An Open Architecture for Affective Traits in a BDI Agent

Bexy Alfonso, Emilio Vivancos and Vicente J. Botti

Department of Informatic Systems and Computing, Universidad Polit´ecnica de Valencia, Valencia, Spain

Keywords:

Agents, Emotions, Personality, Mood, Architecture, BDI, Human Behavior.

Abstract:

Recently an increasing amount of research focuses on improving agents believability by adding affective fea-

tures to the traditional agent-based modeling. This is probably due to the demand of reaching ever more

realistic behaviors on agent-based simulations which extends to several and diverse application ﬁelds. The

present work proposes O3A: an Open Affective Agent Architecture, which extends a traditional BDI agent

architecture improving a practical reasoning with more “human” characteristics. This architecture tries to ad-

dress disperse deﬁnitions combining the main elements of supporting psychological and neurological theories.

1 INTRODUCTION

Artiﬁcial intelligence constantly evolves. New meth-

ods, algorithms and techniques are created or im-

proved in order to achieve more sophisticated solu-

tions. The agents ﬁeld is not far behind. With the vi-

sion of a computational agent as a reactive and proac-

tive entity, with its own goals, desires, sensing and

planning mechanisms, more steps are taken to simu-

late human behavior and human interactions. Never-

theless for a simulation that truly reﬂects how humans

behave, it is necessary to model also the affective side.

Neuroscience methods have found signiﬁcant ev-

idence that emotions are associated to regions in the

brain in charge of controlling the related functions.

They have also demonstrated that these functions are

necessary for the individual because they act as in-

ternal heuristics guiding decisions in uncertain situa-

tions. Psychological and cognitive sciences have also

made important contributions to further research on

emotional computing, considered a “computing that

relates to, arises from, or inﬂuences emotions” (Pi-

card, 1997). One of the challenges to deal with when

addressing issues in emotional computing is to ef-

fectively combine results of several and varied sci-

ences. Speciﬁcally cognitive science has received

special interest by affective computing researchers

due to its suitability for creating computational mod-

els. Among the psychological perspectives of person-

ality and emotion, the cognitive is the most widely

studied because, in some degree, it is contained in

the other perspectives. Results of research in affec-

tive computing have been applied in ﬁelds like educa-

tion, training, therapies and the simulation of disaster

situations.

The aim of this work is to present O3A (an Open

Affective Agent Architecture): general enough to

consider aspects of rational agents as well as their

affective nature and whose components can be cus-

tomized or replaced according to the domain require-

ments. In this article it is shown how this mechanism

of emotion can be integrated into a practical reasoning

architecture. We take the widely accepted BDI (Be-

liefs, Desires and Intentions) architecture of agents as

starting point and we also endow the agent with the

main affectiveconcepts inherited from supporting sci-

ences. We provide a summary of the main concepts

extracted from psychological and neurological litera-

ture that had shed light over our work and brieﬂy com-

ment some signiﬁcant related works. Then we present

our architecture and its main components, pointing

out how it is integrated with a traditional BDI algo-

rithm. Final conclusions provide some annotations of

the work performed.

1.1 Motivation

The BDI agent architecture has been widely ac-

cepted in the agents community because it has impor-

tant advantages compared to other agent architectures

(logic based, reactive, or layered architectures (Weiss,

1999)). It has been able to effectively reﬂect the hu-

man reasoning process having strong philosophical

roots. Besides, many logical and software frame-

works based on the BDI architecture have been de-

veloped. It offers cognitive processes and compo-

320

Alfonso B., Vivancos E. and Botti V..

An Open Architecture for Affective Traits in a BDI Agent.

DOI: 10.5220/0005153603200325

In Proceedings of the International Conference on Evolutionary Computation Theory and Applications (ECTA-2014), pages 320-325

ISBN: 978-989-758-052-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

nents, meaning the processing of perceptions, beliefs,

and goals that are necessary and are directly affected

by the emotional internal state (Castelfranchi, 2000).

That makes this model to become a suitable alterna-

tive in order to represent the practical rational side of

an agent. Although many approaches havetried to im-

prove the BDI architecture with the human emotional

process, they result disperse and confusing on their

deﬁnitions, and also sometimes they don’t follow an

incremental line where one reuses the others results

(Marsella et al., 2010).

According to Castelfranchi (Castelfranchi, 2000)

the basic elements constituting emotions are: beliefs,

evaluations, goals, arousal, and the “tendency towards

action”. Beliefs are individual representations of the

world that activate emotions with a level of arousal,

and motivate the conduct. Then emotions come from

the interpretation of facts and sensations (“recogni-

tion” of emotions). On the other hand individuals

tend to avoid or maybe to pursue some goal if this

leads to a desirable emotional state; in this sense emo-

tions itself can be considered as goals. But also emo-

tions can monitor the goals offering guidance about

the goals consequences and besides they can activate

new goals. Castelfranchi highlighted the difference

between the two kinds of appreciation of events’ eval-

uations: 1) adaptive and non rational, which are auto-

matic, intuitive and unconscious orientations to what

can be wrong or bad, called also primary in the lit-

erature (Ortony and Turner, 1990; Dam´asio, 2005;

Becker-Asano and Wachsmuth, 2010), and 2) declar-

ative or explicit, which is an evaluation based on rea-

soning, can be explained and is closely related to

goals called also secondary. These and other ideas

have laid the ground for structuring O3A.

2 BACKGROUND AND RELATED

WORK

The psychological literature related to affective hu-

man characteristics talks about cognitive concepts

like emotions, moods, feelings, and personality (Fri-

jda, 1987; Castelfranchi, 2000; Ryckman, 2007).

They generally agree in that emotions are reactions

as a consequence of agents, other actions and/or ob-

jects (Ortony et al., 1988). Mood, as emotions, is

considered to be an experiential component too but

it is not necessarily associated with a cause, lasts

longer and has less intensity than emotions (Mehra-

bian, 1997). On the other hand, personality is seen as

a set of individual characteristics which generally in-

ﬂuence motivations and behaviors of the agent (John

and Srivastava, 1999; Ryckman, 2007). Among the

perspectives that address personality and emotions,

the cognitive perspective has special relevance for af-

fective computing due to its suitability to be used in

computational applications. Moreover, in the neuro-

logical ﬁeld we found important works which have

laid the foundations for future applications in artiﬁcial

intelligence and human-computer interaction areas.

LeDoux and Dam´asio made important contributions

in this area (LeDoux, 1998; Dam´asio, 2005). They

found evidences of the relationship between emotions

and the way in which the brain works.

Based on some interdisciplinary works (Mehra-

bian and Russell, 1974; Ortony et al., 1988), sev-

eral approaches have tried to embody agents or virtual

characters with affective traits and expressive func-

tions (Gebhard, 2005; Becker-Asano and Wachsmuth,

2010; Neto and Silva, 2012). Many of them are also

based on the BDI architecture. For example an inter-

esting work is the one proposed in (Parunak et al.,

2006) where authors propose the DETT (Disposi-

tion, Emotion, Trigger, Tendency) model for situated

agents. It is a domain speciﬁc approach that aims to

model agents whose goal is to anticipate the actions of

an enemy in a combat scenario. It proposes a reason-

ing for the agents suitable to perform fast actions, and

it takes the features of the OCC model (Ortony et al.,

1988) for extending a BDI architecture. In this model

emotions inﬂuence perception and analysis. The dis-

position element modulates the appraisal, and so, the

way that emotions are triggered from beliefs. On the

other hand, a tendency is imposed to intentions in that

analysis is modulated by emotions. The analysis pro-

cess together with the agent desires produce the in-

tentions. The EBDI architecture for emotional agents

(Jiang et al., 2007) is another similar work. The au-

thor points out that the way that changes in the en-

vironment affect emotions and how these changes in-

ﬂuence human behavior differ even individually, so

he separates the practical reasoning from the emo-

tion mechanism. But he doesn’t use any psycholog-

ical concept to represent such individual differences.

In his work a distinction between primary and sec-

ondary emotions is made. Primary emotions are con-

sidered reactive responses of the brain and secondary

emotions appear later and can be caused by primary

ones or by more complex chains of thinking. Sim-

ilarly in (Neto and Silva, 2012) an architecture for

emotional agents is presented. This architecture in-

cludes a personality and a mood component and it

describes how affective characteristics inﬂuence per-

ception, motivation, memory and the decision making

in a BDI architecture. Its emotional component uses

the results of the ALMA project (Gebhard, 2005), and

therefore integrates personality, emotions and mood

AnOpenArchitectureforAffectiveTraitsinaBDIAgent

321

components which are modeled through the Five Fac-

tor Model (FFM) of personality (McCrae and John,

1992; Goldberg et al., 1990), the OCC model (Ortony

et al., 1988) and the PAD model (Mehrabian and Rus-

sell, 1974) respectively. This approach focuses on

the cognitive state. The coping actions (Marsella and

Gratch, 2009) are linked to “ﬁlters” that select each

time the percepts or facts that are aligned with the

agent emotional state or plans that doesn’t lead the

agent to an undesirable emotional state. An approach

that integrates the majority of issues of previous ap-

proaches is the one proposed in (Becker-Asano and

Wachsmuth, 2010). The authors propose an archi-

tecture suitable for virtual characters. They include

primary and secondary emotions where the former is

directly linked to expressivecapabilities like facial ex-

pressions and the latter comes as the result of reason-

ing about current events, and by considering expec-

tations and past experiences. In this approach mood

values are in a bipolar scale and move from positiveto

negative. Emotions mix theories from P. Ekman (Ek-

man, 1999) and OCC. This architecture also included

a memory component which is used to generate the

character expectations.

Marsella et. al. present a general computational

appraisal model (Marsella et al., 2010), which tries to

cope with the main issues associated with emotions

and their impact on the cognitive processes and state

of the agent. This architecture is composed of three

elements: the person-environment relationship, the

appraisal variables, and the emotion or affect compo-

nent. These components are linked in such a way that

links have associated a transformation model and the

component in the pointing side of the link needs ele-

ments from the previous component in the link. The

person-environment relationship represents the rela-

tionship between the entities in the social environ-

ment, beliefs, desires or intentions, and the external

events. Other works like (Dastani and Meyer, 2006;

Steunebrink et al., 2012) also propose the syntax and

semantic of a logic-based agent language and the log-

ical formulation of the OCC model of emotions re-

spectively. They specify how external stimuli and the

current state of the agent beliefs and goals may de-

rive in emotions. Speciﬁcally in (Dastani and Meyer,

2006), the authors propose transition rules for the ex-

ecution of actions according the current emotions, so

emotion evaluation, as well as action tendency is de-

scribed through programming constructs. These ap-

proaches do not contradict the proposed architecture.

Conversely they could support some of its compo-

nents (as detailed in section 3.1).

Generally existing approaches offer speciﬁc struc-

tures for particular models of emotions where primary

and secondary emotions are often treated through the

same mechanism. Besides, the concept of a central

core affect (or mood) found in the psychological lit-

erature as well as the individual differences like those

determined by the personality are not always consid-

ered. In section 3 we try to address this issues by

presenting our affectiveagent architecture called O3A

(an Open Affective Agent Architecture).

3 THE O3A ARCHITECTURE

The O3A architecture addresses mental, cognitive and

motivational components of emotions, what, accord-

ing to (Castelfranchi, 2000), makes them a complex,

hybrid subjective state of mind. O3A is based on the

appraisal theory that is currently the most accepted

computational model of emotions. O3A is general be-

cause its components are integrated into a BDI agent-

based architecture aiming to provide a model useful

to build believable agents behaviors in any domain.

3.1 Main Components

The O3A architecture (Figure 1), consists of four

main components in charge of controlling the agents

emotional issues. These components have a well de-

ﬁned interface that allows to easily re-implement any

component if other emotional approach is used in the

future.

The Appraisal Component is in charge of deriving

emotions from perceptions and from the agent state.

Two sub-components make this task. The Emotion

reactive component takes what has been observed in

the environment to obtain the Primary emotions. The

function of this component is based on the idea of

the non-rational, automatic, unconscious and adaptive

evaluation of events stated in (Castelfranchi, 2000),

that derives in primary emotions. In order to imple-

ment this component we have labeled each percept

with a set of “more probable emotions” to be exper-

imented by the agent after perceiving the event. Our

taxonomy of emotions is based on the cognitive per-

spective of emotions proposed by Clore & Collins in

1988 (Ortony et al., 1988), that was later improved by

Ortony in 2003 (Ortony, 2003). This model of emo-

tions, popularly know as OCC, consist of 22 emotions

types with their speciﬁcations and attributes. These

emotions are linked to eliciting conditions, and their

intensity is affected by a set of variables. The emo-

tion deliberative component is in charge of deriving

Secondary emotions, and corresponds to the declara-

tive and explicit evaluation of events that can be ex-

ECTA2014-InternationalConferenceonEvolutionaryComputationTheoryandApplications

322

plained and argumented upon (Castelfranchi, 2000).

As these emotions are the result of more complex

chains of thinking it is necessary to check current be-

liefs, options available, as well as information from

previous results (e.g. how successful or ineffective

has been an action or plan, or the emotional experi-

ence after a plan in previous executions). We use the

ideas proposed in (Marsella and Gratch, 2009) for this

component, where emotions emerge from evaluating

appraisal variables of propositions. A work like (Ste-

unebrink et al., 2012) is also suitable to be used in this

component of the architecture. The authors make log-

ical formalizations to derive emotions starting from

the agent mental attitudes.

The Beliefs Component determines how Current

Mood inﬂuences the percepts before they become the

agent’s Beliefs. It starts from the idea that mood can

intensify or blur perceptions and hence generate dif-

ferent perceptions for each agent (Neto and Silva,

2012; Niedenthal and Setterlund, 1994).

The Mood Component feeds on the agents Person-

ality to establish the agent’s initial mood and to update

the Current Mood. This is based on the idea that indi-

viduals differentiate from each other in the way their

mood changes depending on their personality traits.

An explosive individual may reach a mood with high

levels of arousal more easily than one that has a less

neurotic personality. The Mood component updates

the Current Mood also considering the primary and

secondary emotions elicited. We use the dimensional

representation of the core affect made by A. Mehra-

bian and A. Russell (Mehrabian and Russell, 1974;

Russell and Mehrabian, 1977) in order to describe

mood. A three dimensional space whose dimensions

are Pleassure, Arousal and Dominance (PAD) de-

scribes any emotional state of the agent. This com-

ponent also deals with the concurrence of “evaluation

and appraisal about the same entity/event”, which can

“give rise to convergence and enhancement of the va-

lence, or to conﬂicts” (Castelfranchi, 2000). Another

issue that is addressed in this component is the du-

ration and the return to an “equilibrium” state of the

mood.

The Coping Component decides if the changes ex-

perimented in the current mood deserve to take ac-

tions in the cognitive processes of the agent, deter-

mining in this case the way intentions are selected to

be achieved. As stated in (Castelfranchi, 2000) the

responses of the agents to external events having the

same knowledge, desires, and abilities will be differ-

ent depending on each agent internal state. We use

a plans prioritization strategy but elements of previ-

Personality

Current mood

Emotion

reactive

Mood

component

Coping

component

Results

Beliefs

Percepts

influenced

by mood

Emotion

deliverative

Appraisal

Intentions

Options

Beliefs

component

Primary

emotions

Secondary

emotions

Figure 1: Main components of the O3A architecture.

ous works can also be used, such as the programming

constructs offered in (Dastani and Meyer, 2006), that

offer transition rules for the execution of actions ac-

cording to the current emotions

Just as the mood can be modeled through a di-

mensional representation, there are models like the

Five Factor Model of personality (McCrae and John,

1992) that are useful to build Personality proﬁles.

This model uses a set of ﬁve dimensions to describe

each individual personality quite accurately.

3.2 Integration into a BDI Architecture

Starting from the control cycle of a BDI practical rea-

soning agent offered in (Bordini et al., 2007), we pro-

pose the main cycle for agents under O3A. The algo-

rithm of Figure 2 integrates a traditional BDI agent

architecture with the emotional components proposed

in O3A. Note that this architecture is independent of

the internal implementation of these emotional com-

ponents that can be substituted in the future. In lines

1-3 the initial values for Beliefs (B

), Intentions (I

)

and Current Mood (M

) are respectively set. In par-

ticular, Current Mood is initialized according to the

Personality proﬁle (P). We use the proposal made

in (Mehrabian, 1996) to establish the correlation be-

tween the PAD space (Mehrabian and Russell, 1974)

and the Five Factor Model (McCrae and John, 1992).

Lines 4-31 represent the basic control loop. The main

actions performed in this loop are: observe, execute,

and update options. In line 5 the next percept (ρ)

is observed from the environment. This percept may

have associated a set of the “more probable emotions”

to experiment with this percept. In line 6 Beliefs are

updated considering the agent’s current Beliefs (B),

the new percept (ρ), and the Current Mood (M). In

this algorithm, Desires are considered “candidate op-

tions”, and are determined in line 7 on the basis of

the current Beliefs and Intentions. In lines 8 and 9

This is valid only if, according to the requirements of the do-

main, a centralized processing of a “mood” is not considered and

the emotional internal state is represented directly by the emotions

“evaluated”.

AnOpenArchitectureforAffectiveTraitsinaBDIAgent

323

1: B ← B

; {B

are initial beliefs}

2: I ← I

; {I

are initial beliefs}

3: M

= M = initialize

mood(P); {P: personality}

4: while (true) do

5: get next percept ρ via sensors;

6: B ← get

new beliefs(B, ρ, M);

7: D ← get

options(B, I);

8: PEm ← get

primary Em(ρ);

9: SEm ← get

secondary Em(B, M, D);

10: M ← update

M(PEm, SEm, M, P);

11: I ← filter(B, D, I, M);

12: π ← plan(B, I, Ac); {Ac: set of actions}

13: while not (empty(π) or succeeded(I, B) or

impossible(I, B)) do

14: α ← ﬁrst element of π;

15: execute(α);

16: π ← tail of π;

17: observe environment to get next percept ρ

18: B ← get

new beliefs(B, ρ, M);

19: PEm ← get

primary Em(ρ);

20: SEm ← get

secondary Em(B, M, D);

21: M ← update

M(PEm, SEm, M, P);

22: if (reconsider(I, B, M)) then

23: D ← get

options(B, I);

24: I ← filter(B, D, I, M);

25: end if

26: if not (sound(π, I, B)) then

27: π ← plan(B, I, Ac);

28: end if

29: SuccRate

← get

succ rate(SuccRate

, π);

30: end while

31: end while

Figure 2: Control loop for an emotional BDI agent.

Primary and Secondary emotions are obtained. Sec-

ondary Emotions also consider Desires and Beliefs

because they are the product of a more complex de-

liberative process that may emerge from evaluating

the agent options, current Beliefs (including past ex-

periences and/or expectations) and current Mood. As

it has been posed in section 3.1, some previous ap-

proaches may be used in this last two steps such as

(Dastani and Meyer, 2006; Steunebrink et al., 2012).

In particular we derive PEm directly from percepts

and SEm according to (Marsella and Gratch, 2009).

Current Mood is updated taking into account Pri-

mary and Secondary Emotions as well as the previous

Mood and the agent’s personality (line 10). The main

goal of this step is to perform a transformation from

a set of emotions to a mood in a coherent way. Inten-

tions are updated in the same way in line 11 on the

basis of the selected Desires and the Current Mood.

Then in line 12 the plan function generates a plan for

achieving the selected intentions. The actions of the

loop in lines 13-30 will be executed as soon as the

plan is not empty, succeeded, or impossible. In lines

14-16 the ﬁrst action of the plan is selected and ex-

ecuted, and the plan is updated with the remaining

actions. Lines 18-21 represent a pause that the agent

makes to detect changes in the environment (which is

veriﬁed in line 17), and reconsider its Intentions, de-

riving again Primary Emotions, Secondary Emotions

and Mood as previously in lines 8-10. If it’s worth

to reconsider and to deliberate (Intentions may suffer

changes according to the current state and Mood) the

Desires and Intentions are reevaluated (lines 23-24).

In lines 26-28 a replanning is made in case the cur-

rent plan doesn’t ﬁt well any more with the current

Intentions and Beliefs

. Finally, in line 29 a measure

of the relation between the times the plan has success-

fully fulﬁlled the committed Intentions and the times

it has been executed is saved. This indicator is used

for future deliberations

4 CONCLUSIONS

The O3A architecture is inspired on the most promi-

nent results of psychological and neurological areas.

It offers a general agent structure, with an open com-

ponent implementation in order to be applied in a

wide range of domains. It’s integration into a typical

BDI architecture allows to combine practical ratio-

nal elements with more “human” features, that results

in believable behaviors for the agents. At the same

time it offers an open structure in order to be ﬂexible

enough to adapt to several domains and applications.

A proposal of design for each component of the archi-

tecture is also offered. This approach has many prac-

tical uses in human-computer interaction applications

like education, pathologies treatments, training, enter-

tainment and human simulation behavior in general.

Nevertheless, the main challenge after roughly deﬁne

how the agent reasoning processes are integrated into

the architecture, is the detailed speciﬁcation of each

one of the particular components and how to achieve

a coherent behavior aligned with real situations.

This approach is a work in progress. We are cur-

rently engaged on implementing a practical applica-

tion in order to test the strength of the proposal as well

as to identify improvements and/or necessary mod-

iﬁcations. Speciﬁcally it is been tested in classical

games of experimental economy where humans take

Note that lines 22-28 are kept from the original algorithm in

(Bordini et al., 2007) where the authors point out that reconsidera-

tion is performed only if this leads to a change of intentions and a

replanning is performed if the plan is not sound any more accord-

ing to what the agent wants to achieve (intentions) and it thinks is

the current state of the world (beliefs).

The intention in this step is above all to keep the necessary

records associated to the results of current results of plans executed

and/or achievement of goals in order to reuse this information as a

“memory” for further deliberations.

ECTA2014-InternationalConferenceonEvolutionaryComputationTheoryandApplications

324

decisions that are often biased from the most ‘ratio-

nal’ solution.

ACKNOWLEDGMENTS

This work was supported by the Spanish government

grant MINECO/FEDER TIN2012-36586-C03-01and

HUMBACE (Human and Social Behaviour Models

for Agent-Based Computational Economics) project.

Thanks also to the Research and Development Sup-

port Programme of the UniversidadPolit´ecnica de Va-

lencia PAID-06-11 Project.

REFERENCES

Becker-Asano, C. and Wachsmuth, I. (2010). Affective

computing with primary and secondary emotions in

a virtual human. AAMAS ’10, 20(1):32–49.

Bordini, R. H., H¨ubner, J. F., and Wooldridge, M. (2007).

Programming multi-agent systems in AgentSpeak us-

ing Jason. Wiley.

Castelfranchi, C. (2000). Affective appraisal versus cogni-

tive evaluation in social emotions and interactions. In

Affective Interactions, pages 76–106. Springer.

Dam´asio, A. R. (2005). Descartes’ Error: Emotion, Rea-

son, and the Human Brain. Penguin Group US.

Dastani, M. and Meyer, J.-J. C. (2006). Programming

agents with emotions. In Proceedings of the 2006 con-

ference on ECAI 2006: 17th European Conference on

Artiﬁcial Intelligence August 29 – September 1, 2006,

Riva del Garda, Italy, pages 215–219. IOS Press.

Ekman, P. (1999). Basic Emotions, pages 45–60. John Wi-

ley & Sons, Ltd.

Frijda, N. H. (1987). The Emotions. Studies in Emotion and

Social Interaction. Cambridge University Press.

Gebhard, P. (2005). ALMA: a layered model of affect. In

Proceedings of the fourth international joint confer-

ence on Autonomous agents and multiagent systems,

pages 29–36, New York, USA. ACM.

Goldberg, L. R. et al. (1990). An alternative “description of

personality”: The big-ﬁve factor structure. Journal of

personality and Social Psychology, 59(6):1216–1229.

Jiang, H., Vidal, J. M., and Huhns, M. N. (2007). EBDI:

An architecture for emotional agents. In Proceed-

ings of the 6th International Joint Conference on

Autonomous Agents and Multiagent Systems, pages

11:1–11:3. ACM.

John, O. P. and Srivastava, S. (1999). The Big Five trait

taxonomy: History, measurement, and theoretical per-

spectives. In Pervin, L. A. and John, O. P., edi-

tors, Handbook of Personality: Theory and Research,

pages 102–138. Guilford Press, second edition.

LeDoux, J. E. (1998). The emotional brain: the mysterious

underpinnings of emotional life. A Touchstone book.

Simon & Schuster.

Marsella, S. C. and Gratch, J. (2009). EMA: A process

model of appraisal dynamics. Cognitive Systems Re-

search, 10(1):70–90.

Marsella, S. C., Gratch, J., and Petta, P. (2010). Compu-

tational models of emotion. In A Blueprint for Affec-

tive Computing: A Sourcebook and Manual, Affective

Sciene, chapter 1.2. OUP Oxford.

McCrae, R. R. and John, O. P. (1992). An introduction to

the ﬁve-factor model and its applications. Journal of

personality, 60(2):175–215.

Mehrabian, A. (1996). Analysis of the big-ﬁve personality

factors in terms of the pad temperament model. Aus-

tralian Journal of Psychology, 48(2):86–92.

Mehrabian, A. (1997). Comparison of the PAD and panas as

models for describing emotions and for differentiating

anxiety from depression. Journal of Psychopathology

and Behavioral Assessment, 19(4):331–357.

Mehrabian, A. and Russell, J. A. (1974). An approach to

environmental psychology. MIT Press.

Neto, A. and Silva, F. (2012). A computer architecture

for intelligent agents with personality and emotions.

In Zacarias, M. and Oliveira, J. V., editors, Human-

Computer Interaction: The Agency Perspective, vol-

ume 396 of Studies in Computational Intelligence,

pages 263–285. Springer Berlin Heidelberg.

Niedenthal, P. M. and Setterlund, M. B. (1994). Emotion

congruence in perception. Personality and Social Psy-

chology Bulletin, 20(4):401–411.

Ortony, A. (2003). On Making Believable Emotional

Agents Believable. In Trapple, R. P., Petta, P., and

Payer, S., editors, Emotions in Humans and Artifacts,

chapter 6, pages 189–212. MIT Press.

Ortony, A., Clore, G. L., and Collins, A. (1988). The Cog-

nitive Structure of Emotions. Cambridge University

Press.

Ortony, A. and Turner, T. J. (1990). What’s basic about

basic emotions? Psychological review, 97(3):315.

Parunak, H. V. D., Bisson, R., Brueckner, S., Matthews, R.,

and Sauter, J. (2006). A model of emotions for situ-

ated agents. In Proceedings of the ﬁfth international

joint conference on Autonomous agents and multi-

agent systems, AAMAS ’06, pages 993–995, New

York, USA. ACM.

Picard, R. W. (1997). Affective computing. MIT Press,

Cambridge, MA, USA.

Russell, J. A. and Mehrabian, A. (1977). Evidence for a

three-factor theory of emotions. Journal of research

in Personality, 11(3):273–294.

Ryckman, R. M. (2007). Theories of Personality. PSY 235

Theories of Personality Series. Thomson/Wadsworth.

Steunebrink, B. R., Dastani, M., and Meyer, J.-J. C. (2012).

A formal model of emotion triggers: an approach for

BDI agents. Synthese, 185:83–129.

Weiss, G. (1999). Multiagent systems: a modern approach

to distributed artiﬁcial intelligence. MIT press.

AnOpenArchitectureforAffectiveTraitsinaBDIAgent

325