Skip to Main content Skip to Navigation
Theses

Interaction affective et expressive. Compagnon artificiel-humain

Abstract : The aim of this thesis is to make an Embodied Conversational Agent (ECA) sincere in order to, on one hand, improve its believability from the human's point of view, and on the other hand make it acceptable in an affective relationship between an artificial companion and a human. The first part of this work consists in creating a Multimodal Conversation Language (MCL) for an ECA, made up of Multimodal Conversation Acts (MCA) such as promise, apologise or demand. These MCA allow the agent to appear believable in its sincerity, by multimodally expressing its mental states in an appropriate way. The mental states carried by the MCA are formalised in logics: our will to represent mental states stemming from complex forms of reasoning (based on counterfactual reasoning or on the agent's norms and goals) that are mainly expressed via language (Oatley 1987) led us to design the BIGRE model (Beliefs, Ideals, Goals, Responsibility, Emotions). This model, based on a BDI-like logic (Belief, Desire, Intention), allows us to also represent some particular emotions that we call complex emotions, such as rejoicing, gratitude or regret. The expressive category of MCA contains acts that express these complex emotions (e.g. thank, regret): they are expressed by the language, along with the other appropriate multimodal expressions. We implemented the MCL in the ECA Greta, which enabled an evaluation of this language in terms of believability and sincerity perceived by the human. The second part of this work is about the ECA's reasoning capabilities: in order to allow the agent to reason in the dialogue, that is to update its mental states and its emotions and select its communicative intention, we designed a reasoning engine. This reasoning engine is based on the BDI behaviour cycle of Perception - Decision - Action and on the operators from the BIGRE model, thus enabling the manipulation of mental states resulting from complex reasoning (including complex emotions). The MCA in our language are part of our reasoning engine, and are used to achieve the ECA's communicative intention: for example if the ECA intends to express its gratitude, it builds a plan to achieve this intention, that consists of the MCA thank or congratulate, depending on the intensity of the emotion. One type of communicative intention, triggered by discourse obligations rules, participates in the local regulation of dialogue. Further, since the ECA is emotional, its sincerity brings it to express all its emotions. The generic character of this reasoning engine allowed us to implement it in the ECA Greta (where it is linked with the MCL) as well as in the agent MARC. The multimodal expression of MCA by the agent MARC was made possible by integrating Scherer's checks in the reasoning engine that we adapted to the context of dialogue. An evaluation of the reasoning engine with the agent MARC shows that the mental states deduced by the engine are appropriate to the situation, and that their expression (the expression of the agent's sincerity) is also appropriate.
Document type :
Theses
Complete list of metadatas

Cited literature [118 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00780213
Contributor : Abes Star :  Contact
Submitted on : Wednesday, January 23, 2013 - 3:07:39 PM
Last modification on : Thursday, November 19, 2020 - 12:59:57 PM
Long-term archiving on: : Wednesday, April 24, 2013 - 3:56:05 AM

File

30942_RIVIERE_2012_archivage.p...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00780213, version 1

Collections

STAR | CNRS | LIG | UGA

Citation

Jérémy Riviere. Interaction affective et expressive. Compagnon artificiel-humain. Autre [cs.OH]. Université de Grenoble, 2012. Français. ⟨NNT : 2012GRENM045⟩. ⟨tel-00780213⟩

Share

Metrics

Record views

1276

Files downloads

1257