Sharing intents to provide virtual assistance in a multi-person dialog

• US 10,079,013 B2
• Filed: 11/27/2013
• Issued: 09/18/2018
• Est. Priority Date: 11/27/2013
• Status: Active Grant

• Current Assignee: SRI International, Inc.
• Original Assignee: SRI International, Inc.
• Inventors: Kalns, Edgar T., Bercow, Douglas A., Carpenter, James F.
• Primary Examiner(s): Colucci, Michael
• Application Number:
  US14/092,487
• Publication Number:
  US 20150149177A1
• Time in Patent Office: 1,756 Days
• Field of Search: 704257, 709204, 707774
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding

Abstract

A computing system is operable as virtual personal assistant (VPA) to understand relationships between different instances of natural language dialog expressed by different people in a multi-person conversational dialog session. The VPA can develop a common resource, a shared intent, which represents the VPA'"'"'s semantic understanding of at least a portion of the multi-person dialog experience. The VPA can store and manipulate multiple shared intents, and can alternate between different shared intents as the multi-person conversation unfolds. With the shared intents, the computing system can generate useful action items and present the action items to one or more of the participants in the dialog session.

Claims

1. A computer-implemented method for developing a semantic understanding of a multi-topic natural language dialog session involving at least two human participants, the method comprising, with a computing system:

by a microphone, receiving a plurality of instances of spoken natural language dialog input supplied by a plurality of human participants in the dialog session;

for one or more of the instances of dialog input, determining a domain model to apply to the dialog input;

developing one or more intents based on the determined domain model, an intent comprising a structured representation of semantic content of the one or more instances of dialog input;

storing the intents in a shared dialog context;

selecting an intent from the shared dialog context based on an intent mapping, the intent mapping defining a relationship between the selected intent and a current intent, the current intent comprising a structured representation of semantic content of a current instance of the dialog input;

creating a shared intent by combining a portion of the selected intent with a portion of the current intent;

executing an action by the computing system based on the shared intent;

wherein at least one of the intents is based on a domain model relating to a first topic and at least one other of the intents is based on a different domain model relating to a different topic than the first topic; and

wherein the selected intent is selected based at least partly on a variable temporal proximity relative to the occurrence of the current intent.

2. The method of claim 1, further comprising comparing the portion of the selected intent and the portion of the current intent and determining, based on the comparison, whether the dialog involves a negotiation of a data value by two or more of the human participants.

3. The method of claim 2, further comprising associating the shared intent with the one of the human participants in the dialog session, and creating an electronic action item for the one of the human participants, wherein the electronic action item includes semantic information extracted from the shared intent.

4. The method of claim 3, further comprising creating a second shared intent different than the shared intent by selecting a second intent from the shared dialog context and combining a portion of the second selected intent with a different portion of the current intent than the portion of the current intent used to create the shared intent.

5. The method of claim 4, further comprising associating the second shared intent with one of the human participants in the dialog session, and creating an electronic action item for the one of the human participants, wherein the electronic action item includes semantic information extracted from the second shared intent.

6. The method of claim 5, further comprising associating the shared intent with a different one of the human participants in the dialog session, and creating an electronic action item for the different one of the human participants, wherein the electronic action item includes semantic information extracted from the shared intent.

7. The method of claim 1, wherein the current intent is a partial intent, and the method further comprises creating the shared intent by combining the partial intent with the selected intent.

8. The method of claim 1, wherein the selected intent is a partial intent, the current intent is a partial intent, and the method further comprises creating the shared intent by combining the selected intent with the current intent.

9. The method of claim 1, wherein the selected intent is a complete intent, the current intent is a partial intent, and the method further comprises creating the shared intent by combining the selected intent with the current intent.

10. The method of claim 1, wherein the current intent is a complete intent, the selected intent is a partial intent, and the method further comprises creating the shared intent by combining the selected intent with the current intent.

11. A computing system operable as a virtual personal assistant (VPA), the VPA comprising:

one or more computing devices including instructions embodied in one or more non-transitory machine-readable storage media, wherein the instructions are executable by the one or more computing devices to cause the one or more computing devices to;

by a microphone, receive a plurality of instances of spoken natural language dialog input supplied by a plurality of human participants in a multi-topic dialog session;

for one or more instances of the instances of spoken natural language dialog input, determine a domain model to apply to the dialog input;

extract semantic information from the one or more instances of spoken natural language dialog input;

develop one or more intents based on the determined domain model, the one or more intents comprising a structured representation of the extracted semantic information;

select an intent of the one or more intents based on an intent mapping, the intent mapping defining a relationship between the selected intent and a current intent, the current intent comprising a structured representation of semantic information extracted from a current instance of dialog input supplied by one of the human participants in the dialog session;

combine a portion of the selected intent with a portion of the current intent;

wherein the selected intent is selected based at least partly on a temporal proximity relative to the occurrence of the current intent.

12. The computing system of claim 11, wherein the processor-executable instructions comprise a VPA engine to create an electronic action item on the computing system in response to the shared intent, and wherein the electronic action item comprises semantic information extracted from the shared intent.

13. The computing system of claim 11, wherein at least one of the selected intent and the current intent is not an actionable intent, and the shared dialog understanding module is to combine the portion of the selected intent with the portion of the current intent to create the shared intent as an actionable intent in response to which the VPA can execute an action.

14. The computing system of claim 11, wherein the selected intent is a complete intent, the current intent is a partial intent, and the method comprises creating the shared intent by combining the complete intent with the current intent.

15. The computing system of claim 11, wherein the shared dialog understanding module is to select an intent having the same topic as the current intent.

16. The computing system of claim 11, wherein the intent mapping defines a relationship between a parameter of the current intent and a parameter of the selected intent.

17. The computing system of claim 11, wherein the processor-executable instructions comprise a VPA engine to present a first electronic action item on a first computing device of the computing and to present a second electronic action item on a second computing device of the computing system.

18. The computing system of claim 11, wherein the VPA engine is to generate the first electronic action item in response to a first shared intent, the VPA engine is to generate the second electronic action item in response to a second shared intent, and the first shared intent is different than the second shared intent.

19. The computing system of claim 11, wherein the VPA engine is to generate an electronic message in response to the shared intent and present the electronic message at a computing device of the computing system.