TEXT PARAPHRASING METHOD AND PROGRAM, CONVERSION RULE COMPUTING METHOD AND PROGRAM, AND TEXT PARAPHRASING SYSTEM

• US 20100010803A1
• Filed: 12/21/2007
• Published: 01/14/2010
• Est. Priority Date: 12/22/2006
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Ishikawa, Kai, Akamine, Susumu
• Granted Patent: US 8,447,589 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/247 Thesauruses; Synonyms
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30

Abstract

A paraphrase model of a question text inputted by a user is learned, and a paraphrase expression is generated in real time. When information in text set storage unit is updated, text pair extracting unit extracts a paraphrase text pair from the text set storage unit and stores it in text pair storage unit. Model learning unit learns a question text paraphrase model from the paraphrase text pair in text pair storage unit, and stores it in model storage unit. Text pair extracting unit extracts a paraphrase text pair again from the text set storage unit by using the question text paraphrase model which the model storage unit possesses, and stores it in the text pair storage unit. In case where the stored paraphrase text pair is the same as the paraphrase text pair stored in the text pair storage unit, learning of the question text paraphrase model is ended. Candidate creating unit reads the question text paraphrase model from the model storage unit and generates a paraphrase candidate of the inputted question text.

Claims

1-17. -17. (canceled)

18. A text paraphrasing method for generating, with respect to a question text inputted by a user, a question text of different expression but the same content, said method comprising:

extracting a combination of sentences having a degree of similarity greater than or equal to a set threshold, from among question texts having the same answer texts stored in storage unit, termed “

first sentence extraction step”

;

computing a conversion rule for converting from said combination of sentences extracted in said first sentence extraction step, to a sentence of the same content but different expression, termed “

conversion rule computation step”

;

extracting again from said storage unit a combination of sentences having a degree of similarity greater than or equal to said threshold, using said conversion rule computed in said conversion rule computation step, termed “

second sentence extraction step”

;

judging whether or not all of said combinations of sentences extracted in said first and said second sentence extraction steps match, termed “

judging step”

; and

generating a paraphrase text of said inputted question text, using said conversion rule computed in said conversion rule computation step, termed “

text generation step”

;

wherein

execution of said first sentence extraction step, said conversion rule computation step, and said second sentence extraction step is repeated until it is judged that said combinations of sentences extracted in said first and said second sentence extraction steps all match.

19. The text paraphrasing method according to claim 18, further comprising displaying a generated paraphrase text to a user, wherein said generated paraphrase text is displayed together with a value indicating a degree of similarity with respect to an inputted question text.

20. The text paraphrasing method according to claim 18, wherein said text generation step comprises:

decomposing an inputted question text into a word string;

comparing said conversion rule computed in said conversion rule computation step and said word string, and selecting a plurality of word strings with high degrees of similarity, termed “

selection step” and

sorting said word strings selected in said selection step.

29. A conversion rule computing program for computing, with respect to a question text inputted by a user, a conversion rule for generating a question text of different expression but the same content, said program comprising executing, on a computer:

processings of the steps of the text paraphrasing method according to claim 18.

30. The conversion rule computing program according to claim 29, for executing on a computer, with respect to a storage unit that stores a question text inputted by an input unit, an answer text with respect to said question text, and information indicating a correspondence relationship of said question text and said answer text;

a process of applying said information to one group in which answer texts of the same content are collected, and generating a group of question texts corresponding to answer texts of the same content.

31. The conversion rule computing program according to claim 29, for executing on a computer:

a process of updating information stored in said storage unit; and

,

at an occasion at which said information in said storage unit is updated, a process of starting said first sentence extracting process.

21. A conversion rule computing method for computing, with respect to a question text inputted by a user, a conversion rule for generating question text(s) of different expression but the same content, said method comprising:

extracting a combination of sentences having a degree of similarity greater than or equal to a set threshold, from among question texts having the same answer texts stored in a storage unit, termed “

first sentence extraction step”

;

computing a conversion rule for converting from said combination of sentences extracted in said first sentence extraction step to a sentence of the same content but different expression, termed “

conversion rule computation step”

;

extracting again from said storage unit a combination of sentences having a degree of similarity greater than or equal to said threshold, using said conversion rule computed in said conversion rule computation step, termed “

second sentence extraction step”

; and

judging whether or not all of said combinations of sentences extracted in said first and said second sentence extraction steps match, termed “

judging step”

;

wherein

execution of said first sentence extraction step, said conversion rule computation step, and said second sentence extraction step is repeated until it is judged that said combinations of sentences extracted in said first and said second sentence extraction steps all match.

22. The conversion rule computing method according to claim 21, wherein

a question text inputted by input unit, an answer text with respect to said question text, and information indicating a correspondence relationship of said question text and said answer text are stored in a storage unit,

said information is applied with respect to one group in which answer texts of the same content are collected, and

a group of question texts corresponding to answer texts of the same content is generated.

23. The conversion rule computing method according to claim 21, comprising updating information stored in said storage unit;

wherein

at an occasion at which said information in said storage unit is updated, processing is started in said first sentence extraction step.

32. A text paraphrasing program for generating, with respect to a question text inputted by a user, a question text of different expression but the same content, said program executing on a computer:

processings of the steps of the conversion rule computing method according to claim 21.

33. The text paraphrasing program according to claim 32 for executing, on a computer, a process of displaying a generated paraphrase text together with a value indicating said degree of similarity with respect to an inputted question text.

34. The text paraphrasing program according to claim 32 for executing, on a computer, in said text generating process:

a process of decomposing said inputted question text into a word string;

a selecting process of comparing said conversion rule computed in said conversion rule computing process and said word string, and selecting a plurality of word strings with high degrees of similarity; and

a sorting process of sorting said word strings selected in said selecting process.

24. A text paraphrasing system that generates, with respect to a question text inputted by a user, a question text of different expression but the same content, said system comprising:

an input unit inputting a question text;

a storage unit storing said question text inputted to said input unit and an answer text with respect to said question text;

a sentence extracting unit extracting a combination of sentences having a degree of similarity greater than or equal to a set threshold, from among question texts having the same answer texts stored in said storage unit;

a conversion rule computing unit computing a conversion rule for converting from said combination of sentences extracted in said sentence extracting unit, to a sentence of the same content but different expression; and

a text generating unit generating a paraphrase text of said question text inputted to said input unit by said user, using said conversion rule computed by said conversion rule computing unit.

25. The text paraphrasing system according to claim 24, wherein said storage unit comprises:

a first storage unit storing a question text inputted to said input unit and an answer text with respect to said question text;

a second storage unit for storing a combination of sentences extracted by said sentence extracting unit; and

a third storage unit storing a conversion rule computed by said conversion rule computing unit.

26. The text paraphrasing system according to claim 24, wherein said conversion rule comprises at least a translation model, a language model, and a word string.

27. The text paraphrasing system according to claim 24, comprising:

a display unit displaying said paraphrase text generated by said text generating unit;

wherein

said display unit displays said paraphrase text generated by said text generating unit together with a value indicating a degree of similarity with a question text inputted to said input unit by said user.

28. The text paraphrasing system according to claim 24, comprising:

an information updating unit updating information in said storage unit, according to a command inputted to said input unit;

wherein

said information updating unit transmits a signal for starting extraction of a combination of sentences, when said information in said storage unit is updated, to said sentence extracting unit.