Using ontological information in open domain type coercion

• US 9,508,038 B2
• Filed: 09/06/2012
• Issued: 11/29/2016
• Est. Priority Date: 09/24/2010
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Ferrucci, David A., Kalyanpur, Aditya, Murdock, James W. IV, Welty, Christopher A., Zadrozny, Wlodek W.
• Primary Examiner(s): NGUYEN, THU N
• Application Number:
  US13/605,339
• Publication Number:
  US 20120330921A1
• Time in Patent Office: 1,545 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation

Abstract

A computer-implemented system, method and program product generates answers to questions in an input query text string. The method includes determining, by a programmed processor unit, a lexical answer type (LAT) string associated with an input query; automatically obtaining a candidate answer string to the input query from a data corpus; mapping the query LAT string to a first type string in a structured resource; mapping the candidate answer string to a second type string in the structured resource; and determining if the first type string and the second type string are disjointed; and scoring the candidate answer string based on the determination of the types being disjointed wherein the structured resource includes a semantic database providing ontological content.

Claims

1. A computer-implemented method for automatically generating answers to questions comprising:

receiving an input query;

determining a lexical answer type (LAT) for the input query;

receiving a candidate answer for the input query;

matching the candidate answer to a first instance in an ontology;

matching the first instance to a first instance type in the ontology;

matching the LAT to a second instance type in the ontology;

identifying one or more logical rules in the ontology that relate the second instance type to the first instance type;

producing a score for the candidate answer based on the matching steps; and

modifying the score for the candidate answer based on the logical rule identified that relates the second instance type and the first instance type.

2. The computer-implemented method of claim 1, wherein said identifying one or more logical rules comprises:

identifying an axiom in said ontology indicating a disjoint relation between the first instance type and the second instance type.

3. The computer-implemented method of claim 1, wherein said identifying one or more logical rules comprises:

identifying an axiom in said ontology indicating one of;

an equivalence relation between the first instance type and the second instance type,

a subsumption relation between the first instance type and the second instance type, and

a sibling relation between the first instance type and the second instance type.

4. The computer-implemented method of claim 1, wherein said lexical answer type is a word concept, said ontology providing one or more senses associated with a word concept.

5. The computer-implemented method of claim 4, wherein said matching the lexical answer type to the second instance type in the ontology comprises:

accessing a resource having ontological content;

obtaining, from said resource, a ranking of senses of the obtained type or concept instance.

6. The computer-implemented method of claim 5, wherein said lexical answer type comprises a query LAT string, said matching the lexical answer type to the second instance type in the ontology comprises:

accessing a resource having ontological content;

obtaining one or more types from said resource that match the query LAT string; and

accessing a data corpus, and obtaining from said data corpus a ranked list of second instances based on said obtained one or more types therein.

7. The computer-implemented method of claim 6, wherein said matching the lexical answer type to a second instance type in the ontology further comprises:

accessing said resource having ontological content;

identifying a context in said input query;

computing a similarity measure between the identified context of said input query and contextual data associated with the obtained types from the ontological content.

8. The computer-implemented method of claim 7, wherein said computing a similarity measure comprises:

obtaining a representation of said context; and

implementing a function to compute said similarity between the identified context and contextual data based on the representation.

9. The computer-implemented method of claim 8, wherein said implementing a function comprises:

aligning the context using a BOW (bag of words) approach;

or,

applying a vector model to compute said similarity.

10. The computer-implemented method of claim 8, further comprising:

computing a confidence score based on a weighted or un-weighted combination of one or more of;

said sense rank, said ranked list of said second instances, and said computed similarity measure.

11. The computer-implemented method of claim 4, wherein said candidate answer comprises a candidate answer string, said matching the candidate answer to a first instance in the ontology comprises:

accessing a resource having ontological content;

obtaining one or more types from said resource that match said candidate answer string; and

accessing a data corpus, and obtaining from said data corpus a ranked list of first instances based on a frequency of occurrence of said one or more types therein.

12. The computer-implemented method of claim 11, wherein said matching the candidate answer to a first instance in the ontology further comprises:

accessing said resource having ontological content;

identifying a context in said input query;

computing a similarity measure between the identified context of said input query and contextual data associated with the obtained types from the ontological content.

13. The computer-implemented method of claim 12, further comprising:

computing a confidence score based on a weighted or un-weighted combination of said ranked list of first instances and said computed similarity measure.