ERDA : An Empathic Rational Dialog Agent1 Magalie Ochs (1),(2), Catherine Pelachaud (1) and David Sadek (2) (1) IUT de Montreuil, University Paris VIII, France (2) France Telecom R&D, France
ERDA : An Empathic Rational Dialog Agent2 Overview Motivation and Objectives Rational Dialog Agent Corpora Description A Coding Scheme for Annotating Emotion Elicitation Results of the Annotation Perspectives and Future Work
ERDA : An Empathic Rational Dialog Agent3 Motivation and Objectives (1/3) Expressions of empathic emotions enhance human- machine interaction [Brave et al., 2005] [Klein et al., 1999] [Partala et al., 2004][Prendinger et al., 2005] Empathy: capacity to "put your-self in someone else's shoes" [Poggi, 2004]
ERDA : An Empathic Rational Dialog Agent4 Motivation and Objectives (2/3) Empathic Agent Simulate the process of user's emotion elicitation User's emotion Expression of the emotion To know the circumstances under which user’s emotions are triggered
ERDA : An Empathic Rational Dialog Agent5 Motivation and Objectives (3/3) Objective: To identify under which circumstances a dialog agent should express empathic emotions To identify under which circumstances a user may potentially feel emotion when interacting with a dialog agent Method proposed: To analyze real human-machine dialogs that lead a user to express emotion
ERDA : An Empathic Rational Dialog Agent6 Rational Dialog Agent ) Model of agent based on a formal theory of Interaction (Rational Interaction Theory) BDI-approach: Agent's mental state composed of Beliefs and Intentions. Model of Communicative Acts: Based on speech act theory and BDI approach Description of feasibility preconditions and rational effects in terms of mental attitudes (belief and intention). [Sadek, 1991]
ERDA : An Empathic Rational Dialog Agent7 Corpora Description Real vocal human-machine dialogs corpora Dialogs derived from two vocal applications developed at France Telecom: PlanResto & Dialogue Bourse Users interact in natural language with rational dialog agent Dialogs annotated by two labels of emotion (positive and negative) (two annotators)
ERDA : An Empathic Rational Dialog Agent8 A Coding Scheme for Annotating Emotion Elicitation (1/5) To analyze the features of the dialogs that lead a user to express emotion A coding scheme to annotate the sequence of events that lead to user's expression of emotion Tags of the coding scheme Relevant features of an emotion eliciting-event Computational description
ERDA : An Empathic Rational Dialog Agent9 A Coding Scheme for Annotating Emotion Elicitation (2/5) Consequences of an event on a user's goal: Goal achievement Goal failure Causes of goal failure: Unfeasibility by the agent of the user's goal Belief conflict on the user's goal Goal conflict Transfer of information on the feasibility of the goal Tags of the coding scheme
ERDA : An Empathic Rational Dialog Agent10 Description of tags in terms of mental attitudes (Beliefs and Intentions) Sequence of tags – Condition of Emotion Elicitation Particular mental state of a rational dialog agent Example : Belief conflict on a user's goal belief _conflict user (event, goal, agent) = def K user (Done(event, I user (goal) K agent (I user (goal))) A Coding Scheme for Annotating Emotion Elicitation (3/5)
ERDA : An Empathic Rational Dialog Agent11 A Coding Scheme for Annotating Emotion Elicitation (4/5) Inference of User's Mental State (based on Speech Act Theory): Goals of the user by performing a communicative act: To achieve its intentional and perlocutory effects Belief of the user: the enunciation of the communicative act enables to realize these goals Annotation of 20 dialogs (Intra-coder validity: re- annotation 1 month after) [Austin, 1962] [Searle, 1969]
ERDA : An Empathic Rational Dialog Agent12 A Coding Scheme for Annotating Emotion Elicitation (5/5) An example of annotated dialog extracted from PlanResto […] Agent : What are you looking for? User: the Hippopotamus restaurant mmh in 14th urban district (event1) /* User's goal : G1 : Perlocutory effect : to know an hippopotamus restaurant in the 14th urban district G2 : Intentional effect : the agent knows the user's goal G1 */ Agent: A restaurant in 14th urban district. I have found more than a hundred. You can precise a speciality, for example specific type of food or budget […] User : mmh Agent: A Greek restaurant. […] EXPRESSION OF A NEGATIVE EMOTION
ERDA : An Empathic Rational Dialog Agent13 Results of the Annotation (1/2) Analyze 47 situations that lead a user to express emotion Main circumstance under which negative emotion is expressed: Goal failure (45 situations) Different causes of goal failure: Belief Conflict (30 situations) Goal Conflict (1 situation) Transfer of information on the feasibility of the goal (5 situations)
ERDA : An Empathic Rational Dialog Agent14 Results of the Annotation (2/2) Several successive goal failures required to trigger user's expression of emotion (in average 2) Number of goal failures depends on: The cause of the failure Goal conflict : 1 goal failure Belief conflict : 2 goal failures in average Transfer of information on the feasibility : 1 goal failure in average If an emotion has already been elicited Other conditions of user's expression of negative emotion: Belief Conflict (2 cases) Goal Conflict (3 cases)
ERDA : An Empathic Rational Dialog Agent15 Perspectives and Future Work Analyze of non-acted human-machine dialogs that lead user to express emotions Identification of several circumstances under which a user potentially feels negative emotion Mental states of Conflicts Future Works: Subjective evaluation to validate the believability of the conditions of emotion elicitation identified