Question-answering system and method based on semantic labeling of text documents and user questions

• US 8,666,730 B2
• Filed: 03/12/2010
• Issued: 03/04/2014
• Est. Priority Date: 03/13/2009
• Status: Active Grant

• Current Assignee: Allium US Holding LLC
• Original Assignee: Invention Machine Corporation (S&P Global Inc.)
• Inventors: Todhunter, James, Sovpel, Igor, Pastanohau, Dzianis
• Primary Examiner(s): GUERRA-ERAZO, EDGAR X
• Application Number:
  US12/723,449
• Publication Number:
  US 20100235164A1
• Time in Patent Office: 1,453 Days
• Field of Search: 704 1- 10, 704/257, 704/258, 704/260, 704/270, 704/270.1, 704/275
• US Class Current: 704/9
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation

Abstract

A question-answering system for searching exact answers in text documents provided in the electronic or digital form to questions formulated by user in the natural language is based on automatic semantic labeling of text documents and user questions. The system performs semantic labeling with the help of markers in terms of basic knowledge types, their components and attributes, in terms of question types from the predefined classifier for target words, and in terms of components of possible answers. A matching procedure makes use of mentioned types of semantic labels to determine exact answers to questions and present them to the user in the form of fragments of sentences or a newly synthesized phrase in the natural language. Users can independently add new types of questions to the system classifier and develop required linguistic patterns for the system linguistic knowledge base.

Claims

1. A method for question-answering based on automatic semantic labeling of text documents and user questions, the method comprising:

electronically receiving natural language text documents using at least one computer processor coupled to at least one non-transitory storage medium;

electronically receiving a user question formulated in a natural language;

performing a basic linguistic analysis of the text documents and the user question;

performing semantic labeling of the text documents through semantic analysis, including identifying target words in the text documents using linguistic patterns stored in a database and assigning question types to the target words in the text documents;

storing the semantically labeled text documents in a labeled text documents database;

performing semantic labeling of the user question through semantic analysis, including identifying one or more target words in the user question using linguistic patterns stored in a database and assigning a question type to each of the one or more target words in the user question;

searching the labeled text documents database for text fragments relevant to the semantically labeled user question at least in part by matching a question type assigned to a target word in the labeled text documents to a question type assigned to a target word in the user question, wherein relevance is based on a ranking of the text fragments relative to the semantically labeled user question; and

synthesizing answers to the user question from the relevant text fragments, and electronically presenting the synthesized answer to the user.

2. The method according to claim 1, further comprising:

applying parts-of-speech tags to the text documents and user question to generate tagged text documents and user question;

parsing the tagged text documents and user question to generate parsed and tagged text documents and user question; and

semantically analyzing the parsed and tagged text documents and user question to generate semantically analyzed, parsed, and tagged text documents and user question.

3. The method according to claim 2, wherein applying parts-of-speech tags is performed on preformatted text documents that comprise text with non-natural language symbols removed.

4. The method according to claim 2, wherein the semantic analysis comprises:

recognizing one or more facts in the form of one or more expanded Subject-Action-Object (eSAO) sets in the text documents and user question, wherein each eSAO set has one or more eSAO components; and

recognizing rules in the text documents and user question that reflect regularities of the outside world/knowledge domain in the form of Cause-Effect relations in the eSAO sets, wherein each of the Cause-Effect relations comprises a Cause eSAO and an Effect eSAO.

5. The method according to claim 4, wherein the one or more eSAO components comprises one or more elements of a group consisting of:

subjects, objects, actions, adjectives, prepositions, indirect objects and adverbs.

6. The method according to claim 4, wherein the Cause eSAO comprises one or more eSAO components of the one or more eSAO sets and the Effect eSAO comprises one or more other eSAO components of the one or more eSAO sets.

7. The method according to claim 6, wherein the Cause-Effect relations comprise a sequential operator relating the eSAO components of the Cause eSAO to the other eSAO components of the Effect eSAO with lexical, grammatical, and/or semantic language means.

8. The method according to claim 4, wherein recognizing one or more expanded Subject-Action-Object (eSAO) sets in the text documents and user question comprises recognizing one or more subjects, objects, actions, adjectives, prepositions, indirect objects, and adverbs in one or more sentences of the text documents and user question.

9. The method according to claim 4, wherein recognizing one or more expanded Subject-Action-Object (eSAO) sets and Cause-Effect relations in the text documents and user question comprises accessing a linguistic knowledge base having a database of patterns defining eSAO and Cause-Effect components.

10. The method according to claim 2, wherein identifying target words in the text documents using linguistic patterns stored in the database and assigning question types to the target words in the text documents comprises matching the semantically analyzed, parsed, and tagged text documents against question classification based linguistic patterns in a patterns database that is a part of a linguistic knowledge base using, at least in part, the following steps:

determining part-of-speech tags, syntactic tags, and semantic labels and eSAO sets and Cause-Effect sets for text in the text documents;

generating eSAO labels and Cause-Effect labels; and

matching the eSAO labels and Cause-Effect labels to semantic labels of question types and answer components for target words.

11. The method according to claim 2, wherein identifying one or more target words in the user question using linguistic patterns stored in the database and assigning a question type to each of the one or more target words in the user question comprises matching the semantically analyzed parsed, and tagged user question against question classification based linguistic patterns in a patterns database that is a part of a linguistic knowledge base using, at least in part, the following steps:

determining part of-speech tags, syntactic tags, and semantic labels of words in the user question and determining eSAO sets and cause-effect sets from the words of the user question;

generating eSAO labels and cause-effect labels from the eSAO sets and cause-effect sets; and

matching the eSAO labels and cause-effect labels to semantic labels of question types for target words.

12. The method according to claim 1, wherein searching the labeled text documents database for the text fragments relevant to semantically labeled user questions is based on matching the semantically labeled user questions against sentences from the labeled text documents database, comprising:

matching words and semantic labels; and

building a list of sentences relevant to the user question with indication of a degree of relevance.

13. The method according to claim 1, further including ranking sentences from the labeled text documents database relevant to the user question based on a degree of relevance of each sentence to a user question, determined according to predetermined criteria for matching semantic information from each sentence with semantic information from the user question.

14. The method according to claim 1, wherein synthesizing the answers includes:

synthesizing natural language answers to a user question from relevant sentences is performed in the form of those original fragments of relevant sentences which were marked with labels of answer components on the stage of semantic labeling of text documents, and in the form of new natural language phrases, generated on the basis of linguistic patterns from the linguistic knowledge base, based on the an eSAO format.

15. A computer program product comprising a non-transitory computer-readable medium having stored therein computer-executable instructions for performing a method for question-answering based on automatic semantic labeling of text documents and user questions, the method comprising:

electronically receiving natural language text documents;

electronically receiving a user question formulated in a natural language;

performing a basic linguistic analysis of the text documents and the user question;

performing semantic labeling of the text documents through semantic analysis, including identifying target words in the text documents using linguistic templates and assigning question types to the target words,

storing the semantically labeled text documents in a labeled text documents database;

performing semantic labeling of the user question through semantic analysis, including identifying one or more target words in the user question using linguistic templates and assigning question types to the one or more target words;

searching the labeled text documents database for text fragments relevant to the semantically labeled user question, wherein relevance is based on a ranking of the text fragments relative to the semantically labeled user question; and

synthesizing answers to the user question from the relevant text fragments.

16. The computer program product of claim 15, wherein the method further comprises electronically presenting to the user the answers to his or her questions.

17. A question-answering system that uses automatic semantic labeling of text documents and user questions in electronic or digital form formulated in natural language, the system comprising:

one or more processors;

a linguistic knowledge base that includes a plurality of question-classification based linguistic patterns; and

a linguistic analyzer configured to produce linguistically analyzed text documents and user questions using the one or more processors, the linguistic analyzer comprising a semantic analyzer that includes;

an expanded Subject-Action-Object (eSAO) recognizer and a Cause-Effect recognizer configured to produce semantically analyzed text documents and user questions, including recognizing one or more facts in the form of one or more eSAO sets based on the text documents and on the user questions,

wherein eSAO and Cause-Effect recognition is based on patterns stored in the linguistic knowledge base.

18. The question-answering system according to claim 17, wherein the linguistic analyzer further comprises:

a part-of-speech tagger that receives preformatted text documents based on the text documents and the user questions in electronic or digital format; and

a parser that receives the text documents and the user questions tagged by the part-of-speech tagger and provides parsed text documents and user question to the semantic analyzer,

wherein the part-of-speech tagger and the parser operate with data stored in the linguistic knowledge base.

19. The question-answering system of claim 18, wherein the linguistic analyzer further comprises:

a preformatter that receives the text documents in electronic or digital format and produces the preformatted text documents;

a text documents labeler that matches the semantically analyzed text documents against the question classification based linguistic patterns stored in the linguistic knowledge base and generates semantic relationship labels based on the semantically analyzed text documents and the matching, whereby the semantically labeled text documents are stored in labeled text documents database;

a question labeler that matches the semantically analyzed user question against the question classification based linguistic patterns stored in the linguistic knowledge base and generates semantic relationship labels based on the semantically analyzed user questions and the matching;

a searcher that matches the semantically labeled user question against sentences from the labeled text documents database, wherein the searcher matches words and semantic labels and builds a list of sentences relevant to the user question with an indication of a degree of relevance;

an answer ranker that sorts sentences from the labeled text documents database that are relevant to the user question, the sorting in accordance with the degree of relevance of each sentence to the user question; and

a text synthesizer that generates the natural language answers to the user question from the relevant sentences and electronically presents them to the user.

20. The question-answering system according to claim 19, wherein the preformatter is configured to perform at least one of the following functions:

removal of any symbols in a digital or electronic presentation of the text documents that do not form part of natural language text;

detection and correction of any mismatches or mistakes in text documents; and

partitioning the text into structures of sentences and words.

21. The question-answering system according to claim 19, wherein the text documents labeler is configured to match the semantically analyzed text documents against linguistic patterns by matching words, part-of-speech tags, syntactic tags, eSAO and Cause-Effect sets.

22. The question-answering system according to claim 19, wherein the text documents labeler is configured to generate semantic relationship labels by generating eSAO and Cause-Effect labels and based on matching semantic labels of question types and answer components for target words.

23. The question-answering system according to claim 19, wherein the question labeler is configured to match the semantically analyzed user questions against linguistic patterns by matching words, part-of-speech tags, syntactic tags, labels of question words, eSAO and Cause-Effect sets.

24. The question-answering system according to claim 19, wherein the question labeler is configured to generate the semantic relationship labels by generating eSAO and Cause-Effect labels and based on matching semantic labels of question types for target words.

25. The question-answering system according to claim 19, wherein the text synthesizer is configured to generate natural language answers to the user questions by generating answers in the form of original fragments of relevant sentences, and in the form of new natural language phrases, generated on the basis of linguistic patterns from the linguistic knowledge base based on eSAO format.

26. The question-answering system according to claim 17, wherein the semantic analyzer is also configured to generate Cause-Effect relations from the eSAO sets, wherein each of the Cause-Effect relations comprises a Cause eSAO, an Effect eSAO, and at least one sequential operator relating the Cause eSAO to the Effect eSAO.

27. The question-answering system according to claim 26, wherein each of the eSAO sets comprises eSAO components and the Cause eSAO comprises one or more of the eSAO components and the Effect eSAO comprises one or more of the eSAO components different than the one or more eSAO components of the Cause eSAO.

28. The question-answering system according to claim 27, wherein the one or more eSAO components comprises one or more elements of a group consisting of:

subjects, objects, actions, adjectives, prepositions, indirect objects and adverbs.