic case, and each of the key semantic case and the additional semantic case is assigned a different meaning at each predicate of the natural language question.

8. The question answering system of claim 2, wherein the deep natural language question analysis program code creates the structure of a semantic frame having dependents, including a subject, an object and an adverb with respect to a predicate of the natural language question, as arguments based on results of a morpheme analysis, a syntax analysis and a semantic role analysis of the natural language question.

9. The question answering system of claim 8,

wherein the deep natural language question analysis program code sequentially expresses dependents including a subject, an object and an adverb of a certain predicate in the structure of the semantic frame as arguments;

and

assigns a morpheme analysis result and a named entity analysis result to each argument, and assigns a syntax relation and a semantic role between the certain predicate and each argument.

10. The question answering system of claim 1, wherein the question-intermediate expression creation program code recognizes a question focus from the natural language question, and creates a lexicon level-query format based on a semantic frame including the question focus.

11. The question answering system of claim 10,

wherein the question-intermediate expression creation program code creates a question-intermediate expression including a variable by which the question focus is replaced using a certain sign and a condition which expresses restriction information about the variable;

and

the condition is expressed by a type predicate semantic frame which represents a type of a correct answer for the variable, and a lexicon predicate semantic frame that has a format obtained by replacing the question focus in the semantic frame by the variable.

12. The question answering system of claim 1, wherein the knowledgebase-specialized query creation program code creates a query specialized in a certain knowledgebase by mapping an argument of the question-intermediate expression into a class defined in the certain knowledgebase or an instance, and mapping a predicate of the question-intermediate expression into a property defined in the certain knowledgebase.

13. A question answering method for structured knowledgebase using deep natural language question analysis, the question answering method comprising:

creating a structure of a semantic frame by analyzing a natural language question that is input;

creating a question-intermediate expression of a lexicon level based on the semantic frame;

creating a query used to search in a knowledgebase that is a subject of search, based on the question-intermediate expression; and

finding a correct answer in the knowledgebase that is the subject of search based on the query, and providing an accuracy of the correct answer, a confidence of the correct answer and an evidence for the correct answer.

14. The question answering method of claim 13, wherein the creating of the structure of the semantic frame comprises:

dividing the natural language question in units of morphemes, and assigning a morpheme tag to each of the morphemes;

recognizing an entity of the natural language question based on a result of a morpheme analysis of the natural language question, and assigning a type to the entity;

assigning a dependency relation between word-phrases by analyzing a dependency syntax structure of the natural language question based on a result of the morpheme analysis of the natural language question; and

determining a semantic role of an argument for a predicate of the natural language question based on results of the morpheme analysis and a syntax analysis of the natural language question.

15. The question answering method of claim 13, wherein the generating of the structure of the semantic frame comprises creating the structure of a semantic frame having dependents, including a subject, an object and an adverb with respect to a predicate of the natural language question, as arguments based on results of a morpheme analysis, a syntax analysis and a semantic role analysis of the natural language question.

16. The question answering method of claim 15, wherein the creating of the structure of the semantic frame comprises:

sequentially expressing dependents including a subject, an object and an adverb of a certain predicate in the structure of the semantic frame as arguments; and

assigning a morpheme analysis result and a named entity analysis result to each argument, and assigns a syntax relation and a semantic role between the certain predicate and each argument.

17. The question answering method of claim 13, wherein the creating of the question-intermediate expression of the lexicon level comprises recognizing a question focus from the natural language question, and creating a lexicon level-query format based on a semantic frame including the question focus.

18. The question answering method of claim 17, wherein the creating of the question-intermediate expression of the lexicon level comprises creating the query format by only using a result of a semantic analysis of the natural language question without assuming a certain knowledgebase.

19. The question answering method of claim 17,

wherein the creating of the question-intermediate expression of the lexicon level comprises creating a question-intermediate expression including a variable by which the question focus is replaced using a certain sign and a condition which expresses restriction information about the variable;

and

the condition is expressed by a type predicate semantic frame which represents a type of a correct answer for the variable, and a lexicon predicate semantic frame that has a format obtained by replacing the question focus in the semantic frame by the variable.

20. The question answering method of claim 13, wherein the creating of the query comprises creating a query specialized in a certain knowledgebase by mapping an argument of the question-intermediate expression into a class defined in the certain knowledgebase or an instance, and mapping a predicate of the question-intermediate expression into a property defined in the certain knowledgebase.