Integrating Expectations into Jason for Appraisal in Emotion Modeling

Joaqu

ın Taverner, Bexy Alfonso, Emilio Vivancos and Vicente Botti

Departmento de Sistemas Inform

aticos y Computaci

on, Universitat Polit

ecnica de Val

encia,

Camino de Vera s/n, 46022, Valencia, Spain

Keywords:

Agents, Emotion Modeling, Appraisal, Expectations, Jason.

Abstract:

Emotions have a strong inﬂuence on human reasoning and behavior, thus, in order to build intelligent agents

which simulate human behavior, it is necessary to consider emotions. Expectations are one of the bases for

emotion generation through the appraisal process. In this work we have extended the Jason agent language

and platform for handling expectations. Unlike other approaches focused on expectations handling, we have

modiﬁed the agent reasoning cycle to manage expectations, avoiding complex additional mechanisms such as

monitors. This tool is part of the GenIA

architecture and, hence, is a step towards the standardization of the

emotion modeling process in BDI (Beliefs-Desires-Intentions) agents.

1 INTRODUCTION

Human behavior is not merely rational. It is inﬂu-

enced by individuals’ emotional characteristics such

as personality, emotions or mood. Psychological and

neurological sciences support this statement, offering

the foundation for the research on this ﬁeld (Dam

asio,

1994; Baumeister et al., 2007). In recent years sev-

eral studies have been proposed for modeling emo-

tions using software processes and software agents.

When software agents need to show a human-like be-

havior they need to include emotions in their reason-

ing process. However not many proposals in this ﬁeld

offer general practical tools oriented to the design and

implementation phase.

A key process in emotion modeling is the ap-

praisal process. In psychology, appraisal theories ar-

gue that emotions are the result of the interpretations

and explanations that each individual performs based

on his/her circumstances and concerns. In this evalua-

tion process, an individual interprets his/her relation-

ship with the environment (Scherer, 2001; Lazarus,

1994). As part of this, judgments are performed ac-

cording to certain criteria or variables called appraisal

variables. A set of speciﬁc emotions result from the

appraisal process corresponding to different conﬁg-

urations of these judgments (Lazarus, 1994). There

are several criteria for evaluating an event that takes

place in the individual environment. One of the most

accepted criteria is the implications of the event in

the future. This criteria may result in emotions such

as hope or fear in relation to things that will hap-

pen, or may result in emotions that are reactions to

violations of expectations (Winkielman et al., 1997;

Castelfranchi and Lorini, 2003) such as surprise or

disappointment. Our approach is focused on this kind

of emotions. For computationally modeling the ap-

praisal of an event in relation to its future implica-

tions, it is necessary to evaluate the likelihood of the

event consequences (Golub et al., 2009), as well as the

value of the unexpected and the possibility of change.

Therefore, it may be necessary to count on a repre-

sentation of future expectations. Mechanisms to eval-

uate the probability of events, actions and their con-

sequences may be also required.

According to the OCC model (Ortony et al.,

1988), a widely accepted model of emotions that of-

fers a mechanism of appraisal and a classiﬁcation

of 22 emotion types, one of the appraisal variables

that affects almost all emotion types is related to ex-

pectancy, because expectations affect all emotional

experiences. This is also supported by most ap-

praisal theories like (Scherer, 2001; Roseman, 2001;

Lazarus, 1994). Expectations are an anticipatory

mental component (Ranathunga et al., 2011), which

allow individuals to satisfy their need of anticipating

future events in order to blur the unexpectedness of an

unknown situation.

In order to allow deﬁning expectations in an agent

program we have extended the syntax of the Jason

agent language (Bordini et al., 2007), allowing to de-

ﬁne the probability and a time interval for expecta-

tions that may be used in the appraisal process of

an affective agent. In our approach expectations can

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

Integrating Expectations into Jason for Appraisal in Emotion Modeling.

DOI: 10.5220/0006070202310238

In Proceedings of the 8th International Joint Conference on Computational Intelligence (IJCCI 2016) - Volume 1: ECTA, pages 231-238

ISBN: 978-989-758-201-1

231

be deﬁned in a similar way to beliefs. We have

also created a method for keeping track of fulﬁlled

and not fulﬁlled expectations. Besides, in order to

show how expectations can indirectly inﬂuence an

agent decisions, we use an example where agents

expectations inﬂuence the agent affective state, and

in turn, this affective state inﬂuence further deci-

sions. This extension is part of the development of

the GenIA

architecture (Alfonso et al., 2014; Alfonso

et al., 2016) presented in the section 2.1.

The rest of the paper is organized as follow. Sec-

tion 2 presents the supporting affective theories and

the GenIA

architecture. The method proposed for

deﬁning and monitoring expectations is presented in

Section 3. An example of how to use expectations in

a Jason agent that plays the BlackJack game is pre-

sented in Section 4. Finally, a comparative study with

other similar approach is showed in Section 5, and

Section 6 offers some conclusions.

2 BACKGROUND AND

SUPPORTING THEORIES

When computer scientists model affect, they face two

broad challenges: how to model affect and how to en-

rich artiﬁcial agents architectures and languages to in-

clude those affective models (Reisenzein et al., 2013).

Several psychological theories provide almost com-

plete support for affect-related processes (e.g., emo-

tion generation, and emotions effects on cognition,

expression, and behavior) (Hudlicka, 2014). Con-

sequently, approaches for agent modeling are differ-

ent considering what psychological theories support

them (Reisenzein et al., 2013). For a review about

computational approaches for modeling emotions see

(Marsella et al., 2010), and (Reisenzein et al., 2013).

Nevertheless, the task of systematically recreat-

ing existing emotion theories, and to use the gen-

eral strategy of building formal languages for this,

may be cumbersome. Two more viable strategies

can be used: “1) break up existing emotion theories

into their component assumptions and 2) reformulate

these assumptions in a common conceptual frame-

work”(Reisenzein et al., 2013). In (Alfonso et al.,

2014; Alfonso et al., 2016) it is proposed GenIA

which is a common conceptual framework that sup-

ports different emotion theories.

2.1 The GenIA

Architecture

GenIA

(a General-purpose Intelligent Affective

Agent Architecture) is a BDI (Beliefs-Desires-

Intentions) agent architecture designed to create

affective agents. This architecture is grounded on

general aspects of widely accepted psychological

and neurological theories but it doesn’t makes any

commitment with any particular theory. Nevertheless

it offers a default design that includes an appraisal

process inspired by (Marsella and Gratch, 2009).

GenIA

includes all the processes of a BDI agent

as well as a set of main affective processes according

to a solid theoretical background (see Figure 1). Ac-

cording to GenIA

these are the processes that should

be considered when building appraisal-based models

for affective agents (Alfonso et al., 2014; Hudlicka,

2014; Marsella et al., 2010). As part of the BDI

processes, it includes a belief revision function (brf),

which determines new beliefs starting from a per-

ceptual input and the agent’s current beliefs; an op-

tions generation process (options), which takes the

agent’s current beliefs and intentions to determine its

desires (options or courses of actions available), i.e.

the means to achieve its intentions; a ﬁlter process

(ﬁlter), which determines the agent’s intentions, i.e.

what to do, through a deliberation process that uses

previously-held intentions, and the agent’s current be-

liefs and desires (the new set of intentions will contain

either newly adopted or previously-held intentions);

and action selection function (execute), which returns

the next action to be executed on the basis of cur-

rent intentions. The affective processes, on the other

hand, include an appraisal process, whereby an eval-

uation of the current situation is performed consider-

ing the agent environment, cognitive state, and con-

cerns, and where a set of appraisal variables are de-

rived ; affect generator, where the appraisal variables

that result from the appraisal process are transformed

into a representation of the agent’s affective state; af-

fect regulator, which determines the possible emo-

tional behaviors and coping responses for the given

situation; affective modulator of beliefs which deter-

mines if and how the affective state biases the agent’s

beliefs, contributing to the beliefs maintenance ac-

cording to the affective state; and the affect’s tem-

poral dynamic, which doesn’t depend of any other

process and no other process depends on it, and deter-

mines the duration of the affective state’s components

as well as how their intensities decay over time.

3 PROPOSAL

In this section we present the proposed method for

deﬁning and managing expectations. In order to allow

an agent programmer to deﬁne expectations we have

extended the Jason language including a new structure

for expectations that can have associated a probabil-

ECTA 2016 - 8th International Conference on Evolutionary Computation Theory and Applications

232

appraisal

affect

generator

coping

affective

modulator

of beliefs

execute

filter

options

brf

beliefs

affect

temporal

dynamics

actions

perceptual

input

concerns

internal

events

personality

current

options

external

events

affectively

relevant

events

affective

state

Affective processes

intentions

Affective cycle

Rational processes

Figure 1: GenIA

: a General-purpose Intelligent Affective Agent Architecture that integrates BDI and affective processes.

Sequences are represented as solid line arrows, subprocess as dashed line arrows, and exchange of information as dotted line

arrows.

ity value and a time range. Also, in order to identify

the inﬂuence of the expectations in the agent affective

state, we allow to deﬁne the valence associated with

the fulﬁllment of an expectation, (i.e. positive or

negative). The expectations management, in turn, is

performed at certain points of the agent BDI reason-

ing cycle. To this end, we have included a new step in

this reasoning cycle. Next we brieﬂy introduce Jason

and its characteristics as an agent language (Section

3.1). We also semi formally present the extension of

the Jason language (Section 3.2) and the extension of

its reasoning cycle (Section 3.3).

3.1 Jason Agent Language

Jason (Bordini et al., 2007) is an interpreter for an ex-

tended version of AgentSpeak. Jason agents are BDI

based, thus agents continuously decide actions to per-

form to reach their goals. A Jason agent program can

contain beliefs, goals, and plans. Beliefs represent the

knowledge the agent has about its environment, him-

self or about other agents. Goals represent the agent

desires, and plans consist of a sequence of actions that

are designed to either reaching those goals or simply

respond to events. Events in Jason are generated by

the addition or deletion of beliefs, the addition or fail-

ure of goals, or the addition or failure of test goals

On the other hand the agent life cycle is controlled

by the reasoning cycle (RC) that is continuously exe-

cuted (see Figure 2).

According to the Jason semantic, several steps are

performed in an agent RC. The initial step is in charge

of processing the received messages (ProcMsg). Next

an event is selected to be processed in the SelEv step.

Then, in the step RelPl, a set of relevant plans are se-

lected, and in the step ApplPl a subset of this set is

Test goals are those goals that aim to retrieve informa-

tion from the agent belief base.

RelPl SelAppl

AddIMSelInt

ClrInt

ExecInt

SelEv

ProcMsg

ApplPl

Figure 2: The reasoning cycle of a Jason agent (Bordini

et al., 2007).

beliefs → ( (literal|literal prob) “.” )*

literal prob → [ “∼” ] atomic form prob

atomic form prob→ ( <

<ATOM>

> | <

<VAR>

> )

[“(” list of terms “)”]

[“[”list of terms“]”]

[ t point range ]

t point range→ “<”arithm expr“,”arithm expr“>”

Figure 3: Simpliﬁed extension of the EBNF for including

expectations in the Jason Language.

selected, which contains the applicable plans

. Next a

plan is selected for its execution in the step SelAppl;

this implies the addition of an intended mean (or a

new intention) what is performed in the step AddIM.

Finally, a particular intention is selected for its execu-

tion from the set of intentions (SelInt), and the cycle

ends by clearing the ﬁnished intentions (ClrInt).

3.2 Extension of the Jason Language

We have modiﬁed the syntax of the Jason language

to allow the deﬁnition of expectations with a proba-

bility and a time range. Figure 3 shows a part of the

Jason language EBNF. The complete EBNF of the Ja-

son language can be found at (Vieira et al., 2007).

Figure 3 shows that the rule for deﬁning be-

liefs has been extended so that it can be deﬁned

Relevant plans are those whose triggering event

matches the event being processes, and applicable plans are

those whose conditions for being executed are fulﬁlled ac-

cording to the agent beliefs.

Integrating Expectations into Jason for Appraisal in Emotion Modeling

233

as a literal according to its original deﬁnition or

as a literal prob. When written according to

literal prob, a belief becomes an expectation. Ac-

cording to the deﬁnition of literal prob it is possi-

ble to associate a time range to an expectation, which

must follow the structure of t point range. A time

range allows expectations to be time-bound, so that it

is possible to hypothesize about something for a pe-

riod of time in the present and/or future. Thus a time

range includes an initial time and a ﬁnal time. The ini-

tial time can be expressed either as an arithmetical ex-

pression that can include the reserved word ‘Now’, or

directly the word ‘Now’. We have created this word

‘Now’ for easily referring to the ”current time” (in

milliseconds). The ﬁnal time can be expressed either

as an arithmetical expression or as the reserved word

‘Inﬁnite’ (indicating an undetermined time in the fu-

ture). Moreover, a belief can also become an expecta-

tion if one of its annotations is a probability. A prob-

ability is one annotation with the form prob(P,V),

where P is a numerical value between 0 and 1 and

V is an optional component that can have the values

positive or negative. We have included this com-

ponent to simulate what an appraisal process would

do in order to use expectations for determining their

inﬂuence in the agent affective state. For example

consider the next portion of a Jason code:

belief[prob__(Number,Connotation)]<T1, T2>

In this example the expectation has a structure

similar to the structure of a belief, that has a prob-

ability Number, a valence Connotation and a time

range (T1 represents the initial time and T2 represents

the ﬁnal time). We use the probability to indicate the

level of expectedness of the expectation. Thus we can

use this probability to check the impact generated by

an expectation in the calculation of the affective state

(Golub et al., 2009). We also allow to specify the

valence of expectations to indicate whether the con-

sequences of their fulﬁllment are positive or negative.

Determining the valence of expectations may be one

of the tasks performed by an appraisal process, so

that future extensions of the present approach won’t

need the expectations’ valence component. Finally,

one of the innovations we propose is the possibility

of deﬁning expectations for a time range. Within this

time range the expectations can be fulﬁlled. Once the

time range has ended, it is considered that expecta-

tions haven’t been fulﬁlled. For example, one expec-

tation where the agent believes that the weather will

be cloudy with a probability of 0.5 (at some point be-

tween now and within two hours) and where the agent

considers that a cloudy weather is “something good”

can be written as:

time(cloudy)[prob__(0.5,positive)]<Now,

Now+2*60*60*1000>

If within that time range the time(cloudy) belief

is inserted in the agent belief base (either through the

perception process or by a message received from an-

other agent), then the expectation will be fulﬁlled. If

two hours later no belief time(cloudy) is perceived

or received as a message, the expectation is consid-

ered not fulﬁlled.

3.3 New Step in the Jason Reasoning

Cycle

A Jason agent conﬁguration is deﬁned by a tuple

hag,C, M,T,si (Vieira et al., 2007). The components

of this tuple can be modiﬁed on each step of the agent

reasoning cycle. The ﬁrst component (ag) represents

the agent program which contains a set of beliefs bs

and a set of plans ps. C represents the agent circum-

stance, containing the current set of intentions, events,

and actions to be performed in the agent environment.

M is the component that stores the agent communica-

tion aspects. T stores temporary information includ-

ing relevant plans in relation to an event (R), appli-

cable plans (Ap), and data considered in a particu-

lar reasoning cycle including the current intention (ι),

event (ε), and applicable plan (ρ). Finally s contains

the step of the reasoning cycle being executed, where

s ∈ {ProcMsg, SelEv, RelPl, ApplPl, SelAppl, AddIM,

SelInt, ExecInt, ClrInt}. In our approach we have in-

cluded a new element es in the agent program ag, rep-

resenting the set of expectations of the agent. Also,

the agent temporary information T has been modiﬁed.

We have included a new component Exp in T that rep-

resents the expectations pending of being processed

by any dependent process such as an appraisal pro-

cess. We deﬁne Exp as a tuple h f pe, f ne,n f pe,n f nei

were f pe represents the set of fulﬁlled positive expec-

tations, f ne the set of fulﬁlled negative expectations,

n f pe the set of not fulﬁlled positive expectations, and

n f ne the set of not fulﬁlled negative expectations.

One expectation can be removed due to three rea-

sons: (1) because there is an action in the agent code

to eliminate it, (2) because it has been fulﬁlled or (3)

because it hasn’t been fulﬁlled. In the last two cases,

we keep a record of this expectation. This record is

able to differentiate fulﬁlled from unfulﬁlled expec-

tations and also positive from negative (consequences

of) expectations (fulﬁllment). We use this differen-

tiation for determining the inﬂuence on the affective

state. Following this criteria Exp is updated in two

different moments during the agent reasoning cycle.

Firstly it is updated in the br f (belief revision func-

tion), in charge of updating the agent’s beliefs accord-

ing to what is perceived. Secondly it is updated at

ECTA 2016 - 8th International Conference on Evolutionary Computation Theory and Applications

234

Figure 4: The Jason agent reasoning cycle extended with a new state EvalExp.

a ﬁxed point of the agent reasoning cycle. We have

modiﬁed the br f so that, besides adding and remov-

ing beliefs, expectations are also added and removed.

Every time a belief is inserted we check if that belief

fulﬁlls some expectation. But as an expectation’s time

range may not be active at the time of the insertion or

removal of a belief, we need a way of constantly eval-

uate expectations. In order to evaluate expectations in

a consistent way we have modiﬁed the Jason reason-

ing cycle by including a new step EvalExp as shown

in Figure 4.

The second moment for updating Exp is per-

formed in the EvalExp step. This step has been in-

troduced after ProcMsg and before SelEv, so that the

transition from ProcMsg to SelEv has been removed

and two new transitions from ProcMsg to EvalExp

and from EvalExp to SelEv have been created. This

modiﬁcation implies that, after starting the reasoning

cycle (ProcMsg) the next step is EvalExp, where ex-

pectations are evaluated and updated according to the

transitions rule EvalExp

Exp = ( f pe, f ne,n f pe,n f ne)

hag,C, M, T,EvalExpi → hag,C, M, T

,SelEvi

(EvalExp)

where: T

Exp

f pe

= EvFPE(Exp) ∪ f pe

Exp

f ne

= EvFNE(Exp) ∪ f ne

Exp

n f pe

= EvNFPE(Exp) ∪ n f pe

Exp

n f ne

= EvNFNE(Exp) ∪ n f ne

All new functions introduced in the transition

rule presented determine a set of expectations with

the form (b, p,v,t

), where b is the expectation,

p the probability value, v the expectation valence

(positive or negative), t

the initial time for the

expectation, and t

the ﬁnal time. Given the set bs of

We have followed the same notation as (Vieira et al.,

2007) where attributes are represented as a subindex. For

example, the beliefs set bs of the agent program ag is repre-

sented as ag

agent beliefs, the set es of the agent expectations, and

the current time ct, the previous functions are deﬁned

as follows:

Deﬁnition 1.

EvFPE(es) = {(b, p,v,t

)|b ∈ es, b ∈ bs,

>= ct,ct <= t

,v = ‘positive

}

Deﬁnition 2.

EvFNE(es) = {(b, p,v,t

)|b ∈ es, b ∈ bs,

>= ct,ct <= t

,v = ‘negative

}

Deﬁnition 3.

EvNFPE(es) = {(b, p,v,t

)|b ∈ es, b /∈ bs,

> ct,v = ‘positive

}

Deﬁnition 4.

EvNFNE(es) = {(b, p,v,t

)|b ∈ es, b /∈ bs,

> ct,v = ‘negative

}

4 EXAMPLE

To demonstrate how to use expectations in Jason

agents, we have developed a game based on the

BlackJack cards game (Grifﬁn, 1986). In this game

the player must get cards with a total value of twenty

one for reaching a blackjack. These type of games

are very useful for simulating decision under the in-

ﬂuence of emotions, because it has been widely stud-

ied the role that emotions play in decision-making in

these contexts. Besides, games allow to see changes

in the emotional state, as players usually are happy

when they win and get angry when they lose.

We have created two agents as shown on Figures 5

and 6. On the upper side it is shown a player (an affec-

tive agent). On the bottom side, it is shown the ‘Bank’

which is a Jason agent without emotions and com-

pletely rational. The agent player starts the game by

Integrating Expectations into Jason for Appraisal in Emotion Modeling

235

Figure 5: Graphical Interface. Board situation for the player

(top) and the bank (bottom). The player loses.

Figure 6: Graphical Interface. Board situation for the player

(top) and the bank (bottom). The player wins.

asking a card. Depending on its points and its affec-

tive state it must choose between stand or hit. Figure 5

shows a situation where the player loses and Figure 6

shows a situation where the player wins. With this ex-

ample we have shown that expectations are crucial for

generating emotions since, only by modeling expecta-

tion it has been possible to simulate a variety of situa-

tions completely coherent with what would happen in

the real life. It should be clear that this is just a simple

example for the purpose of showing how expectations

could be used to determine their inﬂuence on the af-

fective state and to determine how this affective state

may impact decisions. We have represented the af-

fective state through the emoticons shown on Figure

Figure 7: Emoticons that represent the different affective

states (depressed, sad, neutral, satisﬁed, happy).

As stated before, the affective agent is able to

decide according to its affective state. To do this,

we have included a new annotation in the label of

the agent plan. This new annotation has the form

[affect (Number)], where Number indicates the

value of the current affective state. We have modi-

ﬁed the function that selects the plan to be executed

to consider the current affective state.

@p2[affect__(2)]

+card(Value)[source(house)] : myPoints(N)

-card(Value)[source(house)];

-+myPoints(N+Value);

!nextPlay(17.0, N+Value).

For example, in the previous fragment of code

there is a plan that will be executed if the affective

state is 2 (sad). The triggering event of this plan

is +card(Value). This event represents that it has

got a new card, and the variable N represents the

total points of the player before the new card. As

part of the actions of the plan, the agent number of

points is updated by adding the value of the new card.

Also as part of the plan, the agent creates a new

goal nextPlay(17.0,N+Value), where 17.0 repre-

sents the points for standing, and N+Value is the total

points the affective agent has.

In this example, the affective state is only mod-

iﬁed by using the evaluations (positive or negative)

of the expectations, as well as the information about

whether they were fulﬁlled or not. The affective state

is a value that can have ﬁve possible ranges of values

one per emoticon of Figure 7: depressed, sad, neu-

tral, satisﬁed, and happy respectively. Following this

example, the affective agent has the expectation:

win(round)[prob__(0.5,positive)]<Now,Now+10000>

According to this expectation the agent is waiting

to win this round

with a probability of 0.5. The agent

also has an expectation where: it is waiting to have

twenty one at some point in the game, it is waiting

to win two consecutive rounds at some point in the

game, and it is waiting to get a card with 10.0 points

in the current round:

get(21.0)[prob__(0.1,positive)] <Now, Infinite>

win_followed(2)[prob__(0.6,positive)] <Now,

In our proposal, we consider that the rounds shall last

for ten seconds

ECTA 2016 - 8th International Conference on Evolutionary Computation Theory and Applications

236

Infinite>

get(10.0)[prob__(0.3,positive)] <Now, Now+10000>

All expectations are positive, so when any of them

is fulﬁlled, this will have a positive effect on the agent

affective state. However, if expectations are not ful-

ﬁlled, this will have a negative effect on the agent af-

fective state. To check the impact that the fulﬁllment

or not fulﬁllment of expectations has in the affective

state we use the probability. Expectations with a high

probability will have less impact (exactly the comple-

ment of the expectation probability) than those with a

low probability. For example, if the player gets a card

with a value of 10.0 units the affective state will be

increased in 0.7, otherwise the affective state will be

decreased in 0.3 units.

5 RELATED WORK

There are some precedents in incorporating expecta-

tions at Jason agents (Craneﬁeld, 2014). Probably the

approach most similar to ours is the one proposed in

(Ranathunga et al., 2010; Ranathunga et al., 2011;

Ranathunga and Craneﬁeld, 2012). The authors in-

corporate a mechanism for expectations management

through monitors. Similar to ours, this approach is

able to detect expectations fulﬁllment and violation at

execution time. This allows to monitor changes in the

environment and the behavior of others to this end.

Nevertheless, in this approach expectation fulﬁllment

is notiﬁed through event generation, allowing to han-

dle those events through plans. By contrast in our

approach expectations are used as appraisal variables

which can be used to determine emotions generated,

and may modify the agent affective state.

In Ranathunga’s approach an expectation is active

if an associate condition is fulﬁlled. Also there is

no time management for expectations, what is detri-

mental to the expectation expressivity. In our ap-

proach expectations can be deﬁne for being fulﬁlled

or violated at a time range, thus an expectation be-

comes active when its time range becomes active. Un-

like Ranathunga’s approach, were there is a “separate

mechanism for obtaining information from the sys-

tem in which the Jason agent is situated” (Ranathunga

et al., 2011), in our approach information from the

system is obtained through the own Jason mecha-

nisms and tools. This makes the sensoring and per-

ception process more transparent to the user.

Finally in Ranathunga’s approach, unlike ours, it

is not possible to express to what extent the agent ex-

pects something (i.e. it is not possible to assign a

probability to expectations). Also, monitors are used

to make the evaluation processes while we use the

own reasoning cycle of the agent. Therefore our pro-

posal avoids the costs associated with the concurrence

derived from the use of monitors. Our approach also

keeps a record of fulﬁlled and not fulﬁlled expecta-

tions for being used and managed by the process that

may need them, such as the appraisal process.

6 CONCLUSIONS

Emotions play an important role in the decision-

making process of humans. Currently tools for allow-

ing the implementation of affective agents are scarce.

As discussed in this paper, expectations handling is

the base for simulating prospective emotion gener-

ation through the appraisal process. We have ex-

tended Jason to introduce expectations representation

and handling in agents as a part of the development of

the GenIA

architecture.

The new proposed method for expectations han-

dling is fully integrated into a BDI agent reasoning

cycle, so that no additional structures (what may com-

promise the agent performance) were needed. Also

this method is able, not only to detect the fulﬁllment

or not fulﬁllment of expectations within a time range,

but also to keep track of this. This record can be used

for determining the inﬂuence of expectations on the

affective state of the agent. The example used shows

that the probability associated to the expectations may

have a great importance on determining the impact of

the fulﬁllment or not of an expectation on the affec-

tive state as well. Clearly, variations on an individ-

ual affective state are not only determined by expec-

tations, nevertheless they are necessary to determine

this inﬂuence. We have conﬁrmed this by observing a

player agent playing a cards game with a feedback of

its affective state. The use of an affective agent as an

artiﬁcial player clearly showed good chances in order

to improve the game experience.

As part of our future work we will improve the de-

fault implementation of GenIA

by creating new meth-

ods for determining the values of other important ap-

praisal variables such as “desirability” or “controlla-

bility”. Also, more sophisticated methods for deter-

mining how these appraisal variables impact the agent

affective state, and how this affective state inﬂuences

the decision making process will be included.

REFERENCES

Alfonso, B., Vivancos, E., and Botti, V. (2016). Towards

Formal Modeling of Affective Agents in a BDI Archi-

Integrating Expectations into Jason for Appraisal in Emotion Modeling

237

tecture. ACM Transactions on Internet Technology, To

appear.

Alfonso, B., Vivancos, E., and Botti, V. J. (2014). An Open

Architecture for Affective Traits in a BDI Agent. In

Proceedings of the 6th ECTA 2014. Part of the 6th

IJCCI 2014, pages 320–325.

Baumeister, R. F., Vohs, K. D., DeWall, C. N., and Zhang,

L. (2007). How emotion shapes behavior: Feed-

back, anticipation, and reﬂection, rather than direct

causation. Personality and Social Psychology Review,

11(2):167–203.

Bordini, R. H., H

ubner, J. F., and Wooldridge, M. (2007).

Programming multi-agent systems in AgentSpeak us-

ing Jason. Wiley.

Castelfranchi, C. and Lorini, E. (2003). Cognitive anatomy

and functions of expectations. Proceedings of IJ-

CAI’03 Workshop on Cognitive Modeling of Agents

and Multi-Agent iterations.

Craneﬁeld, S. (2014). Agents and Expectations, pages 234–

255. Springer International Publishing, Cham.

Dam

asio, A. R. (1994). Descartes’ error: emotion, reason,

and the human brain. Quill.

Golub, S. A., Gilbert, D. T., and Wilson, T. D. (2009). An-

ticipating one’s troubles: the costs and beneﬁts of neg-

ative expectations. Emotion, 9(2):277–281.

Grifﬁn, P. (1986). The Theory of Blackjack: The Compleat

Card Counter’s Guide to the Casino Game of Twenty-

one. Faculty Publishing.

Hudlicka, E. (2014). From Habits to Standards: Towards

Systematic Design of Emotion Models and Affective

Architectures. In Emotion Modeling, pages 3–23.

Springer International Publishing.

Lazarus, R. (1994). Emotion and Adaptation. Oxford Uni-

versity Press.

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

Science, pages 21–46. OUP Oxford.

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

nitive structure of emotions. Cambridge University

Press, Cambridge, MA.

Ranathunga, S. and Craneﬁeld, S. (2012). Expectation and

complex event handling in bdi-based intelligent vir-

tual agents (demonstration). In Proceedings of the

11th International Conference on Autonomous Agents

and Multiagent Systems - Volume 3, pages 1491–1492.

Ranathunga, S., Purvis, M., and Craneﬁeld, S. (2010). In-

tegrating expectation monitoring into jason: A case

study using second life. Presented at the 8th Euro-

pean Workshop on Multi-Agent Systems.

Ranathunga, S., Purvis, M., and Craneﬁeld, S. (2011). Inte-

grating expectation handling into jason. The Informa-

tion Science, (2011/02).

Reisenzein, R., Hudlicka, E., Dastani, M., Gratch, J., Hin-

driks, K., Lorini, E., and Meyer, J.-J. (2013). Compu-

tational modeling of emotion: Toward improving the

inter-and intradisciplinary exchange. IEEE Transac-

tions on Affective Computing, 4(3):246–266.

Roseman, I. J. (2001). A Model of Appraisal in the Emotion

System: Integrating Theory, Research, and Applica-

tions, pages 68–91. Oxford University Press.

Scherer, K. R. (2001). Appraisal considered as a process of

multilevel sequential checking. Appraisal processes

in emotion: Theory, methods, research, 92:120.

Vieira, R., Moreira,

A. F., Wooldridge, M., and Bordini,

R. H. (2007). On the formal semantics of speech-act

based communication in an agent-oriented program-

ming language. J. Artif. Intell. Res.(JAIR), 29:221–

267.

Winkielman, P., Zajonc, R. B., and Schwarz, N. (1997).

Subliminal affective priming resists attributional inter-

ventions. Cognition and Emotion, 11(4):433–465.

ECTA 2016 - 8th International Conference on Evolutionary Computation Theory and Applications

238