Rapid deployment of dialogue system

• US 10,762,892 B2
• Filed: 07/16/2018
• Issued: 09/01/2020
• Est. Priority Date: 02/23/2017
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Semantic Machines, Inc. (Microsoft Corporation)
• Inventors: Rusak, Jesse Daniel Eskes, Hall, David Leo Wright, Wolfe, Jason Andrew, Roth, Daniel Lawrence, Klein, Daniel, Cohen, Jordan Rian
• Primary Examiner(s): Patel, Shreyans A
• Application Number:
  US16/036,622
• Publication Number:
  US 20190103092A1
• Time in Patent Office: 778 Days
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/169 Annotation, e.g. comment data or footnotes
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G10L 15/063 Training
• 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 2015/0638 Interactive procedures
• CPC Class Codes: G10L 2015/225 Feedback of the input speech

Abstract

A method for a dialogue system includes establishing a dialogue session between an application executing on a server and a remote machine. The dialogue session includes one or more utterances received from a user at the remote machine. A natural language processing machine identifies a request associated with a computer-readable representation of an utterance. A dialogue expansion machine generates a plurality of alternative actions for responding to the request. A previously-trained machine learning confidence model assesses a confidence score for each alternative. If a highest confidence score for a top alternative does not satisfy a threshold, the plurality of alternatives including the top alternative are transmitted to a remote machine (which may be the same remote machine or a different remote machine) for review by a human reviewer. After the dialogue system and/or the human reviewer select an alternative, computer-readable instructions defining the selected alternative are executed.

Claims

1. A method for a computerized dialogue system, comprising:

establishing a dialogue session between an application executing on a server and a first remote machine, the dialogue session including one or more utterances received from a user at the first remote machine;

receiving from the first remote machine, by the application on the server, a computer-readable representation of an utterance by the user;

operating, by the application on the server, a natural language processing machine to identify a request associated with the computer-readable representation of the utterance;

operating, by the application on the server, a dialogue expansion machine to generate a plurality of alternatives for responding to the request;

operating, by the application on the server, a previously-trained machine learning confidence model to assess a confidence score for each alternative of the plurality of alternatives, wherein a top alternative of the plurality of alternatives has a highest confidence score;

transmitting the plurality of alternatives including the top alternative to a second remote machine in response to determining that the highest confidence score for the top alternative of the plurality of alternatives does not satisfy a threshold; and

executing computer-readable instructions defining a selected alternative of the plurality of the alternatives based on input received from a human reviewer at the second remote machine.

2. The method of claim 1, wherein the first remote machine and the second remote machine are the same remote machine, and wherein the human reviewer is the human user.

3. The method of claim 1, wherein the first remote machine and the second remote machine are different remote machines, and wherein the human reviewer is not the human user.

4. The method of claim 1, further comprising:

automatically generating grammar rules from the utterance and response; and

applying the automatically generated grammar rules to subsequent requests when generating alternatives to responding to the subsequent requests.

5. The method of claim 4, further comprising:

receiving from the first remote machine, by the application on the server, a subsequent utterance by the user;

operating, by the application on the server, the natural language processing machine to identify a subsequent request associated with the subsequent utterance;

operating, by the application on the server, the dialogue expansion machine to generate a plurality of alternatives for responding to the subsequent request, based on the automatically generated grammar rules;

operating, by the application on the server, the confidence model to assess a confidence score for each alternative of the plurality of alternatives for responding to the subsequent request, wherein a top alternative of the plurality of alternatives for responding to the subsequent request has a highest confidence score; and

executing computer-readable instructions defining the top alternative in response to the subsequent request in response to determining that the confidence score for the top alternative satisfies a second threshold.

6. The method of claim 1, wherein the selected alternative is not the top alternative.

7. The method of claim 1, further comprising receiving, from the first remote machine, a computer-readable indication of a selection of an alternative of the plurality of alternatives, based on input received from the user at the first remote machine.

8. The method of claim 1, further comprising:

transmitting the plurality of alternatives to a plurality of users;

receiving, from each user of the plurality of users, an alternative selected by that user; and

executing computer-readable instructions defining a final selected alternative as the alternative of the plurality of alternatives that was most often selected by the plurality of users.

9. The method of claim 1, wherein the confidence score is generated based on user text representing understood and unknown portions of the user request, wherein the confidence score is relatively higher responsive to a relatively higher proportion of the user text representing understood portions of the user request.

10. A storage subsystem holding instructions executable by a logic subsystem, the instructions comprising:

instructions to establish a dialogue session between an application executing on a server and a first remote machine, the dialogue session including one or more utterances received from a user at the first remote machine;

instructions for the application on the server to receive, from the first remote machine, a computer-readable representation of an utterance by the user;

instructions for the application on the server to operate a natural language processing machine to identify a request associated with the utterance;

instructions for the application on the server to operate a dialogue expansion machine to generate a plurality of alternatives for responding to the request;

instructions for the application on the server to operate a a previously-trained machine learning confidence model to assess a confidence score for each alternative of the plurality of alternatives, wherein a top alternative of the plurality of alternatives has a highest confidence score;

instructions for the application on the server to transmit the alternatives to a second remote machine in response to determining that the highest confidence score for the top alternative of the plurality of alternatives does not satisfy a threshold; and

instructions for the application on the server to execute further computer-readable instructions defining a selected alternative of the plurality of the alternatives, based on input received from a human reviewer at the second remote machine.

11. The storage subsystem of claim 10, wherein the selected alternative is not the top alternative.

12. The storage subsystem of claim 10, the method further comprising receiving, from the first remote machine, a computer-readable selection of an alternative of the plurality of alternatives, based on input received from the user at the first remote machine.

13. The storage subsystem of claim 10, the instructions further comprising:

instructions to transmit the plurality of alternatives to a plurality of users;

instructions to receive, from each user of the plurality of users, a selected alternative selected by that user; and

instructions to execute further computer-readable instructions defining a final selected alternative as the alternative of the plurality of alternatives that was most often selected by the plurality of users.

14. The storage subsystem of claim 10, the instructions further comprising:

instructions to automatically generate grammar rules from the utterance and response; and

instructions to apply the automatically generated grammar rules to subsequent requests when generating alternatives to responding to the subsequent requests.

15. The storage subsystem of claim 14, the instructions further comprising:

instructions to receive from the first remote machine, by the application on the server, a subsequent utterance by the user;

instructions to operate, by the application on the server, the natural language processing machine to identify a subsequent request associated with the subsequent utterance;

instructions to operate, by the application on the server, the dialogue expansion machine to generate a plurality of alternatives for responding to the subsequent request, based on the automatically generated grammar rules;

instructions to operate, by the application on the server, the confidence model to assess a confidence score for each alternative of the plurality of alternatives for responding to the subsequent request, wherein a top alternative of the plurality of alternatives for responding to the subsequent request has a highest confidence score; and

instructions to execute further computer-readable instructions defining the top alternative, in response to the subsequent request in response to determining that the confidence score for the top alternative satisfies a second threshold.

16. The storage subsystem of claim 10, wherein the confidence score is generated based on user text representing understood and unknown portions of the user request, wherein the confidence score is relatively higher responsive to a relatively higher proportion of the user text representing understood portions of the user request.

17. A dialogue system, comprising:

a communication subsystem configured to;

establish a dialogue session between an application executing on a server and a first remote machine, the dialogue session including one or more utterances received from a user at the first remote machine; and

receive from the remote machine, by the application on the server, a computer-readable representation of an utterance by the user;

a natural language processing machine configured to identify a request associated with the utterance;

a dialogue expansion machine configured to generate a plurality of alternatives for responding to the request;

a machine learning confidence model previously-trained to assess a confidence score for each alternative of the plurality of alternatives such that a top alternative of the plurality of alternatives is assessed with a highest confidence score;

wherein the communication subsystem is further configured to transmit the alternatives to a second remote machine in response to determining that the highest confidence score for the top alternative of the plurality of alternatives does not satisfy a threshold; and

the dialogue system further comprising a logic subsystem configured to execute computer-readable instructions defining a selected alternative of the plurality of the alternatives based on input received from a human reviewer at the second remote machine.

18. The dialogue system of claim 17, wherein the communication subsystem is further configured to:

transmit the plurality of alternatives to a plurality of users;

receive, from each user of the plurality of users, an alternative selected by that user; and

execute computer-readable instructions defining a final selected alternative as the alternative of the plurality of alternatives that was most often selected by the plurality of users.

19. The dialogue system of claim 17, wherein the dialogue expansion machine is configured to automatically generate grammar rules from the utterance and response and apply the automatically generating grammar rules to subsequent requests when generating alternatives to responding to the subsequent requests.

20. The dialogue system of claim 19, wherein:

the communication subsystem is further configured to receive, from the first remote machine, a subsequent utterance by the user;

the natural language processing machine is further configured to identify a subsequent request associated with the subsequent utterance;

the dialogue expansion machine is further configured to generate a plurality of alternatives for responding to the subsequent request, based on the automatically generated grammar rules;

the confidence model is further configured to assess a confidence score for each alternative of the plurality of alternatives for responding to the subsequent request, wherein a top alternative of the plurality of alternatives for responding to the subsequent request has a highest confidence score; and

the logic subsystem is further configured to execute computer-readable instructions defining the top alternative in response to the subsequent request in response to determining that the confidence score for the top alternative satisfies a second threshold.