Recombination techniques for natural language generation

• US 10,325,026 B2
• Filed: 09/25/2015
• Issued: 06/18/2019
• Est. Priority Date: 09/25/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Cardillo, Bryan D.
• Primary Examiner(s): Opsasnick, Michael N
• Application Number:
  US14/866,066
• Publication Number:
  US 20170091170A1
• Time in Patent Office: 1,362 Days
• Field of Search: 704 4
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/247 Thesauruses; Synonyms
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06F 40/56 Natural language generation

Abstract

A technique for generating a new equivalent phrase for an input phrase includes receiving a first input phrase for natural language expansion. Tokens that correspond to parts of speech are generated for the first input phrase. An original grammar tree is generated using at least some of the tokens. One or more alternate grammar trees are generated that are different from the original grammar tree but substantially equivalent to the original grammar tree. One or more synonyms for at least one of the tokens are generated. Finally, one or more new phrases are generated based on the one or more alternate grammar trees and the one or more synonyms.

Claims

1. A method of generating one or more new equivalent phrases for an input phrase, comprising:

receiving, by a phrase generating system, a first input phrase for natural language expansion, wherein the first input phrase is a question;

generating, by the phrase generating system, tokens that correspond to parts of speech for the first input phrase;

generating, by the phrase generating system, an original grammar tree using at least some of the tokens;

generating, by the phrase generating system, one or more alternate grammar trees that are different from the original grammar tree but equivalent to the original grammar tree;

generating, by the phrase generating system, one or more synonyms for at least one of the tokens;

generating, by the phrase generating system, a first answer to the first input phrase;

generating, by the phrase generating system, one or more new phrases based on the one or more alternate grammar trees and the one or more synonyms, wherein the one or more new phrases are questions;

receiving, by the phrase generating system, the one or more new phrases as input;

generating, by the phrase generating system, one or more answers to the one or more new phrases; and

determining, by the phrase generating system, using semantic analysis whether the one or more answers are similar to the first answer generated for the first input phrase to ensure that the phrase generating system provides a similar answer to a similar question and thereby improve operation of the phrase generating system.

2. The method of claim 1, further comprising:

tagging, by the phrase generating system, the parts of speech prior to the generating the original grammar tree.

3. The method of claim 2, wherein the tagging further comprises:

indicating whether the tokens correspond to one of a noun, a verb, an adjective, an adverb, and a pronoun.

4. A computer program product for a phrase generating system, the computer program product comprising:

a computer-readable storage device; and

computer-readable program code embodied on the computer-readable storage device, wherein the computer-readable program code, when executed by a phrase generating system, causes the phrase generating system to;

receive a first input phrase for natural language expansion, wherein the first input phrase is a question;

generate tokens that correspond to parts of speech for the first input phrase;

generate an original grammar tree using at least some of the tokens;

generate one or more alternate grammar trees that are different from the original grammar tree but equivalent to the original grammar tree;

generate one or more synonyms for at least one of the tokens;

generate a first answer to the first input phrase;

generate one or more new phrases based on the one or more alternate grammar trees and the one or more synonyms, wherein the one or more new phrases are questions;

receive the one or more new phrases as input;

generate one or more answers to the one or more new phrases; and

determine using semantic analysis whether the one or more answers are similar to the first answer generated for the first input phrase to ensure that the phrase generating system provides a similar answer to a similar question and thereby improve operation of the phrase generating system.

5. The computer program product of claim 4, wherein the computer-readable program code, when executed by the phrase generating system, further configures the phrase generating system to:

tag the parts of speech prior to the generating the original grammar tree.

6. The computer program product of claim 5, wherein the computer-readable program code, when executed by the phrase generating system, further configures the phrase generating system to:

indicate whether the tokens correspond to one of a noun, a verb, an adjective, an adverb, and a pronoun.

7. A phrase generating system, comprising:

a cache memory; and

a processor coupled to the cache memory, wherein the processor is configured to;

receive a first input phrase for natural language expansion, wherein the first input phrase is a question;

generate tokens that correspond to parts of speech for the first input phrase;

generate an original grammar tree using at least some of the tokens;

generate one or more alternate grammar trees that are different from the original grammar tree but equivalent to the original grammar tree;

generate one or more synonyms for at least one of the tokens;

generate a first answer to the first input phrase;

generate one or more new phrases based on the one or more alternate grammar trees and the one or more synonyms, wherein the one or more new phrases are questions;

receive the one or more new phrases as input;

generate one or more answers to the one or more new phrases; and

determine using semantic analysis whether the one or more answers are similar to the first answer generated for the first input phrase to ensure that the phrase generating system provides a similar answer to a similar question and thereby improve operation of the phrase generating system.

8. The phrase generating system of claim 7, wherein the processor is further configured to:

tag the parts of speech prior to the generating the original grammar tree.

9. The phrase generating system of claim 8, wherein the processor is further configured to:

indicate whether the tokens correspond to one of a noun, a verb, an adjective, an adverb, and a pronoun.