Contextual language generation by leveraging language understanding

• US 9,792,281 B2
• Filed: 06/15/2015
• Issued: 10/17/2017
• Est. Priority Date: 06/15/2015
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Sarikaya, Ruhi
• Primary Examiner(s): Nguyen, Cam-Linh
• Application Number:
  US14/740,082
• Publication Number:
  US 20160364382A1
• Time in Patent Office: 855 Days
• Field of Search: 707732, 707708
• CPC Class Codes: G06F 16/3322 using system suggestions G06F16/3325 takes precedence
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30

Abstract

Technology is provided for improving digital assistant performance by generating and presenting suggestions to users for completing a task or a session. To generate the suggestions, a machine learned language prediction model is trained with features extracted from multiple sources, such as log data and session context. When input is received from a user, the trained machine learned language prediction model is used to determine the most likely suggestion to present to the user to lead to successful task completion. In generating the suggestion, intermediate suggestion data, such as a domain, intent, and/or slot, is generated for the suggestion. From the generated intermediate suggestion data for the suggestion, a surface form of the suggestion is generated that can be presented to the user. The resulting suggestion and related context may further be used to continue training the machine learned language prediction model.

Claims

1. A system comprising:

a machine-language-based classifier, wherein the machine-language-based classifier is configured to perform the following actions;

receive a first user input requesting a digital assistant to perform a requested task;

identify a first context for the user;

analyze the first user input to determine first intermediate task data corresponding to the requested first task;

provide the first context and the first intermediate task data as inputs to a machine learned language prediction model, wherein the machine learned language prediction model is based on log data, the log data comprising historical data representing previous interactions between one or more users and one or more digital assistant applications;

receive as output from the machine learned language prediction model, first intermediate suggestion data for generating a first suggestion for the user, wherein the first suggestion is for a second task to be requested based on the inputs to the machine learned language prediction model;

receive a second user input, wherein the second user input is the same as the first user input;

receive a second context for the user, wherein the second context is different from the first context;

analyze the second user input to determine second intermediate task data corresponding to the requested task;

provide the second context and the second intermediate task data as inputs to the machine learned language prediction model; and

receive as output from the machine learned language prediction model, second intermediate suggestion data for generating a second suggestion for the user, wherein the second suggestion is different from the first suggestion; and

an output module, wherein the output module is configured to present the first suggestion and the second suggestion to the user.

2. The system of claim 1, wherein the first intermediate suggestion data includes at least one data type selected from the group consisting of a domain, an intent, and a slot for generating the first suggestion to the user.

3. The system of claim 1, further comprising a language generation suggestion module configured to generate a surface form of the first suggestion based on the first intermediate suggestion data.

4. The system of claim 1, further comprising a log data module configured to receive the first intermediate suggestion data, the first suggestion, and the first context to the log data.

5. The system of claim 1, further comprising:

a machine learned language prediction module, wherein the machine learned language prediction module is configured to perform the following actions;

receive updated log data; and

train the machine learned language prediction model with the updated log data to generate an updated machine learned language prediction model.

6. The system of claim 2, wherein the machine learned language prediction model includes a predictive model for domains, a predictive model for intents, and a predictive model of slots.

7. A computer-implemented method comprising:

receiving a first user input requesting a digital assistant to perform a requested task;

identifying a first context for the user;

analyzing the first user input to determine first intermediate task data corresponding to the requested first task;

providing the first context and the first intermediate task data as inputs to a machine learned language prediction model, wherein the machine learned language prediction model is trained from log data, the log data comprising historical data representing previous interactions between one or more users and one or more digital assistant applications;

receiving as output from the machine learned language prediction model, first intermediate suggestion data for generating a first suggestion for the user, wherein the first suggestion is for a second task to be requested based on the inputs to the machine learned language prediction model;

presenting the first suggestion to the user;

receiving a second user input, wherein the second user input is the same as the first user input;

receiving a second context for the user, wherein the second context is different from the first context;

analyzing the second user input to determine second intermediate task data corresponding to the requested task;

providing the second context and the second intermediate task data as inputs to the machine learned language prediction model;

receiving as output from the machine learned language prediction model, second intermediate suggestion data for generating a second suggestion for the user, wherein the second suggestion is different from the first suggestion; and

presenting the second suggestion to the user.

8. The method of claim 7, wherein the first intermediate suggestion data includes at least one data type selected from the group consisting of:

a domain, an intent, and a slot for generating the first suggestion to the user.

9. The method of claim 7, further comprising generating a surface form of the first suggestion based on the intermediate suggestion data.

10. The method of claim 7, further comprising:

receiving updated log data;

training the machine learned language prediction model while offline with the updated log data to generate an updated machine learned language prediction model.

11. The method of claim 7, further comprising:

receiving updated log data;

training the machine learned language prediction model at runtime with the updated log data to generate an updated machine learned language prediction model.

12. The method of claim 7, wherein the machine learned language prediction model includes a predictive model for domains, a predictive model for intents, and a predictive model of slots.