Crowdsourced, grounded language for intent modeling in conversational interfaces

• US 9,754,585 B2
• Filed: 04/03/2012
• Issued: 09/05/2017
• Est. Priority Date: 04/03/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Brockett, Christopher John, Choudhury, Piali, Dolan, William Brennan, Ju, Yun-Cheng, Pantel, Patrick, Sophy, Noelle Mallory, Volkova, Svitlana
• Primary Examiner(s): COLUCCI, MICHAEL C
• Application Number:
  US13/438,751
• Publication Number:
  US 20130262114A1
• Time in Patent Office: 1,981 Days
• Field of Search: 704243, 704277, 7042701, 704257, 704235, 704209, 704200, 715853, 715758, 715256, 715254, 715234, 709204, 707794, 707769, 707760, 707706
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

Different advantageous embodiments provide a crowdsourcing method for modeling user intent in conversational interfaces. One or more stimuli are presented to a plurality of describers. One or more sets of describer data are captured from the plurality of describers using a data collection mechanism. The one or more sets of describer data are processed to generate one or more models. Each of the one or more models is associated with a specific stimulus from the one or more stimuli.

Claims

1. A system comprising:

one or more processors;

a model generation mechanism configured to;

identify a first set of data;

based on the identified first set of data, select a plurality of users;

present the first set of data to the selected plurality of users; and

process a second set of data captured from the selected plurality of users in response to the first set of data presented to the selected plurality of users to generate metadata documents that correspond to respective users of the selected plurality of users;

a clarification mechanism configured to;

process the metadata documents;

based on the processing, identify ambiguities in the metadata documents, the ambiguities comprising relative terms or descriptive words having more than one meaning;

generate a set of clarifying questions based on the identified ambiguities; and

present the set of clarifying questions to the selected plurality of users;

a data collection mechanism configured to capture clarifying data from the selected plurality of users in response to the presentation of the set of clarifying questions;

the model generation mechanism further configured to;

refine the metadata documents based on the clarifying data; and

generate a model that provides natural language descriptions of the first set of data based on the refined metadata documents and the clarifying data; and

a conversational interface configured to be trained using the model generated by the model generation mechanism to determine a meaning behind a verbal communication from a user.

2. The system of claim 1, wherein the metadata documents further correspond to the first set of data.

3. The system of claim 1, wherein the conversational interface is further configured to be trained using a training algorithm in conjunction with the model.

4. The system of claim 1, wherein the conversational interface receives a verbal input that triggers a machine state change.

5. The system of claim 1, wherein the first set of data includes a stimuli and a plurality of questions associated with the stimuli.

6. The system of claim 5, wherein the stimuli is a representation of a state or change of state of an action or event that can be applied to a human sensory receptor.

7. An apparatus, comprising:

a processor programmed to;

identify a first set of data;

based on the identified first set of data, select a plurality of users; and

present the first set of data to the selected plurality of users;

a data collection mechanism configured to capture a second set of data from the selected plurality of users in response to the first set of data presented to the selected plurality of users;

a model generation mechanism configured to;

process the second set of data; and

generate a model that provides natural language descriptions of the first set of data, the generated model including a set of metadata documents associated with the first set of data; and

a clarification mechanism configured to;

process the metadata documents; and

based on the processing, identify ambiguities in the set of metadata documents, the ambiguities comprising relative terms or descriptive words having more than one meaning; and

generate a set of clarifying questions based on the identified ambiguities, wherein the set of clarifying questions are presented to the selected plurality of users, and wherein the data collection mechanism captures clarifying data from the selected plurality of users in response to the presentation of the set of clarifying questions;

the model generation mechanism further configured to;

refine the metadata documents based on the clarifying data; and

refine the generated model based on the refined metadata documents and the clarifying data.

8. The apparatus of claim 7, wherein the first set of data is a representation of a state or change of state.

9. The apparatus of claim 7, wherein the selected plurality of users are a crowd of people.

10. The apparatus of claim 7, wherein the second set of data corresponds to the first set of data presented to the selected plurality of users.

11. The apparatus of claim 7, wherein the model generation mechanism includes one or more ranking algorithms used to process the second set of data.

12. The apparatus of claim 7 further comprising:

a knowledge repository; and

a knowledge harvesting mechanism configured to interact with the knowledge repository to generate a set of word associations.

13. The apparatus of claim 12, wherein the model generation mechanism uses the set of word associations generated by the knowledge harvesting mechanism along with the one or more ranking algorithms to process the second set of data and generate the set of metadata documents.

14. The apparatus of claim 7, wherein the stimuli is an image.

15. A method comprising:

identify a first set of data;

based on the identified first set of data, select a plurality of users;

presenting the first set of data to the selected plurality of users for determining a meaning behind a verbal communication from a user in conversational interfaces, the first set of data comprising a stimuli and questions regarding the stimuli;

capturing a second set of user data in response to the second set of data from the selected plurality of users using a data collection mechanism;

processing, by a processor, the second set of data;

based on processing the second set of data, generate metadata documents that correspond to respective users of the selected plurality of users;

process the metadata documents;

based on processing the metadata documents, identify ambiguities in the second set of data, the ambiguities comprising relative terms or descriptive words having more than one meaning;

generate, by the processor, a set of clarifying questions based on the identified ambiguities;

present the set of clarifying questions to the selected plurality of users;

capture clarifying data from the selected plurality of users in response to the presentation of the set of clarifying questions;

refine the metadata documents based on the clarifying data; and

generate, by the processor, one or more models that provide natural language descriptions of the first set of data based on the refined metadata documents and the clarifying data.

16. The method of claim 15, wherein the second set of data comprises a collection of natural language descriptions associated with the first set of data.

17. The method of claim 15, wherein processing the second set of data comprises filtering and clustering the second set of data to generate the metadata documents, wherein the metadata documents correspond to a set of data from the first set of data, and wherein a subset of metadata documents in the metadata documents correspond to a respective user from the selected plurality of users.

18. The method of claim 15, wherein processing the second set of data comprises refining the second set of data using a set of word associations generated by a knowledge harvesting mechanism to generate the metadata documents.

19. The method of claim 15 further comprising:

training a conversational interface using the one or more models.

20. The method of claim 19, wherein training a conversational interface using the one or more models comprises determining a meaning behind a verbal communication from a user.