Ontology driven dictionary generation and ambiguity resolution for natural language processing

• US 10,268,673 B2
• Filed: 12/01/2017
• Issued: 04/23/2019
• Est. Priority Date: 06/12/2012
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Bostick, James E., Ganci, Jr., John M., Kaemmerer, John P., Trim, Craig M.
• Primary Examiner(s): Lerner, Martin
• Application Number:
  US15/829,810
• Publication Number:
  US 20180095948A1
• Time in Patent Office: 508 Days
• Field of Search: 704 1, 704 9, 704 10
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri
• CPC Class Codes: G06F 40/169 Annotation, e.g. comment data or footnotes
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Natural language processing ambiguity resolution includes storing an ontological library specifying a set of grammatical rules associated with ontological classes and sub-classes. A dictionary for each class and sub-class is generated by compiling all word instances belonging to the class or sub-class. A phrase is received having at least one ambiguous word that is associated with at least one class and at least two sub-classes. The ambiguous word is annotated with the ontological classes and sub-classes to which the ambiguous word belongs and the grammatical rules associated with the ontological classes and sub-classes to which the ambiguous word belongs are retrieved from the ontological library. One or more ontological classes or sub-classes are eliminated based on the retrieved grammatical rules and a surviving ontological class and sub-class are determined to be an accurate ontological class and sub-class for the ambiguous word to resolve the natural language processing ambiguity.

Claims

1. A computer implemented method for natural language processing ambiguity resolution, comprising:

storing, in computer memory, an ontological hierarchy of classes and sub-classes, wherein each class or sub-class represents a distinct grammatical function or grammatical property;

storing, in computer memory, a set of grammatical rules, wherein each grammatical rule stores a permissible positional relation between one class or sub-class and another class or sub-class;

storing, in computer memory, a dictionary for each class and subclass comprising compiled word instances belonging to the class or sub-class;

receiving, using a computer processor, a phrase comprising at least two words that are each associated in the dictionary with at least one class and at least two sub-classes;

annotating, using the processor, the at least two words with possible classes and sub-classes to which the at least two words belong, based on the dictionary; and

until each of the at least two words is annotated with a single possible class and sub-class;

eliminating, using the processor, a possible class or sub-class for an ambiguous word of the at least two words, based on the grammatical rules and possible classes or sub-classes of one or more other words in the phrase, and

either selecting, using the processor, a next ambiguous word of the at least two words that remains annotated with more than one possible class and sub-class to be eliminated, or determining, using the processor, that each of the at least two words is annotated with a single possible class and sub-class.

2. The method of claim 1, wherein the eliminating a possible class or sub-class comprises:

retrieving, using the processor, a class or sub-class of a word adjacent to an ambiguous word;

comparing, using the processor, the class or sub-class of the word adjacent to an ambiguous word with a required class or sub-class according to a grammatical rule associated with one of the possible classes or sub-classes of the ambiguous word; and

eliminating, using the processor, the one of the possible classes or sub-classes of the ambiguous word if the class or sub-class of the adjacent word does not match a class or sub-class required by the grammatical rule.

3. The method of claim 1, wherein the at least two words comprise the entire phrase.

4. A system for natural language processing ambiguity resolution, comprising:

a computer processor; and

computer memory, the computer memory storing

an ontological hierarchy of classes and sub-classes, wherein each class or sub-class represents a distinct grammatical function or grammatical property;

a set of grammatical rules, wherein each grammatical rule stores a permissible positional relation between one class or sub-class and another class or sub-class;

a dictionary for each class and subclass comprising compiled word instances belonging to the class or sub-class; and

a series of program instructions, wherein the program instructions are executable by the computer processor to;

receive a phrase comprising at least two words that are each associated in the dictionary with at least one class and at least two sub-classes;

annotate the at least two words with possible classes and sub-classes to which the at least two words belong, based on the dictionary; and

until each of the at least two words is annotated with a single possible class and sub-class;

eliminate a possible class or sub-class for an ambiguous word of the at least two words, based on the grammatical rules and on the possible classes or sub-classes of one or more other words in the phrase, and

either select a next ambiguous word of the at least two words that remains annotated with more than one possible class and sub-class to be eliminated, or determine that each of the at least two words is annotated with a single possible class and sub-class.

5. The system of claim 4, wherein the eliminating the possible class or sub-class comprises:

retrieving a class or sub-class of a word adjacent to an ambiguous word;

comparing the class or sub-class of the word adjacent to an ambiguous word with a required class or sub-class according to a grammatical rule associated with one of the possible classes or sub-classes of the ambiguous word; and

eliminating the one of the possible classes or sub-classes of the ambiguous word if the class or sub-class of the adjacent word does not match a class or sub-class required by the grammatical rule.

6. The system of claim 4, wherein the at least two words comprise the entire phrase.

7. A computer implemented method for natural language processing ambiguity resolution, comprising:

storing, in computer memory, an ontological hierarchy of classes and sub-classes, wherein each class or sub-class represents a distinct linguistic property;

storing, in computer memory, a set of grammatical rules, wherein each grammatical rule stores a permissible positional relation between one class or sub-class and another class or sub-class;

storing, in computer memory, a dictionary for each class and subclass comprising compiled single-word or multiword instances belonging to the class or sub-class;

receiving, using a computer processor, a phrase comprising at least two words or multiword instances that are each associated in the dictionary with at least one class and at least two sub-classes;

annotating, using the processor, the at least two words or multiword instances with possible classes and sub-classes to which the at least two words or multiword instances belong, based on the dictionary; and

until each of the at least two words or multiword instances is annotated with a single possible class and sub-class;

eliminating, using the processor, a possible class or sub-class for an ambiguous word or multiword instance of the at least two words or multiword instances, based on the grammatical rules and possible classes or sub-classes of one or more other words or multiword instances in the phrase, and

either selecting, using the processor, a next ambiguous word or multiword instance of the at least two words or multiword instances that remains annotated with more than one possible class and sub-class to be eliminated, or determining, using the processor, that each of the at least two words or multiword instances is annotated with a single possible class and sub-class.

8. The method of claim 7, wherein the eliminating a possible class or sub-class comprises:

retrieving, using the processor, a class or sub-class of a word or multiword instance from the phrase adjacent to an ambiguous word or multiword instance;

comparing, using the processor, the class or sub-class of the adjacent word or multiword instance with a required class or sub-class according to a grammatical rule associated with one of the possible classes or sub-classes to which the ambiguous word or multiword instance belongs; and

eliminating, using the processor, the one of the possible classes or sub-classes to which the ambiguous word or multiword instance belongs if the class or sub-class of the adjacent word or multiword instance does not match a class or sub-class required by the grammatical rule.

9. The method of claim 7, wherein the at least two words or multiword instances comprise the entire phrase.