Semantic re-ranking of NLU results in conversational dialogue applications

• US 9,269,354 B2
• Filed: 03/11/2013
• Issued: 02/23/2016
• Est. Priority Date: 03/11/2013
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Gandrabur, Simona, Lavalle, Jean-Francois, Tremblay, Ral
• Primary Examiner(s): Hang, Vu B
• Application Number:
  US13/793,854
• Publication Number:
  US 20140257794A1
• Time in Patent Office: 1,079 Days
• Field of Search: 704/1, 704/2, 704/3, 704/7, 704/9, 704/10, 704/231
• US Class Current: 1/1
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• 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

Abstract

A human-machine dialog system is described which has multiple computer-implemented dialog components. A user client delivers output prompts to a human user and receives dialog inputs from the human user including speech inputs. An automatic speech recognition (ASR) engine processes the speech inputs to determine corresponding sequences of representative text words. A natural language understanding (NLU) engine processes the text words to determine corresponding NLU-ranked semantic interpretations. A semantic re-ranking module re-ranks the NLU-ranked semantic interpretations based on at least one of dialog context information and world knowledge information. A dialog manager responds to the re-ranked semantic interpretations and generates the output prompts so as to manage a dialog process with the human user.

Claims

1. A method comprising:

receiving, by a computing system, data generated based on spoken user responses to prompts associated with a dialogue, the data comprising a portion corresponding to one or more of the spoken user responses spoken before a subsequently spoken response of the spoken user responses;

generating, by at least one processor of the computing system, a list of natural language understanding (NLU)-ranked semantic interpretations for the subsequently spoken response;

determining, by the computing system and based on the portion corresponding to the one or more of the spoken user responses spoken before the subsequently spoken response, a plurality of key-value pairs corresponding to different possible resolutions for an anaphora in the subsequently spoken response; and

re-ranking, by the computing system and based on the plurality of key-value pairs, the list of NLU-ranked semantic interpretations to identify a semantic interpretation that resolves the anaphora in the subsequently spoken response based on a context of the dialogue determined from the one or more of the spoken user responses spoken before the subsequently spoken response.

2. The method of claim 1, comprising determining, by the computing system, that a key-value pair of the plurality of key-value pairs corresponds to the semantic interpretation that resolves the anaphora based on a determination that the key-value pair comprises:

a context type corresponding to the context of the dialogue; and

a context value indicating that the semantic interpretation is context based.

3. The method of claim 1, wherein determining the plurality of key-value pairs comprises determining, for each element of a plurality of elements of the list, a key-value pair comprising a context type and a context value.

4. The method of claim 3, wherein determining the key-value pair comprises determining at least one of a state of the dialogue, an expectation of the dialogue, a focus of the dialogue, or a selection based on the one or more of the spoken user responses.

5. The method of claim 3, comprising determining, by the computing system and for each key-value pair of the plurality of key-value pairs, a confidence score indicating a level of confidence that a possible resolution of the different possible resolutions corresponding to the key-value pair resolves the anaphora.

6. The method of claim 5, comprising re-ranking, for each key-value pair of the plurality of key-value pairs and based on the confidence score, an element of the plurality of elements corresponding to the key-value pair within the list.

7. A system comprising:

at least one processor; and

a memory storing instructions that when executed by the at least one processor cause the system to;

receive data generated based on spoken user responses to prompts associated with a dialogue, the data comprising a portion corresponding to one or more of the spoken user responses spoken before a subsequently spoken response of the spoken user responses;

generate a list of natural language understanding (NLU)-ranked semantic interpretations for the subsequently spoken response;

determine, based on the portion corresponding to the one or more of the spoken user responses spoken before the subsequently spoken response, a plurality of key-value pairs corresponding to different possible resolutions for an anaphora in the subsequently spoken response; and

re-rank, based on the plurality of key-value pairs, the list of NLU-ranked semantic interpretations to identify a semantic interpretation that resolves the anaphora in the subsequently spoken response based on a context of the dialogue determined from the one or more of the spoken user responses spoken before the subsequently spoken response.

8. The system of claim 7, wherein the instructions, when executed by the at least one processor, cause the system to determine that a key-value pair of the plurality of key-value pairs corresponds to the semantic interpretation that resolves the anaphora based on a determination that the key-value pair comprises:

a context type corresponding to the context of the dialogue; and

a context value indicating that the semantic interpretation is context based.

9. The system of claim 7, wherein the instructions, when executed by the at least one processor, cause the system to determine, for each element of a plurality of elements of the list, a key-value pair comprising a context type and a context value.

10. The system of claim 9, wherein the instructions, when executed by the at least one processor, cause the system to determine the key-value pair based on at least one of a state of the dialogue, an expectation of the dialogue, a focus of the dialogue, or a selection based on the one or more of the spoken user responses.

11. The system of claim 9, wherein the instructions, when executed by the at least one processor, cause the system to determine, for each key-value pair of the plurality of key-value pairs, a confidence score indicating a level of confidence that a possible resolution of the different possible resolutions corresponding to the key-value pair resolves the anaphora.

12. The system of claim 11, wherein the instructions, when executed by the at least one processor, cause the system to re-rank, for each key-value pair of the plurality of key-value pairs and based on the confidence score, an element of the plurality of elements corresponding to the key-value pair within the list.

13. One or more non-transitory computer-readable media comprising instructions that when executed by one or more computers cause the one or more computers to:

receive data generated based on spoken user responses to prompts associated with a dialogue, the data comprising a portion corresponding to one or more of the spoken user responses spoken before a subsequently spoken response of the spoken user responses;

generate a list of natural language understanding (NLU)-ranked semantic interpretations for the subsequently spoken response;

determine, based on the portion corresponding to the one or more of the spoken user responses spoken before the subsequently spoken response, a plurality of key-value pairs corresponding to different possible resolutions for an anaphora in the subsequently spoken response; and

re-rank, based on the plurality of key-value pairs, the list of NLU-ranked semantic interpretations to identify a semantic interpretation that resolves the anaphora in the subsequently spoken response based on a context of the dialogue determined from the one or more of the spoken user responses spoken before the subsequently spoken response.

14. The one or more non-transitory computer-readable media of claim 13, wherein the instructions, when executed by the one or more computers, cause the one or more computers to determine that a key-value pair of the plurality of key-value pairs corresponds to the semantic interpretation that resolves the anaphora based on a determination that the key-value pair comprises:

a context type corresponding to the context of the dialogue; and

a context value indicating that the semantic interpretation is context based.

15. The one or more non-transitory computer-readable media of claim 13, wherein the instructions, when executed by the one or more computers, cause the one or more computers to determine, for each element of a plurality of elements of the list, a key-value pair comprising a context type and a context value.

16. The one or more non-transitory computer-readable media of claim 15, wherein the instructions, when executed by the one or more computers, cause the one or more computers to determine, for each key-value pair of the plurality of key-value pairs, a confidence score indicating a level of confidence that a possible resolution of the different possible resolutions corresponding to the key-value pair resolves the anaphora.

17. The one or more non-transitory computer-readable media of claim 16, wherein the instructions, when executed by the one or more computers, cause the one or more computers to re-rank, for each key-value pair of the plurality of key-value pairs and based on the confidence score, an element of the plurality of elements corresponding to the key-value pair within the list.