Semantically-driven extraction of relations between named entities

• US 8,370,128 B2
• Filed: 09/30/2008
• Issued: 02/05/2013
• Est. Priority Date: 09/30/2008
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Brun, Caroline, Hagege, Caroline
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): Borsetti, Greg
• Application Number:
  US12/241,347
• Publication Number:
  US 20100082331A1
• Time in Patent Office: 1,589 Days
• Field of Search: 704 9- 10
• US Class Current: 704/9
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A system and method of developing rules for text processing enable retrieval of instances of named entities in a predetermined semantic relation (such as the DATE and PLACE of an EVENT) by extracting patterns from text strings in which attested examples of named entities satisfying the semantic relation occur. The patterns are generalized to form rules which can be added to the existing rules of a syntactic parser and subsequently applied to text to find candidate instances of other named entities in the predetermined semantic relation.

Claims

1. A method of developing rules for text processing comprising:

identifying a semantic relation involving at least two named entities, each named entity being represented in the semantic relation by its class;

providing a set of example instances, each example instance consisting of a combination of named entities which satisfy the semantic relation;

retrieving text strings from text, each of the retrieved strings including at least two of the named entities of one of the example instances in the set, the text strings being retrieved regardless of any lexico-syntactic relationship involving the named entities;

extracting, from the retrieved text strings, syntactic patterns, each syntactic pattern involving at least two of the named entities of one of the example instances in the respective retrieved text string;

generating generalized rules for text processing based on the extracted syntactic patterns, the rules being of a form for identifying candidate instances of the semantic relation in text; and

wherein the retrieving of text strings, extracting of syntactic patterns, and generating of rules are computer implemented.

2. The method of claim 1, wherein the extracting comprises parsing the retrieved text strings with a syntactic parser.

3. The method of claim 1, further comprising adding the generated rules to existing rules of a syntactic parser.

4. The method of claim 1, further comprising applying the generated rules to text to identify new example instances of named entities and repeating the retrieving, extracting, and generating steps using the new example instances.

5. The method of claim 1, wherein the classes are selected from the group consisting of PERSON, EVENT, DATE, ORGANIZATION, PLACE, ARTIFACT, and MONETARY EXPRESSION.

6. The method of claim 1, wherein the at least two named entities represented in the semantic relation are each represented by a different class.

7. The method of claim 6, wherein the classes are selected from DATE, PLACE, and EVENT and the semantic relation is selected from the group consisting of DATE-of-EVENT, PLACE-of-EVENT, DATE-and-PLACE-of EVENT.

8. The method of claim 1, wherein the providing of the set of example instances of named entities which satisfy the semantic relation include retrieving attested instances from a knowledge base.

9. The method of claim 1, wherein the extracting of each syntactic pattern includes identifying a set of dependencies, each of the dependencies involving at least one of the named entities of one of the example instances in the respective retrieved text string and combining at least some of the dependencies to form a pattern.

10. The method of claim 9, wherein the generating of the generalized rules includes replacing the named entity in each dependency with its corresponding class and optionally further includes normalizing other words in the dependencies.

11. The method of claim 1, wherein the generating of the generalized rules includes generating rules in the language of a syntactic parser configured for processing text.

12. The method of claim 1, further comprising applying the generated rules to text and labeling the text based on candidate instances of the semantic relation identified in the text.

13. The method of claim 1, further comprising using the new rules in information extraction to retrieve text which includes an instance of the semantic relation in response to an input of a selected named entity which is in one of the classes.

14. The method of claim 1, wherein the providing a set of example instances of combinations of named entities which satisfy the semantic relation comprises providing a set of example instances of a tuple consisting of the named entity classes, each example instance consisting only of an instance of each of the named entity classes in the tuple.

15. The method of claim 1, wherein the retrieving text strings from text, includes searching a corpus of text for instances of the example instances and retrieving all text strings from the corpus as an instance when each of the named entities of an example is present.

16. A computer program product comprising a non-transitory storage medium encoding instructions, which when executed on a computer provided with the example instances, causes the computer to perform the method of claim 1.

17. A system for developing rules for text processing comprising:

memory which stores instructions for performing the method of claim 1; and

a processor in communication with the memory which executes the instructions.

18. A system for developing rules for text processing comprising:

memory for storing a set of example instances of combinations of named entities which satisfy a semantic relation involving at least two named entities, wherein each named entity is represented in the semantic relation by its class, each of the example instances consisting only of a named entity instance of each of the named entity classes in the combination;

a text retriever for retrieving text strings from text by searching the text with the example instances, each of the retrieved strings including at least two of the named entities of one of the example instances in the set, the text strings being retrieved regardless of any lexico-syntactic relationship involving the named entities;

a pattern extractor which extracts, from the retrieved text strings, syntactic patterns, each syntactic pattern involving at least two of the named entities of one of the example instances in the respective retrieved text string; and

a rules generator which generates generalized rules for text processing based on the extracted syntactic patterns.

19. The system of claim 18, wherein the rules generator is configured for adding the generalized rules to the parser'"'"'s existing rules.

20. A method of identifying candidate instances of at least one of a date of an event and a place of an event, comprising:

identifying attested instances of combinations of named entities which are the date and place of an event, each of the instances consisting only of a DATE, PLACE, and EVENT tuple;

searching for the attested instances and retrieving text strings from text, each of the retrieved strings including at least two of the named entities of one of the attested instances, the text strings being retrieved regardless of any lexico-syntactic relationship involving the named entities;

extracting, from the retrieved text strings, syntactic patterns, each syntactic pattern involving at least two of the named entities of one of the attested instances in the respective retrieved text string;

with a processor, generating generalized rules for text processing based on the extracted syntactic patterns; and

applying the rules to text to identifying candidate instances of at least one of the date and the place of an event.