ACCEPTABILITY IN INTERACTION

From Robots to Embodied Conversational Agents

Sylvie Pesty

UMR 5217 - Laboratory LIG, University of Grenoble, Grenoble, France

Dominique Duhaut

Valoria – University South-Britanny, Vannes, France

Keywords: Embodied Conversational Agents, Acceptability, Companion robot, Emotions, and personality.

Abstract: This paper will propose to increase the traditional definition of system acceptability to take into account the

heavy interaction between a man and a real or virtual system. We introduce the notion of “socially credible”

which describes the need of emotions and personality in intelligent environment. This notion is measured by

experiments using a plush robot Emi and a virtual embodied conversational agent.

1 INTRODUCTION

The rapid development of modern information

technologies, information systems and intelligent

environment brings both opportunities and

challenges to contemporary organizations. One of

the major problems of these technologies is: what

are the different factors that impact the user

acceptance? Indeed what ever are the qualities of the

intelligent environment, if it is not used because it is

perceived by the user as non useful, then the

developed system misses its goal.

In this paper, we want to focus on the factors

usually retrain to approach the problem of

technology acceptance. Has we will show, they are

not longer enough if we look to systems having a

heavy interaction with human. By heavy interaction

we mean, interaction with an intelligent environment

which changes the quality of the human life: for

instance at home for disabled people. In this case the

problem is not to perform a task but to “feel good”.

This interaction introduces a new dimension which it

not any more functional but psychological. The

problem then is to build the intelligent environment

tp become “socially credible”.

In this paper we will introduce the traditional

parameters of acceptance: utility, usability and social

acceptance. These parameters are enough to model

the acceptance of systems used for a short time but

we will explain why it is necessary to introduce a

new parameter. In a second part we will propose our

implementation of the social credibility based on

knowledge and values representation, and on

interaction based on emotion and personality.

Finally we will report some preliminary experiments

on a robot and ECA to illustrate the approach.

2 INFORMATION

TECHNOLOGY ACCEPTANCE

In this first section we want to introduce the major

models of acceptance technology and discuss their

limits.

2.1 Technology Acceptance Model

The Technology Acceptance Model, first introduced

by Davis in 1986, proposes that applications usage

and adoption can be predicted based upon the factors

of perceived ease of use and perceived usefulness

Davis in 1989.

Based on this first work Nielsen in 1993

proposes the model of acceptability (figure 1) and

expresses that “User interfaces are now a much more

important part of computers then they used to be”.

In this model we can see that the fundamental

points are utility and usability which implied that the

intelligent environment in which the human is living

must be well done from a technical point of view. In

365

Pesty S. and Duhaut D. (2011).

ACCEPTABILITY IN INTERACTION - From Robots to Embodied Conversational Agents.

In Proceedings of the International Conference on Computer Graphics Theory and Applications, pages 365-370

DOI: 10.5220/0003405703650370

 SciTePress

robotics or virtual environment this means that the

system must provide functions that are efficient and

sure

Figure 1: J. Nielsen 93 Acceptability model.

In this model the box “social acceptability” is

mentioned but not really developed..

2.2 Theory of Planned Behavior

In his work Ajzek in 1985, introduces the theory of

planned behavior from which a description social

acceptability can be derived (figure 2).

Figure 2: I. Ajzek theory of planned behaviour.

Venkatesh in 2003 proposes to unified Unified

Theory of Acceptance and Use of Technolog

(UTAUT). His work is based on eight models

reviewed are the theory of reasoned action, the

technology acceptance model, the motivational

model, the theory of planned behavior, a model

combining the technology acceptance model and the

theory of planned behavior, the model of PC

utilization, the innovation diffusion theory, and the

social cognitive theory.

Here again there is limit in this approach and this

limit is that the model does not take into account the

heavy interaction that human have with the

intelligent environment including a companion

robot.

2.3 Socially Credible

If a system must be credible to interact with in the

day life, it is necessary to take into account the

human specific dimension: emotion and personality.

This means that we cannot have a single device used

with success by everybody because each of us is

different.

Figure 3: Human identity and feelings.

It is then necessary to deal with the personal

identity and the history of the person which is

already described in works on social psychology but

also to build an intelligent system which can be

perceived like a social creature having its own

identity and personality. We will propose some way

to implement this and some preliminary

experimentations will be presented in the next

sections.

2.4 Acceptability in Interaction

Based on the three previous approaches it is possible

to summarize it in one final model in which we add

the socially credible factor in figure 4.

Figure 4: Full model for human acceptability for systems

with interaction.

Introducing this in the model we have to think about:

how can we realize such credible social interaction.

The next section will expose the four main

parameters used in our projects.

GRAPP 2011 - International Conference on Computer Graphics Theory and Applications

366

3 BUILDING SOCIAL

CREDIBILITY

If a system is socially credible then we can interact

with, having the feeling that it is a kind of “human

species”. For this, the system must share we the

human some knowledge and values and must be able

to express things with personality and emotion.

3.1 Sharing Knowledge

The knowledge shared with the human is divided in

three levels.

Figure 5: Three kind of knowledge

The first level is connected to a standard

knowledge. This is the first data base to develop and

this is currently an active research part. Moreover,

this level often generates an action on the

environment from the robot. This part of the

knowledge is implemented in all the personal robots

and is used to understand the meaning of everyday

sentences.

The second level requires the robot to search for

information to understand the sentence. This means

that it must have access to social knowledge and in

this case, the internet is a good means for carrying

out this type of research. The difficult part here is to

filter all the information to obtain what is relevant.

The third level requires that the robot learn the

personal history of the human from the human. This

is possible only by a dynamic acquisition of

information. The sources of this information can be

multiple: family, doctor, friend, neighbor. This

personal knowledge is the fundamental issue to be in

empathy with someone.

3.2 Sharing Values

Working on the semantic of the sentences is not

enough to provide a social interaction. For instance

if the robot tells the human: “You are right and I

agree with you” then in the impact of this sentence

on the human we can distinguish different levels.

At a first level the robot expresses that it agrees

with the action performed “because it is the right

action at the right time at the right place”. But this

does not build empathy it is just sympathy, a good

friend approving your action.

At the second level the robot expresses empathy

by saying that it feels the same emotion of the

human and that it is sharing it. Here the robot shares

the emotion associated to the action performed by

the human. This second level is empathy but is very

limited in the time, it is just focused on the local

action.

At a third level the same sentences can mean that

the robot agrees on the fact the human is reliable,

offering security “because it is the right action at the

right time at the right place”, and the human s

always like this. This means that the robot feels

confident with the human. In this case the empathy

is enduring. The reason is that the underlying

message is sharing values with the human.

To build this kind of interaction, to each kind of

knowledge standard, social, personal we must build

“meta semantic information” where a list of values

must be coded.

What Kind of Values can we Retrain?

In D.L. Liedner & all 2006 looks for the linkage

between information technology acceptance and

culture. For this purpose they show that some

cultural value can be retrained and we think that they

are the good departure point to code the knowledge

has described previously.

Those values are:

Equality

Progressivism

Community

Sympathy

Emotionality

Optimism

Freedom

Superiority of culture

Deterministic

Objective

Neutrality

Progress

Adventurous

Glamorous

Subordination

Conservatism

Isolation

Antipathy

Sensibility

Pessimism

Enslavement

Inferiority of culture

Uncertainty

Subjective

Partiality

Retreat

Routine

Dull

ACCEPTABILITY IN INTERACTION - From Robots to Embodied Conversational Agents

367

Known

Order

Friend

Mythical

Rationality

Virtuosity

Aesthetic

Unknown

Chaos

Enemy

Factual

Subjectivity

Basic needs

Practical

Having this list of values the idea is to code all

the words of the used language between the human

and the robot by and static value or a function on the

model used to code emotions Le Tallec & all 2009.

3.3 Emotion in Communication

In the EmotiRob project we already coded emotions

in sentences and we believe that on the same ideas it

might be possible to code values defined in the

previous section.

Coding Emotions in a Sentence

The language is coded has follow Le Tallec & all

2009:

<noun> coded by a static emotional number « à

priori »

Wolf = -2 very negative emotion

Mother = +2 very positive emotion

<adjective> coded by a function

Pretty = X → X + 1 this adjective increases the

positive number of the associated noun

<verbe> coded by a function

Break(X,Y) = (X,Y) → –Y breaking something

inverse the broken thing

Have(X,Y) = (X,Y) → X*Y the friends of friends

are friends …

With this kind of knowledge representation it has

been possible to test 178 sentences manually coded

by 5 different persons. The response of the system is

more than 90% in concord with the reference

sentences.

The idea behind the language is to allow

expressive communication between not only

software agents but also between human and agents.

This language takes into consideration aspects such

expectations, conditions of success, among other

characteristics that are present in human

communication. In Berger 2005, the conditions of

success and satisfaction are explicitly defined as

well as the elements from the conversational

background. The thirty two formalized conversation

acts are:

Assertive: confirm, deny, think, say, remember,

inform and contradict;

Commissives: commit oneself, promise, guarantee,

accept, refuse, renounce and give;

Directives: request, ask a question, suggest, advise,

require, command and forbid;

Declaratives: declare, approve, withdraw, cancel;

Expressives: thank, apologize, congratulate,

compliment, complain, protest, greet.

Connected to this five class of behaviour Berger

2006 define some logical rules to know when one of

them as to be applied.

3.4 Personality in Communication

The personality is expressed in the computation of

the emotions. So a computational model of emotion

is developed in Dang in 2008.

Definition of emotion: an emotion is the process

that characterizes the human body’s response to a

stimulus or event.

By stimulus or event we mean: external changes

in the environment of the body, absence of external

changes in the environment although one expected,

and internal body changes.

By human body response we mean:

physiological changes inside the body, external

expressions of the body and also … no change.

Based on this definition we propose the

following model in Fig.6.

In this model Sensation is the basic starting

point. The sensation is generated by an event,

something which really exists or not, but which

generates a physiological change in the body and/or

by sending subjective information (from Intuition) to

the sense-organs: touch, hear, see … This sensation

will be processed in two ways.

First, the Physiological Interpretation will

directly interpret this initial signal into a body

reaction (the heart races …) and will also alert the

module Behaviour.

On the other hand, the Cognitive Interpretation

will interpret the signals received from Sensation

into cognitive information about the environment

situation.

The Behaviour will then calculate the response

from the information coming for the perceptions

based on the Internal Cognitive State. This response

is sent to the Body where the physical reaction will

take place.

The MBTI model of personality, proposes four

categories to build personality. Our model

completely covers these four categories. The first

one is the attitude splited in Extraversion (E) or

Introversion (I). In the generic model this particular

feature is integrated in the Mood and Behaviour

modules. Secondly, perception category of the

MBTI is completely covered by the generic

architecture.

GRAPP 2011 - International Conference on Computer Graphics Theory and Applications

368

Figure 6: The Grace model of emotion.

The Sensing is constructed with the two

Interpretation modules and the Intuition by the

Intuition module. Then, the third category is

decisions: Thinking (T) or Feeling (F). We cover

these two approaches by the way that the Behaviour

module of the generic model is coded. Last category

of MBTI - Judging (J) or Perceiving (P) can be

coded at the Interpretation level. In fact, it is a level

of interest for the sensation that will be used. For

instance, a sensation directly concerning a person

will be more interesting for someone who is

Perceiving.

4 EXPERIMENT

In this part we want to report on the introduction of a

part of the socially acceptable parameters defined

previously, in real systems : one robot and one ECA.

4.1 On Emi Robot

The Emi robot is a fully autonomous robot designed

in the ANR EmotiRob project to build emotion

interaction with children.

A full description of the robot and its realization

can be find in Saint-Aimé 2010. It is a 10 degree of

freedom robot : 2 for the eyesbrow, 4 for the mouth,

2 for the head and 2 for the body (pan-till).

In this first experiment we want to measure the

recognition of emotion on the Emi robot (figure 7).

For this experiment, children around 10 years old are

the human users which, for a series of sentences

given, will determine the emotion expressed by the

robot. We used the Wizard of Oz technique to make

the child believe that EmI it understanding the

dialog.

Saint-Aimé 2010 shows that the emotion recognition

is very good and that the behavior of the robot is

credible. One question is not solved behind this

result is what is the part is the human

Figure 7: Emi robot and simulator.

believes in this result. Indeed the human can see in

the robot what he is expecting even is it is not the

real intention of the robot.

4.2 On ECA

The general goal of researchers in the field of

Embodied Conversational Agents (ECAs) is to

develop interactive systems that are more natural

and easy to use, closer to the human user. ECAs

must be credible or “believable”, the most general of

these terms, used to describe anything we accept as

true, even in the absence of absolute proof. Ortony

2003 said that a major question is how to make an

agent a believable agent. Bates 1993 explained the

crucial role of emotion in believable agent. Thus,

ECAs must be endowed with refined communicative

capabilities and the challenge is to build ECAs,

which are capable to reason about emotions, to

predict and understand human emotions, and to

process emotions in reasoning and during the

interaction with a human user.

In the ANR CECIL project (Complex Emotions

in Communication, Interaction, and Language), we

endowed an ECA with the capabilities to express its

emotions by means of different modalities including

facial expressions, gestures, and language. We

merged speech act theory, emotion theory

, and logic.

We used the logic in order to provide a

systematic analysis of expressive speech acts, that is,

speech acts that are aimed at expressing a given

emotion (e.g. to apologize, to thank, to reproach, to

rejoice, to regret, to deplore, etc.).

A description of

the expressive speech acts can be found in Guiraud

ACCEPTABILITY IN INTERACTION - From Robots to Embodied Conversational Agents

369

2011. We put forward a Multimodal Conversation

Language that enables agents to form expressive

dialogues, mainly deliberative dialogue such as

negotiation, advice seeking, bargaining and setting

up appointments. Part of the thirty-two expressive

conversation acts described in Berger 2005 has been

used to constitute this language. We are currently

experimenting the language on small dialogue

scenarios between the Greta ECA (figure 8) and a

human user. A preliminary evaluation of this

experiment shows that the agent communicates its

emotions during the interaction and tends to be

“credible” with regard to the human user.

Figure 8: Greta ECA see Guiraud 2011.

5 CONCLUSIONS

In this paper we propose to discuss social credibility

for human acceptance of intelligent environment.

We propose to implement this with three different

levels of shared knowledge, values representation in

knowledge and with emotion and personality

expression in communication. We report on some

preliminary experimentation in this field.

ACKNOWLEDGEMENTS

A part of this work has been realized with the

support of the French ANR agency under the Psirob

project and the CECIL project.

REFERENCES

Ajzen, I., 1985. From Intention to Actions: A Theory of

Planned Behaviour. In:Kuhl, J., Beckman, J. (Eds.),

Action Control: From Cognitions to Behaviour.

Springer-Verlag, New York, pp. 11–39.

Bates J., 1994. The role of emotion in believable agents.

Communications of the ACM, 37(7):122–125.

Berger A., Pesty S., 2005. Towards a Conversational

Language for Artificial Agents in Mixed Community.

In Fourth International Central and Eastern European

Conference on Multi-Agent Systems CEEMAS'05,

Springer Verlag, Budapest, pages 31-40.

Davis, F. D., 1986. A technology Acceptance Model for

empirically testing new end-user information systems:

theory and results. Phd dissertation, MIT Sloan School

of Management, Cambridge.

Davis, F. D., 1989. Perceived usefulness, perceived ease

of use, and user acceptance of information technology.

MIS Quarterly 13, 319–339.

Dang, T. Letellier-Zarshenas,S., Duhaut, D. 2008

Comparison of recent architectures of emotions. In

10th International Conference on Control,

Automation, Robotics and Vision – ICARCV 2008.

Guiraud N., Lorini E., Longin D., Rivière J., Pesty S.,

2011. The face of emotions: a logical formalization of

expressive speech acts. Proceedings of the Tenth

International Conference on Autonomous Agents and

Multiagent Systems, AAMAS (in press)

Leidner D. L. & all Review: A review of culture in

information systems Mis Quarterly vol 30 N°2 june

2006

Le Tallec M., Villaneau J., Antoine J-Y., Savary A.,

Syssau-Vaccarella A, Emologus - a compostional

model of emotion detection based on the

propositionnal content of spoken utterances,

Proceedings Text, Speech and Dialogue, TSD'2010,

Brno, 2010.

Nielsen, J., 1993. Usability Engineering. Academic Press,

New-York.

Ortony, A., 2003. On making believable emotional agents

believable. In R. Trappl, P. Petta & S. Payr (Eds.),

Emotions in humans and artifacts. Cambridge, MA.:

MIT Press

Poggi I., Pelachaud C., De Rosis F., V. Carofiglio V., De

Carolis B., 2005. Greta: A believable embodied

conversational agent. in O. Stock and M. Zancarano,

eds, Multimodal Intelligent Information Presentation,

Kluwer, 2005

Saint-Aimé, S. Conception et réalisation d’un robot

compagnon expressif basé sur un modèle calculatoire

des emotions, Phd university Bretagne-Sud 2010

V. Venkatesh, M. Morris & all, September 2003 User

Acceptance of information technology: toward a

unified view MIS Quarterly Vol. 27 No. 3, pp. 425-

478.

GRAPP 2011 - International Conference on Computer Graphics Theory and Applications

370