MULTI-MODAL MODELING OF TEMPORAL INTERACTION SEQUENCES

• US 20140212854A1
• Filed: 01/31/2013
• Published: 07/31/2014
• Est. Priority Date: 01/31/2013
• Status: Active Application

• Current Assignee: SRI International, Inc.
• Original Assignee: SRI International, Inc.
• Inventors: Divakaran, Ajay, Siddiquie, Behjat, Khan, Saad, Lubin, Jeffrey, Sawhney, Harpreet S.
• Application Number:
  US13/755,783
• Publication Number:
  US 20140212854A1
• Time in Patent Office: Days
• US Class Current: 434/236
• CPC Class Codes: G09B 19/00 Teaching not covered by other main groups of this subclass teaching or practice apparatus for gun-aiming or gun-laying F41G3/26

Abstract

A multi-modal interaction modeling system can model a number of different aspects of a human interaction across one or more temporal interaction sequences. Some versions of the system can generate assessments of the nature or quality of the interaction or portions thereof, which can be used to, among other things, provide assistance to one or more of the participants in the interaction.

Claims

1. A method for predicting a behavioral event to occur in an interaction involving at least two participants, at least one of the participants being a person, the method comprising, with a computing system:

detecting, from multi-modal data captured by one or more sensing devices, a plurality of different behavioral cues expressed by the participants during the interaction;

recognizing a plurality of temporal interaction sequences, each of the temporal interaction sequences occurring over a time interval during the interaction, and involving a pattern of the behavioral cues, and at least two of the participants being involved in at least one of the temporal interaction sequences;

determining the nature of the interaction based on the recognized temporal interaction sequences; and

predicting the behavioral event to occur in the interaction based on the determined nature of the interaction.

2. The method of claim 1, wherein the predicted event comprises a change in the nature of the interaction.

3. The method of claim 1, wherein the predicted event comprises a change in the emotional state of at least one of the participants during the interaction.

4. The method of claim 1, wherein the plurality of different behavioral cues comprises verbal content and non-verbal cues.

5. The method of claim 1, comprising communicating a suggestion relating to the predicted event to one or more of the participants.

6. The method of claim 5, comprising communicating the suggestion during the interaction.

7. An interaction assistant embodied in one or more machine-readable storage media and accessible by a computing device to assist with an interaction involving a natural-language dialog between a person and the computing device, by:

detecting, from multi-modal data captured by at least one sensing device, a plurality of different behavioral cues expressed by the person during the natural-language dialog;

recognizing a temporal interaction sequence comprising a pattern of the behavioral cues occurring over a time interval during the natural-language dialog;

deriving, from the temporal interaction sequence, an assessment of a portion of the natural-language dialog involving the person; and

formulating a portion of the natural-language dialog involving the computing device based on the assessment.

8. The method of claim 7, wherein the computing device comprises a mobile computing device and the method comprises detecting the behavioral cues using one or more sensing devices of the mobile computing device.

9. The method of claim 7, comprising recognizing a plurality of temporal interaction sequences occurring over different time intervals, and deriving the assessment from the plurality of temporal interaction sequences.

10. The method of claim 9, wherein at least two of the different time intervals are defined by different time scales.

11. The method of claim 7, wherein the plurality of different behavioral cues comprises one or more non-verbal cues relating to one or more of:

a gesture, a body pose, a head pose, an eye gaze, a facial expression, a voice tone, a voice loudness, and another non-verbal expression.

12. The method of claim 11, comprising detecting one or more verbal behavioral cues from the multi-modal data, semantically analyzing the verbal content of the one or more verbal behavioral cues, and deriving the assessment based on the semantic analysis of the verbal content.

13. The method of claim 12, comprising deriving the assessment based on a combination of the verbal and non-verbal cues.

14. A method for assessing a person'"'"'s emotional state during an interaction involving the person and at least one other participant, the method comprising, with a computing system:

detecting, from multi-modal data captured by one or more sensing devices, a plurality of different behavioral cues expressed by the participants during the interaction;

recognizing a plurality of temporal interaction sequences, each temporal interaction sequence comprising a pattern of the behavioral cues occurring over a time interval during the interaction, and at least one of the temporal interaction sequences involving the person and at least one other participant;

assessing the person'"'"'s emotional state during each of the temporal interaction sequences based on the behavioral cues involved in the temporal interaction sequence;

detecting changes in the participant'"'"'s behavior over a period of time defining the duration of the interaction; and

evaluating the participant'"'"'s behavior over the period of time based on the detected changes in behavior and the time intervals in which they occurred.

15. The method of claim 14, wherein at least some of the temporal interaction sequences have overlapping time intervals.

16. The method of claim 14, wherein the behavioral cues comprise verbal content and non-verbal cues.

17. The method of claim 14, wherein the multi-modal data comprises a stream of audio data and a stream of visual data, and the method comprises extracting the behavioral cues from the multi-modal data and assessing the person'"'"'s emotional state using a graphical model.

18. The method of claim 17, comprising fusing the captured multi-modal data and using a discriminative probabilistic model to recognize the temporal interaction sequences.