Semantic concept scorer based on an ensemble of language translation models for question answer system

Patent metadata

Field	Value
Publication number	US-2020327198-A1
Application number	US-201916379593-A
Country	US
Kind code	A1
Filing date	Apr 9, 2019
Priority date	Apr 9, 2019
Publication date	Oct 15, 2020
Grant date	—

Abstract

A system includes a memory having instructions stored therein. The system also includes at least one processor in communication with the memory. The at least one processor is configured to execute the instructions to receive an input question in a source language, generate a first similarity score based at least in part on a source-language operand, generate a target-language operand based at least in part on the source-language operand, generate a second similarity score based at least in part on the target-language operand, generate a semantic concept score based at least in part on the first similarity score and the second similarity score, generate a set of ranked answers to the input question (“ranked answer set”) based at least in part on the semantic concept score, and output the ranked answer set.

First claim

What is claimed is:

1 . A method, comprising:
receiving an input question in a source language; generating a first similarity score based at least in part on a source-language operand; generating a target-language operand based at least in part on the source-language operand; generating a second similarity score based at least in part on the target-language operand; generating a semantic concept score based at least in part on the first similarity score and the second similarity score; generating a set of ranked answers to the input question (“ranked answer set”) based at least in part on the semantic concept score; and outputting the ranked answer set.

2 . The method of claim 1 , wherein generating the semantic concept score includes using a machine learning model.

3 . The method of claim 2 , wherein generating the first similarity score includes:
generating a first constituent similarity score based at least in part on the input question and a corresponding source-language candidate-answers set, generating a second constituent similarity score based at least in part on the source-language candidate-answers set and a corresponding source-language evidence set, and generating a third constituent similarity score based at least in part on the source-language candidate-answers set and a corresponding source-language concepts-triples set; and wherein generating the second similarity score includes: generating a fourth constituent similarity score based at least in part on a target-language input question and a corresponding target-language candidate-answers set, generating a fifth constituent similarity score based at least in part on the target-language candidate-answers set and a corresponding target-language evidence set, and generating a sixth constituent similarity score based at least in part on the target-language candidate-answers set and a corresponding target-language concepts-triples set.

4 . The method of claim 3 , wherein the source-language concepts-triples set includes a source-language subject-verb-object triple, and wherein the target-language concepts-triples set includes a corresponding target-language subject-verb-object triple.

5 . The method of claim 4 , further comprising generating the source-language concepts-triples set based at least in part on a question domain.

6 . The method of claim 5 , further comprising determining a target language for the target-language operand based at least in part on the question domain.

7 . The method of claim 6 , further comprising using at least one structure mapping engine to perform at least one step selected from the group consisting of:
generating the source-language concepts-triples set and determining the target language.

8 . The method of claim 7 , wherein generating the first similarity score includes:
applying an adjustment factor based at least in part on the question domain to the first similarity score, and generating the second similarity score includes applying the adjustment factor to the second similarity score.

9 . A system, comprising:
a memory having instructions stored therein; and at least one processor in communication with the memory, wherein the at least one processor is configured to execute the instructions to:
receive an input question in a source language;
generate a first similarity score based at least in part on a source-language operand;
generate a target-language operand based at least in part on the source-language operand;
generate a second similarity score based at least in part on the target-language operand;
generate a semantic concept score based at least in part on the first similarity score and the second similarity score;
generate a set of ranked answers to the input question (“ranked answer set”) based at least in part on the semantic concept score; and
output the ranked answer set.

10 . The system of claim 9 , wherein the at least one processor is further configured to execute the instructions to:
generate a first constituent similarity score based at least in part on the input question and a corresponding source-language candidate-answers set, generate a second constituent similarity score based at least in part on the source-language candidate-answers set and a corresponding source-language evidence set, generate a third constituent similarity score based at least in part on the source-language candidate-answers set and a corresponding source-language concepts-triples set, generate the first similarity score based at least in part on the first constituent similarity score, the second constituent similarity score, and the third constituent similarity score, generate a fourth constituent similarity score based at least in part on a target-language input question and a corresponding target-language candidate-answers set, generate a fifth constituent similarity score based at least in part on the target-language candidate-answers set and a corresponding target-language evidence set, generate a sixth constituent similarity score based at least in part on the target-language candidate-answers set and a corresponding target-language concepts-triples set, and generate the second similarity score based at least in part on the fourth constituent similarity score, the fifth constituent similarity score, and the sixth constituent similarity score.

11 . The system of claim 10 , wherein the source-language concepts-triples set includes a source-language subject-verb-object triple, and wherein the target-language concepts-triples set includes a corresponding target-language subject-verb-object triple.

12 . The system of claim 11 , wherein the at least one processor is further configured to execute the instructions to generate the source-language concepts-triples set based at least in part on a question domain.

13 . The system of claim 12 , wherein the at least one processor is further configured to execute the instructions to determine a target language for the target-language operand based at least in part on the question domain.

14 . The system of claim 13 , wherein the at least one processor is further configured to execute the instructions to:
apply an adjustment factor based at least in part on the question domain to the first similarity score, and apply the adjustment factor to the second similarity score.

15 . A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by at least one processor to cause the at least one processor to:
receive an input question in a source language; generate a first similarity score based at least in part on a source-language operand; generate a target-language operand based at least in part on the source-language operand; generate a second similarity score based at least in part on the target-language operand; generate a semantic concept score based at least in part on the first similarity score and the second similarity score; generate a set of ranked answers to the input question (“ranked answer set”) based at least in part on the semantic concept score; and output the ranked answer set.

16 . The computer program product of claim 15 , wherein the program instructions are further executable by the at least one processor to cause the at least one processor to:
generate a first constituent similarity score based at least in part on the input question and a corresponding source-language candidate-answers set; generate a second constituent similarity s

Assignees

• IBM

Inventors

• Ding Chengmin
• Filoti Octavian F
• Vernier Stanley J
• Decker Renee F
• Shek Elinna

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G06F40/40

  Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30) · CPC title

• G06F40/35

  Discourse or dialogue representation · CPC title