System and method for generating a multi-lingual and multi-intent capable semantic parser based on automatically generated operators and user-designated utterances relating to the operators

• US 10,579,738 B2
• Filed: 04/05/2018
• Issued: 03/03/2020
• Est. Priority Date: 04/05/2017
• Status: Active Grant

• Current Assignee: VoiceBox Technologies Corporation (Microsoft Corporation)
• Original Assignee: VoiceBox Technologies Corporation (Microsoft Corporation)
• Inventors: Johnson, Mark Edward
• Primary Examiner(s): Gauthier, Gerald
• Application Number:
  US15/946,687
• Publication Number:
  US 20180300313A1
• Time in Patent Office: 698 Days
• Field of Search: 704 2, 704 9, 704 10, 704257, 704 4, 704251, 709206, 707769
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

The disclosure relates to systems and methods for generating semantic parsers based on automatically generated operators and user-designated utterances relating to the operators for use in natural language processing. The system may automatically generate multiple operators that each express a respective computer-executable instruction that resolves a request. These operators may be expressed in a manner that is machine-readable and not necessarily for consumption by a human user. The system may generate a canonical statement that expresses the request in a first manner that a human user would be able to understand. The system may generate a task, such as crowd-sourced task, for a human user to provide an utterance that conveys the canonical statement in a second manner different than the first manner. By doing so, the system may rapidly build operators and learn how humans would utter requests resolved by instructions encoded in the operators for building semantic parsers.

Claims

1. A system of generating a semantic parser for natural language processing, comprising:

one or more physical processors programmed to;

generate an operator that represents a computer-executable instruction that resolves a request;

generate at least one canonical statement of the operator for presentation to a human user, wherein the at least one canonical statement expresses the request in a first manner, and wherein the at least one canonical statement is generated after generating the operator;

generate a task, after the operator has been generated and to be performed by the human user, to make an utterance that conveys the canonical statement in a second manner different than the first manner;

receive the utterance;

associate the utterance with the operator; and

generate the semantic parser based on the association of the utterance with the operator.

2. The system of claim 1, wherein the operator is a de-lexicalized operator.

3. The system of claim 2, wherein the de-lexicalized operator replaces an entity with a named entity type.

4. The system of claim 2, wherein the de-lexicalized operator utilizes an attribute-value pair format comprising an attribute and a corresponding value of the attribute.

5. The system of claim 4, wherein the attribute-value format includes a head attribute.

6. The system of claim 4, wherein the at least one canonical statement is generated by converting the attribute-value pair of an operator using a rule-based process.

7. The system of claim 6, wherein the converting of the attribute-value pair comprises mapping the attribute-value pair to a text string.

8. The system of claim 1, wherein at least two different canonical statements in two different languages are generated for the operator.

9. The system of claim 8, wherein the operator is a de-lexicalized operator.

10. The system of claim 9, wherein the de-lexicalized operator utilizes an attribute-value pair format comprising an attribute and a corresponding value of the attribute.

11. A computer-implemented method of generating a semantic parser for natural language processing, comprising:

generating an operator that represents a computer-executable instruction that resolves a request;

generating at least one canonical statement of the operator for presentation to a human user, wherein the at least one canonical statement expresses the request in a first manner, and wherein the at least one canonical statement is generated after generating the operator;

generating a task, after the operator has been generated and to be performed by the human user, to make an utterance that conveys the canonical statement in a second manner different than the first manner;

receiving the utterance;

associating the utterance with the operator; and

generating the semantic parser based on the association of the utterance with the operator.

12. The method of claim 11, wherein the operator is a de-lexicalized operator.

13. The method of claim 12, wherein the de-lexicalized operator replaces an entity with a named entity type.

14. The method of claim 12, wherein the de-lexicalized operator utilizes an attribute-value pair format comprising an attribute and a corresponding value of the attribute.

15. The method of claim 14, wherein the attribute-value format includes a head attribute.

16. The method of claim 14, wherein the at least one canonical statement is generated by converting the attribute-value pair of an operator using a rule-based process.

17. The method of claim 16, wherein the converting of the attribute-value pair comprises mapping the attribute-value pair to a text string.

18. The method of claim 11, wherein at least two different canonical statements in two different languages are generated for the operator.

19. The method of claim 18, wherein the operator is a de-lexicalized operator.

20. The method of claim 19, wherein the de-lexicalized operator utilizes an attribute-value pair format comprising an attribute and a corresponding value of the attribute.