Automatic answer rephrasing based on talking style

Patent metadata

Field	Value
Publication number	US-11030990-B2
Application number	US-201916561929-A
Country	US
Kind code	B2
Filing date	Sep 5, 2019
Priority date	Oct 17, 2017
Publication date	Jun 8, 2021
Grant date	Jun 8, 2021

Abstract

Techniques are described that facilitate automatically providing entities with rephrased versions of standard answers. In one embodiment, a computer-implemented is provided that comprises determining, by a device operatively coupled to a processor, a talking style of a plurality of talking styles that an entity is associated with based on reception of natural language input from the entity proposing a question related to a defined topic. The method further comprises selecting, by the device based on the talking style, an answer rephrasing model from a plurality of answer rephrasing models respectively configured to generate different rephrased versions of a standard answer to the question, and employing, by the device, the answer rephrasing model to generate a rephrased version of the standard that corresponds to the talking style.

First claim

What is claimed is:

1. A system, comprising:
a memory that stores computer executable components;
a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise:
a classification component that determines a talking style of a plurality of talking styles that an entity is associated with based on reception of input from the entity proposing a question related to a defined topic; and
a model selection component that selects, based on the talking style, an answer rephrasing model from a plurality of answer rephrasing models respectively configured to generate different rephrased versions of a standard answer to the question, wherein respective answer rephrasing models of the plurality of answer rephrasing models employ a sequence-to-sequence type of recurrent neural network architecture.

2. The system of claim 1 , wherein the computer executable components further comprise:
a clustering component that determines the plurality of talking styles based on analysis of parallel corpus question and answer data comprising different versions of questions related to the defined topic that respectively correspond to same standard answers.

3. The system of claim 1 , wherein different answer rephrasing models of the plurality of answer rephrasing models are tailored to different talking styles of the plurality of talking styles.

4. The system of claim 3 , wherein the different answer rephrasing models respectively employ a neural network structure and wherein the computer executable components further comprise:
a modeling component that generates the different answer rephrasing models by training the different answer rephrasing models using parallel corpus question and answer data that comprises different versions of same standard answers and respectively associates the different versions with the different talking styles.

5. The system of claim 1 , wherein the respective answer rephrasing models employ a copy function and a restricted generative function to determine the different rephrased versions of the standard answer to the question.

6. The system of claim 1 , wherein the respective answer rephrasing models determine the different rephrased versions of the standard answer to the question based on encoding the standard answer and the question, thereby resulting in an encoded answer and an encoded question.

7. The system of claim 6 , wherein the respective answer rephrasing models determine the different rephrased versions of the standard answer to the question using either a copy function or a restricted generative function to decode respective parts of the encoded answer and the encoded question.

8. The system of claim 7 , wherein the respective answer rephrasing models comprise a selection component that determines whether to apply the copy function or the restricted generative function to decode the respective parts of the encoded answer and the encoded question.

9. The system of claim 7 , wherein the copy function selects a term or phrase from the standard answer.

10. The system of claim 7 , wherein the restricted generative function selects an alternative term or phrase from an alignment table data that correlates one or more alternative terms or phrases with standard terms or phrases of the standard answer.

11. The system of claim 1 , wherein the defined topic comprises health and medicine.

12. A computer implemented method, comprising:
determining, by a device operatively coupled to a processor, a talking style of a plurality of talking styles that an entity is associated with based on reception of input from the entity proposing a question related to a defined topic; and
selecting, by the device based on the talking style, an answer rephrasing model from a plurality of answer rephrasing models respectively configured to generate different rephrased versions of a standard answer to the question, wherein respective answer rephrasing models of the plurality of answer rephrasing models employ a sequence-to-sequence type of recurrent neural network architecture.

13. The computer implemented method of claim 12 , further comprising:
determining, by the device, the plurality of talking styles based on analysis of parallel corpus question and answer data comprising different versions of questions related to the defined topic that respectively correspond to same standard answers.

14. The computer implemented method of claim 12 , wherein different answer rephrasing models of the plurality of answer rephrasing models are tailored to different talking styles of the plurality of talking styles.

15. The computer implemented method of claim 14 , wherein the different answer rephrasing models respectively employ a neural network structure and wherein the method further comprises:
a modeling component that generates the different answer rephrasing models by training the different answer rephrasing models using parallel corpus question and answer data that comprises different versions of same standard answers and respectively associates the different versions with the different talking styles.

16. The computer implemented method of claim 12 , wherein the respective answer rephrasing models employ a copy function and a restricted generative function to determine the different rephrased versions of the standard answer to the question.

17. A computer program product for automatically generating rephrased versions of answers to questions, wherein the rephrased versions are tailored to different talking styles, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processing component to cause the processing component to:
determine a talking style of a plurality of talking styles that an entity is associated with based on reception of input from the entity proposing a question related to a defined topic; and
select, based on the talking style, an answer rephrasing model from a plurality of answer rephrasing models respectively configured to generate different rephrased versions of a standard answer to the question, wherein respective answer rephrasing models of the plurality of answer rephrasing models employ a sequence-to-sequence type of recurrent neural network architecture.

18. The computer program product of claim 17 , wherein a restricted generative function to determine the different rephrased versions of the standard answer to the question.

Assignees

• IBM

Inventors

• Jiang Jian Min
• Ni Yuan
• Tang Guo Yu
• Xie Guo Tong
• Zhao Shi Wan

Classifications

• G10L13/047

  Architecture of speech synthesisers · CPC title

• G06F40/56 Primary

  Natural language generation · CPC title

• G10L2015/223

  Execution procedure of a spoken command · CPC title

• G10L15/18

  using natural language modelling · CPC title

• G10L13/033 Primary

  Voice editing, e.g. manipulating the voice of the synthesiser · CPC title