Scalable endpoint-dependent natural language understanding

• US 10,229,687 B2
• Filed: 03/10/2016
• Issued: 03/12/2019
• Est. Priority Date: 03/10/2016
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Jeong, Minwoo, Sarikaya, Ruhi
• Primary Examiner(s): Colucci, Michael C
• Application Number:
  US15/067,082
• Publication Number:
  US 20170263255A1
• Time in Patent Office: 1,097 Days
• Field of Search: 704257, 704208, 704235, 704244, 709223
• CPC Class Codes: G06F 16/33 Querying
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications

Abstract

A computer-implemented technique is described for processing a linguistic item (e.g., a query) in an efficient and scalable manner. The technique interprets the linguistic item using a language understanding (LU) system in a manner that is based on a particular endpoint mechanism from which the linguistic item originated. The LU system may include an endpoint-independent subsystem, an endpoint-dependent subsystem, and a ranking component. The endpoint-independent subsystem interprets the linguistic item in a manner that is independent of the particular endpoint mechanism. The endpoint-dependent subsystem interprets the linguistic item in a manner that is dependent on the particular endpoint mechanism. The ranking component generates final interpretation results based on intermediate results generated by the endpoint-independent subsystem and the endpoint-dependent subsystem, e.g., by identifying the most likely interpretation of the linguistic item.

Claims

1. A system comprising:

a processing device; and

a storage resource storing instructions which, when executed by the processing device, cause the processing device to;

receive text-based or voice-recognized words input by a user to a particular physical device;

interpret the text-based or voice-recognized words in a manner that is independent of the particular physical device to obtain a device-independent candidate meaning for the text-based or voice-recognized words, the device-independent candidate meaning being applicable to a plurality of different physical devices;

interpret the text-based or voice-recognized words in a manner that is dependent on the particular physical device to obtain a device-dependent candidate meaning for the text-based or voice-recognized words;

generate final interpretation results by ranking the device-independent candidate meaning for the text-based or voice-recognized words relative to the device-dependent candidate meaning for the text-based or voice-recognized words and selecting one of the device-independent candidate meaning or the device-dependent candidate meaning as the final interpretation results; and

output the final interpretation results.

2. The system of claim 1, wherein the instructions, when executed by the processing device, cause the processing device to:

calculate a first score for the device-independent candidate meaning, the first score reflecting a first likelihood that the device-independent candidate meaning is the true meaning of the text-based or voice-recognized words;

calculate a second score for the device-dependent candidate meaning, the second score reflecting a second likelihood that the device-dependent candidate meaning is the true meaning of the text-based or voice-recognized words; and

rank the device-independent candidate meaning relative to the device-dependent candidate meaning based at least on the first score and the second score.

3. The system of claim 2, wherein the instructions, when executed by the processing device, cause the processing device to:

select the device-independent candidate meaning for the final interpretation results in an instance when the first score exceeds the second score; and

select the device-dependent candidate meaning for the final interpretation results in another instance when the second score exceeds the first score.

4. The system of claim 1, provided on a server located remotely from the particular physical device.

5. The system of claim 1, wherein the instructions, when executed by the processing device, cause the processing device to:

use a device-independent interpreter component to interpret the text-based or voice-recognized words with respect to a particular domain and output the device-independent candidate meaning; and

use a device-dependent interpreter component to interpret the text-based or voice-recognized words with respect to another domain and output the device-dependent candidate meaning.

6. The system of claim 5, wherein the device-independent interpreter component or the device-dependent interpreter component includes a corresponding domain determination component configured to determine a respective domain for the text-based or voice-recognized words.

7. The system of claim 5, wherein the device-independent interpreter component or the device-dependent interpreter component includes an intent determination component configured to determine an intent associated with the text-based or voice-recognized words.

8. The system of claim 5, wherein the device-independent interpreter component or the device-dependent interpreter component includes a slot value determination component configured to extract one or more slot values for individual words.

9. The system of claim 5, wherein the instructions, when executed by the processing device, cause the processing device to:

filter interpreter results produced by the device-dependent interpreter component based at least on device information that identifies the particular physical device that has been used to provide the text-based or voice-recognized words.

10. The system of claim 1, wherein the instructions, when executed by the processing device, cause the processing device to:

determine different probability values for the device-independent candidate meaning for the text-based or voice-recognized words and the device-dependent candidate meaning for the text-based or voice-recognized words; and

select, as the final interpretation results, a particular candidate meaning having the highest probability.

11. The system of claim 1, wherein the instructions, when executed by the processing device, cause the processing device to:

use one or more machine-learned models to obtain the device-independent candidate meaning, the device-dependent candidate meaning, or the final interpretation results.

12. The system of claim 1, wherein the instructions, when executed by the processing device, cause the processing device to:

use one or more rule-application engines to obtain the device-independent candidate meaning, the device-dependent candidate meaning, or the final interpretation results.

13. The system of claim 1, wherein the instructions, when executed by the processing device, cause the processing device to

cause a computer-implemented action specified by the final interpretation results.

14. A method implemented by one or more computing devices, the method comprising:

receiving one or more voice-recognized or text-based words provided by a particular device in response to interaction by a user with an input device associated with the particular device; and

processing the one or more words by;

interpreting the one or more voice-recognized or text-based words using a device-independent interpreter that outputs a first candidate command conveyed by the one or more voice-recognized or text-based words without considering the particular device that provided the one or more voice-recognized or text-based words;

interpreting the one or more voice-recognized or text-based words using a device-dependent interpreter that outputs a second candidate command conveyed by the one or more voice-recognized or text-based words by explicitly considering the particular device that provided the one or more voice-recognized or text-based words;

determining respective likelihood scores for the first candidate command output by the device-independent interpreter and the second candidate command output by the device-dependent interpreter;

based at least on the respective likelihood scores, selecting a selected command from the first candidate command and the second candidate command; and

performing a computer-implemented action in response to the selected command.

15. The method of claim 14, further comprising:

considering hardware capabilities of the particular device when interpreting the one or more voice-recognized or text-based words using the device-dependent interpreter.

16. The method of claim 14, further comprising:

ranking the first candidate command relative to the second candidate command based at least on the respective likelihood scores; and

selecting the selected command based at least on the ranking.

17. The method of claim 14, wherein the device-independent interpreter and the device-dependent interpreter are configured to perform word interpretation with respect to different domains.

18. The method of claim 17, wherein the first candidate command involves searching for a hotel and the second candidate command involves playing a media item.

19. A computer-readable storage medium storing computer-readable instructions which, when executed by one or more processor devices, cause the one or more processor devices to perform acts comprising:

interpreting a specific sequence of words in a device-independent manner that is independent of a particular device from which the specific sequence of words has been received to obtain a device-independent candidate meaning of the specific sequence of words;

interpreting the specific sequence of words in a device-dependent manner that is dependent on the particular device from which the specific sequence of words has been received to obtain a device-dependent candidate meaning of the specific sequence of words that is different from the device-independent candidate meaning;

performing a comparison of respective probability values for the device-independent candidate meaning and the device-dependent candidate meaning; and

based at least on the comparison, identifying a candidate interpretation result item that is a most likely interpretation of the specific sequence of words, the candidate interpretation result item being selected from among the device-independent candidate meaning and the device-dependent candidate meaning.

20. The computer-readable storage medium of claim 19, wherein interpreting the specific sequence of words in the device-independent manner does not consider hardware capabilities of the particular device, and interpreting the specific sequence of words in the device-dependent manner explicitly considers the hardware capabilities of the particular device.