Context interpretation in natural language processing using previous dialog acts

• US 10,726,831 B2
• Filed: 05/20/2014
• Issued: 07/28/2020
• Est. Priority Date: 05/20/2014
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Di Fabbrizio, Giuseppe, Bharathi, Shishir Sridhar, Shi, Ying, Mathias, Lambert
• Primary Examiner(s): Sirjani, Fariba
• Application Number:
  US14/283,017
• Publication Number:
  US 20150340033A1
• Time in Patent Office: 2,261 Days
• Field of Search: None
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 2015/225 Feedback of the input speech
• CPC Class Codes: G10L 2015/228 of application context

Abstract

Features are disclosed for processing and interpreting natural language, such as interpretations of user utterances, in multi-turn dialog interactions. Context information regarding interpretations of user utterances and system responses to the user utterances can be maintained. Subsequent user utterances can be interpreted using the context information, rather than being interpreted without context. In some cases, interpretations of subsequent user utterances can be merged with interpretations of prior user utterances using a rule-based framework. Rules may be defined to determine which interpretations may be merged and under what circumstances they may be merged.

Claims

1. A computer-implemented method comprising:

under control of a natural language processing system comprising one or more computing devices configured with specific computer-executable instructions,

storing context information associated with a first natural language input and a first response to the first natural language input, wherein the context information comprises a semantic representation of the first natural language input and a semantic representation of the first response to the first natural language input, and wherein the first response to the first natural language input comprises a prompt, by the natural language processing system, for information from a user;

receiving (i) a second natural language input, and (ii) a third natural language input, wherein the second natural language input is received in response to the first response and prior to receiving the third natural language input, and wherein the third natural language input is received in response to a second response, by the natural language processing system, to the second natural language input;

determining that the second response does not comprise a second prompt, by the natural language processing system, for the information prompted by the first response;

determining, subsequent to determining that the second response does not comprise the second prompt, that the first response comprises the prompt for information;

determining that the third natural language input is associated with a same domain as the first natural language input;

determining that a semantic representation of the third natural language input comprises a slot value corresponding to the information prompted by the first response;

determining, based at least partly on the third natural language input relating to the information prompted by the first response, to merge the semantic representation of the third natural language input with the semantic representation of the first natural language input;

generating natural language processing results for the third natural language input by merging the semantic representation of the third natural language input with the semantic representation of the first natural language input instead of a semantic representation of the second natural language input; and

generating a third response to the third natural language input based at least partly on the natural language processing results.

2. The computer-implemented method of claim 1, wherein the first natural language input comprises text or speech.

3. The computer-implemented method of claim 1, wherein the semantic representation of the third natural language input is merged with the semantic representation of the first natural language input by replacing a slot value of the semantic representation of the first natural language input with a slot value of the semantic representation of the third natural language input.

4. The computer-implemented method of claim 1, wherein the semantic representation of the third natural language input is merged with the semantic representation of the first natural language input by replacing an intent of the semantic representation of the first natural language input with an intent of the semantic representation of the third natural language input.

5. The computer-implemented method of claim 1, wherein the natural language processing results are generated based at least partly on a context interpretation rule.

6. The computer-implemented method of claim 5, wherein the context interpretation rule relates to determining whether a domain associated with the first natural language input is equivalent to a domain associated with the third natural language input.

7. The computer-implemented method of claim 5, wherein the context interpretation rule relates to determining whether at least one of an intent or a slot associated with the semantic representation of the first natural language input is equivalent to at least one of an intent or a slot associated with the semantic representation of the third natural language input.

8. The computer-implemented method of claim 5, wherein the context interpretation rule relates to the semantic representation of the first response to the first natural language input.

9. The computer-implemented method of claim 8, wherein the context interpretation rule further relates to determining whether the semantic representation of the first response to the first natural language input corresponds to an elicitation of information associated with the semantic representation of the first natural language input.

10. The computer-implemented method of claim 8, wherein the context interpretation rule further relates to determining whether the semantic representation of the first response to the first natural language input corresponds to a confirmation of information associated with the semantic representation of the first natural language input.

11. The computer-implemented method of claim 1, further comprising re-ranking at least a portion of the natural language processing results.

12. The computer-implemented method of claim 1, further comprising filtering at least a portion of the natural language processing results.

13. The computer-implemented method of claim 1, further comprising determining that the semantic representation of the third natural language input is to be merged with the semantic representation of the first natural language input based at least partly on the semantic representation of the third natural language input comprising a value for a content slot, wherein the semantic representation of the first natural language input comprises the content slot.

14. The computer-implemented method of claim 1, further comprising determining that the semantic representation of the third natural language input is to be merged with the semantic representation of the first natural language input based at least partly on the semantic representation of the third natural language input comprising a value for an intent, wherein the prompt for information comprises a prompt for information regarding an intent of the first natural language input.

15. A system comprising:

a computer-readable memory storing executable instructions; and

one or more processors in communication with the computer-readable memory, wherein the one or more processors are programmed by the executable instructions to at least;

store context information associated with a first natural language input and a first response by the system to the first natural language input, wherein the context information comprises a semantic representation of the first natural language input and a semantic representation of the first response, and wherein the first response comprises a prompt for information;

receive (i) a second natural language input and (ii) a third natural language input, wherein the second natural language input is received in response to the first response and prior to receiving the third natural language input, and wherein the third natural language input is received in response to a second response by the system to the second natural language input;

determine that the second response does not comprise a second prompt by the system for the information prompted by the first response;

determine, subsequent to determining that the second response does not comprise the second prompt, that the first response comprises the prompt for information;

determine that the third natural language input is associated with a same domain as the first natural language input;

determine that a semantic representation of the third natural language input comprises a slot value corresponding to the information prompted by the first response;

determine, based at least partly on the third natural language input relating to the information prompted by the first response, to merge the semantic representation of the third natural language input with the semantic representation of the first natural language input;

generate natural language processing results for the third natural language input by merging the semantic representation of the third natural language input with the semantic representation of the first natural language input instead of a semantic representation of the second natural language input; and

generate a third response to the third natural language input based at least partly on the natural language processing results.

16. The system of claim 15, wherein the semantic representation of the first natural language input comprises a data structure comprising:

a first data element representing a domain of the first natural language input;

a second data element representing an intent of the first natural language input; and

one or more additional data elements representing one or more slots associated with the intent.

17. The system of claim 15, wherein the one or more processors are further programmed by the executable instructions to add, to a ranked set of natural language processing results, a merged semantic representation comprising the semantic representation of the third natural language input merged with the semantic representation of the first natural language input, wherein the ranked set of natural language processing results comprises a semantic representation of the third natural language input, and wherein the merged semantic representation is associated with ranking data used to add the merged semantic representation at a particular position in the ranked set of natural language processing results.

18. The system of claim 15, wherein the one or more processors are further programmed by the executable instructions to determine that a semantic representation of the third natural language input comprises a value associated with a slot of an intent of the semantic representation of the first natural language input, and wherein the information prompted by the first response is associated with the slot.

19. The system of claim 15, wherein the first natural language input, the second natural language input, and the third natural langue input are each individual dialog turns of a multi-turn dialog, and wherein the one or more processors are further programmed to:

determine to iterate through a first subset of dialog turns of the multi-turn dialog; and

iterate through the first subset of dialog turns to determine whether to merge the semantic representation of the third natural language input with individual semantic representations of a plurality of semantic representations associated with the first subset of dialog turns, wherein the one or more processors do not iterate through a second subset of dialog turns of the multi-turn dialog.