Natural language processing utilizing propagation of knowledge through logical parse tree structures

Patent metadata

Field	Value
Publication number	US-9588961-B2
Application number	US-201414506810-A
Country	US
Kind code	B2
Filing date	Oct 6, 2014
Priority date	Oct 6, 2014
Publication date	Mar 7, 2017
Grant date	Mar 7, 2017

Abstract

Mechanisms are provided for processing logical relationships in natural language content. A logical parse of a first parse of a natural language content is generated by identifying latent logical operators within the first parse indicative of logical relationships between elements of the natural language content. The logical parse comprises nodes and edges linking nodes. At least one knowledge value is associated with each node in the logical parse. The at least one knowledge value of at least a subset of the nodes in the logical parse is propagated to one or more other nodes in the logical parse based on propagation rules. A reasoning operation is performed on the logical parse to generate a knowledge output indicative of knowledge associated with one or more of the logical relationships between elements of the natural language content.

First claim

What is claimed is:

1. A method, in a data processing system comprising at least one processor and at least one memory, the at least one memory comprising instructions executed by the at least one processor that configure the at least one processor to implement a natural language processing system that performs natural language processing on natural language content at least by processing logical relationships in the natural language content, the method comprising:
generating, by the natural language processing system of the data processing system, a logical parse hierarchical representation of a first parse of a natural language content by identifying latent logical operators within the first parse indicative of logical relationships between elements of the natural language content, wherein the logical parse hierarchical representation comprises nodes and edges linking nodes;
associating, by the natural language processing system, at least one knowledge value with each node in the logical parse hierarchical representation;
propagating, by the natural language processing system, the at least one knowledge value of at least a subset of the nodes in the logical parse hierarchical representation to one or more other nodes in the logical parse hierarchical representation based on propagation rules; and
performing, by the natural language processing system, a reasoning operation on the logical parse hierarchical representation to generate a knowledge output indicative of knowledge associated with one or more of the logical relationships between elements of the natural language content.

2. The method of claim 1 , wherein the at least one knowledge value associated with each node in the logical parse hierarchical representation comprises one or more supporting evidence measure knowledge values and a relevance knowledge value.

3. The method of claim 2 , wherein the one or more supporting evidence measure knowledge values comprises a fuzzy logic truth supporting evidence value indicative of an amount of supporting evidence that an atomic logic term of a corresponding node is true/met, and a fuzzy logic falsity supporting evidence value indicative of an amount of supporting evidence that the atomic logic term is false/not met.

4. The method of claim 3 , wherein the relevance knowledge value is a fuzzy logic value indicative of a measure of relevancy of the atomic logic term to reasoning operations operating on the natural language content.

5. The method of claim 3 , wherein the fuzzy logic truth supporting evidence value and fuzzy logic falsity supporting evidence value for each node in the hierarchical representation are set according to results of one or more queries applied to a content of an external source to identify evidence in support of, or against, the atomic logical term associated with the node being true/met.

6. The method of claim 3 , wherein propagating the at least one knowledge value of at least a subset of the nodes in the logical parse hierarchical representation to one or more other nodes in the logical parse hierarchical representation based on propagation rules comprises, for an OR logical operator node in the hierarchical representation:
propagating downward, in the hierarchical representation, a falsity supporting evidence value of the at least one knowledge value of a parent node to child nodes of the OR logical operator node and not propagating a truth supporting evidence value of the parent node to the child nodes of the OR logical operator;
propagating upward, in the hierarchical representation, a maximum truth supporting evidence value and a minimum falsity supporting evidence value, of the child nodes of the OR logical operator, to the parent node; and
propagating sideways, to sibling nodes in the hierarchical representation, a product of a source node's falsity supporting evidence value and the parent node's truth supporting evidence value as a truth supporting evidence value for the sibling node.

7. The method of claim 3 , wherein propagating the at least one knowledge value of at least a subset of the nodes in the logical parse hierarchical representation to one or more other nodes in the logical parse hierarchical representation based on propagation rules comprises, for an AND logical operator node in the hierarchical representation:
propagating downward, in the hierarchical representation, a truth supporting evidence value of the at least one knowledge value of a parent node to child nodes of the AND logical operator node and not propagating a falsity supporting evidence value of the parent node to the child nodes of the AND logical operator;
propagating upward, in the hierarchical representation, a minimum truth supporting evidence value and a maximum falsity supporting evidence value, of the child nodes of the AND logical operator, to the parent node; and
propagating sideways, to sibling nodes in the hierarchical representation, a product of a source node's truth supporting evidence value and the parent node's falsity supporting evidence value as a falsity supporting evidence value for the sibling node.

8. The method of claim 3 , wherein propagating the at least one knowledge value of at least a subset of the nodes in the logical parse hierarchical representation to one or more other nodes in the logical parse hierarchical representation based on propagation rules comprises, for a NOT logical operator node in the hierarchical representation:
propagating downward and upwards, in the hierarchical representation, a truth supporting evidence value and a falsity supporting evidence value of the at least one knowledge value of nodes associated with the NOT logical operator node by swapping the truth supporting evidence value and falsity supporting evidence value of parent nodes with child nodes; and
preventing sideways propagation of knowledge values of nodes associated with the NOT logical operator.

9. The method of claim 1 , wherein the at least one knowledge value for a subset of the nodes is set at least by:
associating at least one domain specific annotation, from a predetermined set of domain specific annotations, with the subset of the nodes;
performing a search of a corpus of content based on the at least one domain specific annotation and content of subset of the nodes to identify supporting content in the corpus of content;
setting at least one support knowledge value associated with the at least one domain specific annotation based on results of the search of the corpus;
determining whether there is at least one intersection of the at least one domain specific annotation with atomic logical terms of the subset of the nodes;
setting a relevance knowledge value associated with the at least one domain specific annotation based on results of determining whether there is at least one intersection; and
transferring the at least one support knowledge value and relevance knowledge value to the subset of the nodes.

10. The method of claim 1 , wherein propagating the at least one knowledge value of at least a subset of the nodes in the logical parse hierarchical representation to one or more other nodes in the logical parse hierarchical representation based on propagation rules comprises applying different propagation rules for different logical operator nodes in the logical parse hierarchical representation to propagate the at least one knowledge value upwards, downwards, and sideways in the hierarchical representation.

11. A computer program product comprising a non-transitory computer readable medium having a computer readable program stored therein, wherein the computer readable program, when executed in a data processing system, configures the data processing sys

Assignees

• IBM

Inventors

• Bruno Nicholas V
• Nielsen Robert L
• Sakhitab Saeid
• Smythe Jared M D
• Woods Eric

Classifications

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06F17/2785 Primary

  Physics · mapped topic