System and method for providing question and answers with deferred type evaluation

• US 8,332,394 B2
• Filed: 05/23/2008
• Issued: 12/11/2012
• Est. Priority Date: 05/23/2008
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Fan, James, Ferrucci, David, Gondek, David C., Zadrozny, Wlodek W.
• Primary Examiner(s): Nguyen, Loan T
• Application Number:
  US12/126,642
• Publication Number:
  US 20090292687A1
• Time in Patent Office: 1,663 Days
• Field of Search: None
• US Class Current: 707/723
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A system, method and computer program product for conducting questions and answers with deferred type evaluation based on any corpus of data. The method includes processing a query including waiting until a “Type” is determined AND a candidate answer is provided; the Type is not required as part of a predetermined ontology but is a lexical/grammatical item. Then, a search is conducted for evidence that the candidate answer has the required LAT (e.g., as determined by a matching function leveraging a parser, a semantic interpreter and/or a pattern matcher). Alternately, the LAT may be matched to a known Ontological Type and then a candidate answer searched in an knowledge-base or database determined by that type. Then, all the evidence from all the different ways to determine that the candidate answer has the expected lexical answer type (LAT) is combined and one or more answers provided.

Claims

1. A computer-implemented method of generating answers to questions based on a corpus of data comprising the steps of:

receiving an input query;

computing lexical answer types (LATs) for the input query;

extracting candidate answers from the corpus;

matching candidate answers to instances in a knowledge resource;

matching instances to instance types in the knowledge resource;

matching LATs to instance types in the knowledge resource;

producing a type coercion score based on the matching steps, wherein the type coercion score reflects a degree that a candidate answer is coerced to a lexical answer type in order to eliminate the need for the lexical answer type to be part of a predetermined ontology; and

returning a candidate answer based on the type coercion score.

2. The computer-implemented method of claim 1, wherein the input query comprises a string, a string with context, or a string with context wherein the context includes another string or data structure.

3. The computer-implemented method of claim 1, wherein said computing LATs includes implementing a detection rule for detecting said LAT in a question or a candidate answer.

4. The computer-implemented method of claim 1, wherein said computing LATs includes implementing a parser and/or semantic interpreter.

5. The computer-implemented method of claim 1, wherein the instance types and LATs comprise a lexical string, the matching is performed by using string matching.

6. The computer-implemented method of claim 1, wherein the instance types and LATs comprise a lexical string or a set of strings, the matching including checking for one or more of synonym or hyponym relation between the LAT and an instance type.

7. The computer-implemented method as claimed in claim 1, further comprising:

providing a previously obtained candidate answer ranking function operating on a collection of correctly scored examples by applying machine learning technique to a corpus of scored matching LAT type pairs.

8. The computer-implemented method of claim 1, wherein the step of returning the candidate answer is based on additional scores in addition to the type coercion score.

9. A system for generating answers to questions based on a corpus of data comprising:

a memory;

a processor device in communication with the memory that performs a method comprising;

receiving an input query;

receiving an input query;

computing lexical answer types (LATs) for the input query;

extracting candidate answers from the corpus;

matching candidate answers to instances in a knowledge resource;

matching instances to instance types in the knowledge resource;

matching LATs to instance types in the knowledge resource;

producing a type coercion score based on the matching steps, wherein the type coercion score reflects a degree that a candidate answer is coerced to a lexical answer type in order to eliminate the need for the lexical answer type to be part of a predetermined ontology; and

returning a candidate answer based on the type coercion score.

10. The system as claimed in claim 9, wherein the instance types and LATs comprise a lexical string, the matching is performed by string matching.

11. The system as claimed in claim 9, wherein the instance types and LATs comprise a lexical string or a set of strings, said matching including checking for one of synonymy or hyponymy relation between the LAT and an instance type.

12. The system as claimed in claim 9, further comprising:

applying, by said processor device, a machine learning technique to a corpus of scored matching LAT and instance type pairs to provide a candidate answer ranking function.

13. The system of claim 9, wherein the step of returning the candidate answer is based on additional scores in addition to the type coercion score.

14. A non-transitory machine medium readable by a machine, tangibly embodying a program of instructions executable by the machine to perform method steps for generating

answers to questions based on a corpus of data, said method steps including the steps of:

receiving an input query;

computing lexical answer types (LATs) for the input query;

extracting candidate answers from the corpus;

matching candidate answers to instances in a knowledge resource;

matching instances to instance types in the knowledge resource;

matching LATs to instance types in the knowledge resource;

producing a type coercion score based on the matching steps, wherein the type coercion score reflects a degree that a candidate answer is coerced to a lexical answer type in order to eliminate the need for the lexical answer type to be part of a predetermined ontology; and

returning a candidate answer based on the type coercion score.

15. The non-transitory machine medium readable by a machine as claimed in claim 14, wherein said computing a LAT includes one of:

implementing a detection rule for detecting said LAT in a question or a candidate answer, or implementing a parser and/or semantic interpreter.

16. The non-transitory machine medium readable by a machine of claim 14, wherein the instance types and LATs comprise a lexical string, the matching is performed by using string matching.

17. The non-transitory machine medium readable by a machine of claim 14, wherein the instance types and LATs comprise a lexical string or a set of strings, the matching including checking for one or more of synonym or hyponym relation between the LAT and a type.

18. The non-transitory machine medium readable by a machine as claimed in claim 14, wherein said returning the candidate answer is based on additional scores in addition to the type coercion score.