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
Copyright
c
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