Modeling Emotion Understanding in Stories: Insights from Traumatic Brain Injured Patients

Abstract

In this paper, we describe a case study in emotion understanding in stories that leverages the contribution of traumatic brain injured people. In particular, we focus on moral emotions, leveraging the di erences in moral functioning that characterizes traumatic brain injured patients. By comparing the understanding of the moral and emotional aspects of characters' behavior in traumatic brain injured patients and in a control group of neurologically healthy subjects, we observed slight, yet meaningful di erences in the two groups. We describe the test methodology and results, discussing their implications for the design of rehabilitation applications that leverage virtual characters.

Keywords

• emotions
• narrative understanding
• rehabilitation

-

Emotions play a fundamental role in stories, as acknowledged by scholars for centuries, with contributions that range from philosophy [ 2 ] and psychology [ 8 ] to narratology [ 15 ] and drama studies [ 24 ]. Emotions in stories have been studied also from a media-speci c perspective, by taking into account the expressive means enabled by the media through which the story is delivered, from lm [ 27 ] and videogame [ 28 ] to literary studies [ 15 ].

By creating an emotional bond with the characters, through mechanism such as sympathy, the audience experiences the process of identi cation that contemporary aesthetics sees as the core of the narrative experience [ 23, 13 ]. The study of how characters' emotions are understood by the audience is relevant to understanding how this bond is constructed. In this paper, we try to shed light on the understanding of characters' emotions through the insight provided by a group of people with traumatic brain injuries (TBI). Our attempt leverages the discrepancy observed in moral and emotional functioning in TBI patients to assess the role of the rational and emotional components in the understanding of stories. Traumatic brain injured patients are characterized by impairments of social cognition, as described by Milders et al. [ 21 ]. Although it is very difcult to nd clear indications on the characteristics of such impairments, it is clear that they involve moral functioning. According to Milders et al. [ 21 ], moral behavioural consequences of TBI tend to be more di cult than physical impairments for caregivers to deal with. Although there is no theoretical account on these moral impairments, the main di erences on moral functioning for braindamaged patients seem to pertain to tasks involving moral emotions [ 14, 22 ].

Computational models of characters, delivered by the AI research in virtual agents [ 3, 25 ], provide a formal model of the characters' behavior and emotions that can be compared with the expectations of the audience about what the characters will do and feel. Moreover, computational models of characters provide a formal framework for designing and implementing experiments about story understanding. Once validated, these models can be straightforwardly employed to drive the behavior of virtual characters in applications that range from entertainment to education and rehabilitation.

In previous work, Battaglino and Damiano [ 5 ] demonstrated through experiments with users that a computational model of agent (the Moral Emotional Agent, or MEA [ 6 ]) that encompasses moral values and emotions actually matches the expectations of the audience about the behavior and the emotions of characters in stories that concern moral dilemmas. Our methodology is aimed at comparing the di erences in the understanding of emotions, and of moral emotions in particular, in traumatic brain injured patients and control subjects with no traumatic brain impairment. The approach we propose adapts to our case study the methodology described by Battaglino and Damiano [ 5 ] to evaluate the MEA model: the characters' emotions generated by the agent model were compared with the emotions attributed to the character by the human users, while here we compare them with the test and control group to study the di erences between the two groups.

This paper is organized as follows. In Section 2 we describe the agent model we adopt for modelling story characters and its validations through experiments conducted with human subjects. The scenarios employed for the original experiments, re-used here for investigating the understanding of emotions by TBI patients, are described in Section 3. The experiment is described in 4 and its results are discussed in Section 5. Conclusion ends the paper. 2

Modelling story characters with the MEA agent Moral values are especially relevant to the understanding of stories. According to Bruner, who stressed the importance of stories in cognitive and social psychology, one of the major functions of stories is the transmission of a culture's values [ 8 ]. In Bruner's view, narrative characters are taken by the audience to have not only beliefs and desires, but also values. In drama, the importance of values is accrued by the fact, clearly stated by Elder Olson, that \Our emotions are evidently functions of a system of values, and are regulated by that system" [ 24 ]. The Moral Emotional Agent (MEA) [ 6 ] accounts for the relevance of values through an explicit moral component: in the MEA model, the emotional state of the agent in response to a given situation is determined not only by its congruence with the agent's goals but also by its congruence with the agent's values. The experiments described by Battaglino and Damiano in [ 5 ] show that the predictions of the model about the story characters' actions and emotions actually match the expectations of the users. 2.1

The MEA model

Designed to implement virtual characters for linear and interactive storytelling, the MEA model integrates into the well known Belief Desire Intention (BDI) model [ 7 ] an emotional component that relies on the cognitive theory of emotions by Ortony, Clore and Collins [ 1 ] (often referred to as OCC model). The BDI model, inspired by the pioneering work of Bratman [ 7 ] on bounded rationality, provides a theoretically sound and formally speci ed basis for the implementation of virtual agents [ 4, 18 ]. In the BDI model, the agent forms practical intentions (or plans) based on her/his beliefs in order to achieve her/his desires (or goals).

Here, we provide a sketch of the MEA model, focusing on the role of emotions in its deliberation component; details can be found in [ 6 ]. Formally, the agent features a set of goals, which form the motivational component of the agent. The agent's deliberation (Deliberation Process in Fig. 1) depends on the importance of its goals and on the possibility for the agent to achieve them. Each goal is associated with an importance of success and an importance of failure, and with three di erent set of conditions: adoption conditions, success conditions and failure conditions. When the agent believes that one of the adoption conditions of a goal is veri ed in the world, the goal becomes an active intention and can compete for being selected to become the active focused intention; if this is the case, the agent tries to nd sequences of actions (i.e. plans) to achieve it. For each plan, the agent computes its cost and probability of success.

Following Van Fraassen's suggestion that moral values should be prioritized [ 12 ], the moral dimension of the agent is formed by a set of values organized in a scale. Each value is associated with a priority that speci es the importance of the value for the agent, and with a set violation conditions. When one of the violation conditions holds in the state of the world, the value is at stake. If the violation condition does not hold anymore (and no other violation conditions have become true), the value is brought back to balance.

The values are the moral drive of the agent: they constrain the behavior of the agent to its moral dimension and allow the agent to appraise the behavior of self and others in moral terms.

After the deliberation phase, the agent trades o its values against its goals to choose a plan for execution (see Planning and Anticipatory Appraisal Process in Fig. 1). The role of values in the agent's deliberation is mediated by the emotional appraisal: for each plan, the agent appraises the consequences of the plan in terms of the emotional states it would generate (Anticipatory Emotional Appraisal). To do so, the agent calculates the Expected Emotional Reward utility for each plan, based on the emotions generated by the expected consequences of

Perceive   (update  knowledge)  

Events  

Appraisal  and  Affect  

Processes   (generation  of   emotions)  

Execution  Process   (executes  the  first  action   of  the  current  plan)  

Deliberation  Process   (forming  active   intentions  and  the   active  focused   intention)  

Planning  and   Anticipatory  Appraisal  

Processes   (finding  plans  and   choose  the  best  plan)   plans on the agent's goals and values. The agent will choose the plan that has the best trade-o between positive emotions and negative emotions.

The model was implemented into FAtiMa [ 11 ], a modular architecture designed to develop emotional agents. 2.2

Goals and values in emotional appraisal

Following an established practice in virtual agents, the emotional component of the MEA model relies on the paradigm of appraisal theories of emotions [ 25, 19 ]. In particular, the MEA model translates the appraisal process described by OCC [ 1 ] in terms of goal and value processing. In practice, the appraisal process generates a desirability or (undesirability ) variable when an agent's goal is achieved (or not achieved); it generates a probability variable depending on the probability that an agent's plan succeeds; nally, it generates a praiseworthiness (or blameworthiness ) variable when an agent's value is balanced (or put at stake)by the execution of some action. Given the appraisal variables, then, the a ect generation process generates emotions according to the following domain-independent rules: { Joy (or Distress) if a desirable (undesirable) appraisal variable is generated; { Hope (or Fear) if a likelihood appraisal variable is high (or low); { Pride (or Shame) if praiseworthy (blameworthy) appraisal variable is generated and the responsibility is self-caused; { Admiration (or Reproach) if praiseworthy (blameworthy) appraisal variable is generated and the responsibility is other-caused.

The intensity of the emotions is based on multiplicative relationship between the importance of values and goals, the e ort made (i.e. the cost of the plan) and the probability of success of the plan.

When both appraisal variables for values and goals are generated, the following compound emotions arise: Grati cation (Joy and Pride), Gratitude (Joy and Admiration), Remorse (Distress and Self-Reproach), Anger (Distress and Reproach).

Thanks to the anticipatory emotional appraisal, the behavior of the agent is compelled by its moral values: for example, consider an agent who desires to eat a chocolate candy, but should steal in order to obtain one. In this case, the action that achieves the agent's goal threatens the moral dimension of the agent putting at stake the value \honesty". The emotional state governs the con ict that may arise from the agent's desires and moral values: the agent will opt for stealing only if the positive emotions generated by the perspective achievement of the goal of eating the chocolate candy outbalance the negative, moral emotions generated by the blameworthy action of stealing. The same rules are employed in the agent also for the appraisal of the events and the actions executed by self or others (Appraisal and A ect Processes in Fig. 1). 3

Narrative Scenarios

The experiment described by Battaglino and Damiano in [ 5 ] is based on a testing procedure that compares the predictions of the MEA model about the narrative characters with the expectations of human subjects. The experiment included three narrative scenarios, created with the help of a drama expert, and inspired by well known literary works. An implementation of the MEA agent model [ 10 ] was employed to model the behavior of a character in each narrative situation, in order to guarantee that the described characters' behavior and emotions were aligned with the choices and the emotional states predicted by the model. For each scenario, two di erent courses of action (namely, two plans) were generated by manipulating the priority of the values of the agent, but only one matched the actual character's behavior in the narrative situation that inspired the scenario (for simplicity, it was termed the \right" plan).

For each scenario, the subjects were asked to act and feel according to the characters' beliefs and values (as if they were doing practice in an acting school), choosing between the two courses of action generated by the system and selecting the appropriate emotions for the chosen option. The actions and emotions selected by the testers were compared with the actions and the emotional states generated by the model, in order to assess its validity and coverage. The experiment revealed that the predictions of the agent model substantially matched the expectations of the testers about the characters. Also, a comparison with the scale of values declared by the testers revealed that it has not in uence on their expectations about the character's behavior and emotions [ 5 ].

Scenario 1. Wallace and uncle George's roses: Wallace and Charlie are

cousins. They live in the country where uncle George has a nursery of precious roses he brings to gardening contests. Uncle George is very jealous of his roses. Charlie wants to make a gift to his girlfriend and asks Wallace to give him the key of the nursery to get one.

In this scenario, Wallace has to choose whether to be loyal to his cousin Charlie or to uncle George. The system nds two plans: the plan giving key contains the action of giving the key to Charlie, thus deceiving uncle George, and puts at stake the value Honesty (priority 7:0), so Wallace will feel Shame emotion for putting at stake this value; the plan refusing key contains the action of refusing to give the key to Charlie and it puts at stake the value Loyalty to Charlie (priority 8:5), so Wallace will again feel Shame.

Assuming that the two plans have the same probability of success, Wallace's anticipatory appraisal leads him to choose the plan with the highest EER: in any case, Wallace will feel Shame, but will choose the course of actions that brings him to a state in which the Shame intensity is lower, giving key: he gives the key to Charlie and feels Shame for putting at stake the value Honesty. Scenario 2. At school!: Tom is bullied by his classmate Pier. Pier has taken from Tom the role of director of the school newspaper, putting around lies about him. A few days later, Tom sees that Pier has forgotten his backpack with all his stu in the locker room. Tom digs in Pier's backpack and nds evidence that Pier copied the class test. Tom is now uncertain about what to do, whether to take revenge against Pier or to pass through this situation.

In this scenario, Tom has to choose if he wants to take vengeance or not. The system nds two plans: the plan revenging contains the action of humiliating Pier and puts at stake the value Pity (priority 7:5), so Tom will feel Shame, but it also brings back to balance the value Justice (priority 8:5) put at stake by Pier, and satis es the goal of being the director again, so Tom will feel Joy (for satisfying his goal) and Pride (for restoring his value Justice) as well (which, combined, give the Grati cation emotion); the plan letting go contains the action of letting it and has no e ects on Tom's goals and values: If Tom performs this plan, the situation doesn't change and the value Pity is not put at stake: since the value Justice was put at stake by Pier before, Tom may still feel residual Anger towards Pier. Assuming that the two plans have the same probability of success, Tom chooses to execute the plan revenging and takes revenge on Pier: Tom will feel Shame for putting at stake the value Pity, but will also feel Shame and Grati cation (Joy and Pride).

Scenario 3. A di cult choice: New York, 2003. Mark and Lucy are married and they have a beautiful baby. Lucy has agreed to go a couple of years to Italy for the job of her dreams: working as a curator of a famous art gallery in Rome. Mark, however, has always wanted to be a judge in New York. Just when Lucy has o cially accepted her job in Italy, Mark gets the seat as a judge in New York. Mark's desires are of being with his family and having the work of his dreams as well. Now, he has to choose whether to have the job or to stay with his family. The system nds two plans: the plan staying contains the action of staying in New York without his family, which satis es the goal of being a judge and threatens the goal of being with the family, putting at stake the value Family (8:0 priority), so he will feel Joy for satisfying his goal of being a judge, but also Shame (for putting at stake a value) and Distress (for threatening the goal of being with the family), which combine into the emotion of Remorse. The plan leaving contains the action of going to Italy, which satis es the goal of being with the family but threatens his goal of being a judge and puts at stake the value Happiness (priority 8:5), so, again, Mark will feel Shame for putting at stake a value, Distress for threatening his goal of being a judge, but Joy for satisfying his goal of being with the family. The system chooses the plan staying since it puts at stake the value with a lower priority (Happiness), due to the equal importance of success of both goals: in any case Mark will feel Shame, but the anticipatory appraisal would make him choose the course of actions that brings him to a state of a airs in which the Remorse is lower. 4

Testing moral emotion understanding in TBI patients The methodology described in Section 3 was adapted to a new sample made up of a group of TBI patients and a group of neurologically healthy participants. In the experiment, both groups (the test and control groups) were asked to execute the test; then, we compared the answers of the two groups with the predictions of the model and observed the di erences between them. The goal was to gain insight on the moral and emotional functioning of the TBI patients and on the di erences with the group of neurologically healthy subjects (which was known to correspond to the predictions of the agent model from the previous experiment). The MEA agent functioned here as a formally represented model of the understanding of the character's behavior and emotions, validated through the experiment described in Section 3.

Our hypothesis was that the impairment of moral emotion reasoning that characterizes TBI patients would let di erences emerge concerning moral emotions, without necessarily a ecting their capability to reason about the moral implications of actions. Finding such pattern would also con rm the two-fold nature of the utilitarian and emotional nature embedded in the design of the MEA agent. 4.1

Description of the test

Test procedure and material. The experiment was conducted online, via a text-based web interface. For each scenario, a short text introduced the character and her/his values, then a narrative situation was described that put a stake the character's values. The scale of values was presented to participant not in a numerical format but with a gurative scale, in order to make the values priorities apparent at rst glance (see Fig. 3). The task of identifying the expected course of action and emotions for the character was introduced to the participants as a game: the participants were asked to pretend they were exercising identi cation in an acting class, in order to leverage their capability to take the point of view of the character and behave \as if" they were the in the character's shoes. For each scenario, a pair of alternative actions were submitted to the participants, who also had to select a set of appropriate accompanying emotions. Each narrative situation was encoded in formal terms as described in Section 2 and the resulting formalization was employed to generate the behavior of the main character.

Protocol'

Summary Character’s scale of values

Play the game!

First, two TBI patients, whose cognitive pro le was similar to that of other TBI participants, were asked to participate in the study in order to test if their cognitive impairments (i.e. reading di culties, attention de cits) would cause them to nd the tasks di cult or impossible to complete. Therefore, the methodologies were slightly modi ed in their verbal content in order to be simpler and shorter (as patients were very slow in reading, taking more than 60 min. for each test).

Participants. The clinical sample was made up of TBI patients recruited from the Puzzle rehabilitation center in Turin (Italy); the healthy participants were volunteers. 13 TBI patients and 13 control subjects participated in the experiment. TBI patients received the tests from the experimenter, while healthy participants completed the tasks at home, via an online form. The experimenter's help was needed to avoid noise from reading/comprehension/movement impairments often occurring in this kind of patients, as emerged in the aforementioned trials.

Measures. The participants' answers were transformed into scores in order to allow comparisons between groups. For this test, the following variables were taken into account, and a score was computed for each: { Action total score (score=1 each time participants guessed the agent's `right' action according to the MEA model); { Emotion total score (score=1 each time participants chose the emotion predicted according to the MEA model); { Moral emotion score (score=1 each time participants chose the emotion predicted according to the MEA model; only for moral emotions, being them admiration, contempt, grati cation, gratitude, pride, shame, anger, blame); { Non moral emotion score (score=1 each time participants chose the emotion predicted according to the MEA model; only for non moral emotions, being them joy, sadness, hope and fear); { Error total score (score=1 each time participants both chose an emotion in disagreement with the MEA model and missed those predicted by the model); { Moral emotion error score (score=1 each time participants both chose an emotion in disagreement with the MEA model and missed those predicted by the model; only for moral emotions); { Non moral emotion score (score=1 each time participants both chose an emotion in disagreement with the MEA model and missed those predicted by the model, only for non moral emotions). 4.2

Results

We performed statistical tests to compare the performance of the two groups. We compared the means using Mann-Withney test (suitable for di erent-sized samples), using SPSS Statistics (IBM). Sample size was n=13 for TBI patients and n=11 for healthy participants (due to participants' drop-out).

Table 4 summarizes the results we found. No statistically signi cant di erence was found between groups. Nevertheless, the mean di erence for the moral emotion total score is close to signi cance (p. < .063). Mean di erence for the        

Measure  Action score  Emotion total score  Moral emotion score  Non moral emotion score  Error total score  Error moral emotion score  Error non moral emotion score  3.31 (1.109) 

TBI 

M (SD)  Action total score is also worntohnmmentioning; TBI patients don't seem to di er actio moral  oral  emoti from healthy subjects for this ekminotdi of task, thus suggesting it to be more similar ntot emoti on  to a reasoning thaanlscto aonn emotoino nal tottaaslsk. Percentages fornoancmtiooranlecmhoice did not Group di er amon g the towreo  grsocuorpes  (ssecoereT  abcleore5 ); theermroroscroeren otice_aebrrloersdcoiree rence fomrotrahleem_errorscore  Controlsr st sceMneaarni o (\A2t.0s0c hoo1l."36) , wi1t.h09 TBI2.4p5a tients per1f0o.2r7m  ing sligh2t.l6y4  better than the conNtr ol group, 1s1h ould11n  ot su11r prise11a s controls ha11d  to becom1e1 used to the methodSotldo. gies by.6t3h2e m1s.e5l6v7e s. .831  2.067  2.832  1.027 

As tDheeviattiaobn le shows, scores for emotions (both moral and non-moral) are TBI  higher fMoeranT  BI pa2.r1t5i cipa2n.1t5s : al1t.2h3o ugh3.3a8p parently 1s1u.3r1p rising (ac3c.3o1r ding to our hypothesis), this ca1n3 be explaine1d3 by ob13s erving the er1r3o  r scores. For both moral

N  13  13  and non-moral emotions, in fact, healthy participa2.n3t9s4  show lo1w.1e0r9 scores, thus Std.  .801  .801  .832  .768 

Deviation  Total  suggestiMnegan  1.17 

that th2e.08T  BI1p.7a9t ients' bett2e.9r6 performan1c0e.8c3 ould be a3s.c0r0i bed to their tendencNy  to oversele24c t emo2t4i ons (2a4s  it ca2n4 be seen in Ta2b4 le 6). For e2x4a mple, in the scenarioSt\dA. t scho.o7l1"7,  T1B.2I50p art.i8c1i6p an1t.s54s6e lected gra2t.i5t9u9d e, pride1.a1n03d  admiration as conseDqeuvieatniotn emotions for the plans, while none of the controls did. However, w e believe larger samples are needed to further assess this di erence.   5  Discussi on Our work moves from the fact that, according to current literature, people that su ered from traumatic-brain injury tend to exhibit impaired moral functioning, even in the chronic phase of the injury. By \impaired moral functioning" we mean the meta-cognitive and interpersonal aspects of human behavior (e.g. being emphatic). With respect to the original experiment [ 5 ], our experiment was performed with TBI patients and with a control group of neurologically intact subjects. The goal of our experiment was to compare the expectations of TBI individuals about the narrative characters with the behavior and emo  FTiagb.le6 .2: OEvmeortvioienw chooficteh feor ethmeo ttwioon gsrosueplesc. te d by the TBI and control group for each s cenario (the rst column of each scenario shows the emotion type predicted by the sqyusateli me m).oFzrioonmi htaonpnoto scbeolttot oi mso,gtgheetti TrsBtI gcrhoeu hpanisnno osnc-emltoo rl’aalzieomneo tgiiounssta,?t hsecesneacorino dpegrr oup (slicgehntargiore. y) is moral emotions, the third group (dark grey) is compound emotions. tions generated by the MEA agent model, and to compare both with the control group, in search for insight about the nature of the impairment of TBI patients and the moral-emotional reasoning in general. To the best of our knowledge, no contributions thoroughly describe or help assessing these impairments.

Nonetheless, many authors have focused on the di erences for moral judgment that can be found in brain-damaged individuals. Greene [ 14 ] and Martins et al. [ 20 ] illustrate how brain damaged individuals di er from healthy participants for moral personal dilemmas. According to Greene's perspective [ 14 ], personal moral dilemmas are those engaging emotional responses while evaluating the dilemma, whereas non-personal and non moral dilemmas elicit no emotional response. When facing a moral personal dilemma, TBI participants di er from neurologically intact individuals whilst they don't seem to di er when the dilemma elicits no emotion. According to Greene [ 14 ], personal moral dilemmas activate medial front gyrus, superior frontal gyrus while non-personal moral dilemmas engage dorsolateral and prefrontal areas; the same pattern can be observed when evaluating non-moral dilemmas. According to Moretti, Dragone and DiPellegrino [ 22 ], TBI individuals whose brain damage includes areas such as the ventromedial prefrontal cortex are able to perform moral judgements (in terms of right-wrong), but seem compromised in the emotional counterpart. Authors describe this as a \moral emotions selective impairment". Also, Hutcherson and colleagues [ 17 ] described the interplay between utilitarian and emotional appraisal during moral judgment involving to moral values. According to their work, utilitarian and emotional moral appraisals are computed independently and then integrated in a moral value response. We believe the emotional appraisal is what makes TBI patients' performance impaired, thus causing nonstandard moral judgments. Thus our results seem in line with current literature on emotional processing in brain damaged individuals; taken altogether these ndings add further evidence to the suitability of the MEA model previously evaluated [ 5 ]. The di erence we found is, in fact, what could be expected according to current studies on TBI.

We also had the secondary aim of gaining preliminary data for the development of a rehabilitative tool where TBI patients face moral situations, training their ability to perform moral reasoning and reasoning on their moral emotions. In this perspective, the results obtained were encouraging. On the one side, the experiment con rms the feasibility of this type of rehabilitation, accrued by the interest showed by the comments of the TBI patients: A said: \this test allowed me to think about the values that shape my moral judgment, helping me understand how every action comes after deep and elaborate reasoning"; R. believes this test \is important, as it makes my head start". On the other side, the indirect validation of the MEA model it provides opens the way to the exploitation of virtual agents in narrative-based rehabilitation application. The literature about serious games for rehabilitation mainly focuses on the design and implementation of empathetic agents who respond to the emotional states of the patients by providing emotional cues through verbal and non verbal behavior [ 16 ]. Our work goes in a complementary direction: here, we hypothesize the use of virtual agents for narrative games that train the recognition of emotions and empathy in patients who have some impairment in this area. Following an established practice in rehabilitation [ 26 ], virtual characters may be exploited to train patients to appraise moral emotions in self and others. 6

Conclusions

Although it was not possible for us to nd the key features of the impaired moral functioning in people who su ered from TBI described in literature, it is clear how, in order to shed more light on such di erences, research must focus on emotion processing (rather than on reasoning). This follows what is stated by the mental models theory perspective on moral reasoning [ 9 ]: according to this perspective, moral reasoning is just regular reasoning that happens to concern moral issues. The TBI participants included in our study were in a late stage of their rehabilitation, thus their reasoning could be compared to those of neurologically intact individuals; nonetheless they di er from controls more signi cantly in their performance for those tasks requiring (moral or non-moral) emotional processing. We believe this indicates emotions as the key feature to investigate in order to analyze moral impairments following TBI.

We also believe that the interaction with virtual characters whose emotional state includes a moral emotional component could be useful to help TBI patients to train their emotional moral functioning and their understanding of moral (but also non-moral) emotions. So, our future work includes further investigations on the moral and emotional functioning of TBI patients with the help of virtual character by using larger groups of patients and more focused experiments.

References

1. A. Ortony , A.C. , Clore , G. : The cognitive structure of emotions . Cambridge University Press Cambridge ( 1988 )

2. Aristotle: Poetics. Oxford University Press ( 2013 )

3. Aylett , R. , Louchart , S. , Dias , J. , Paiva , A. , Vala , M. , Woods , S. , Hall , L. : Unscripted narrative for a ectively driven characters . IEEE Computer Graphics and Applications 26 , 42{ 52 ( 2006 )

4. Bates , J. , Loyall , A.B. , Reilly , W.S.: Integrating reactivity, goals, and emotion in a broad agent . In: In Proceedings of the Fourteenth Annual Conference of the Cognitive Science Society . Citeseer ( 1992 )

5. Battaglino , C. , Damiano , R.: A character model with moral emotions: Preliminary evaluation . In: 5th Workshop on Computational Models of Narrative . p. 24 ( 2014 )

6. Battaglino , C. , Damiano , R. , Lesmo , L. : Emotional range in value-sensitive deliberation . In: AAMAS . pp. 769 { 776 ( 2013 )

7. Bratman , M. : Intention , Plans, and Practical Reason . Harvard University Press, Cambridge (MA) ( 1987 )

8. Bruner , J.: The narrative construction of reality . Critical Inquiry 18 ( 1 ), 1 { 21 ( 1991 )

9. Bucciarelli , M. , Khemlani , S. , Johnson-Laird, P.N.: The psychology of moral reasoning. Judgment and Decision making 3(2 ), 121 ( 2008 )

10. Cristina Battaglino , Rossana Damiano, J.D. : Telling the di erence between asking and stealing: Moral emotions in value-based narrative characters . In: Seventh Intelligent Narrative Technologies Workshop (INT7)

11. Dias , J. , Paiva , A. : Fatima modular: Towards an agent architecture with a generic appraisal framework . Proceedings of the International Workshop on Standards for Emotion Modeling ( 2011 )

12. Fraassen , B.C.V. : Values and the heart's command . Journal of Philosophy 70 ( 1 ), 5 { 19 ( 1973 )

13. Giovannelli , A. : In Sympathy with Narrative Character , Journal of Aesthetics and Art Criticism , vol. I, pp. 83 { 95 . Wiley Blackwell ( 2009 )

14. Greene , J. , Haidt , J.: How (and where) does moral judgment work? Trends in cognitive sciences 6 ( 12 ), 517 { 523 ( 2002 )

15. Hogan , P.C. : A ective narratology: The emotional structure of stories . U of Nebraska Press ( 2011 )

16. Hudlicka , E.: Virtual a ective agents and therapeutic games . Arti cial Intelligence in Behavioral and Mental Health Care p. 81 ( 2015 )

17. Hutcherson , C.A. , Montaser-Kouhsari , L. , Woodward , J. , Rangel , A. : Emotional and utilitarian appraisals of moral dilemmas are encoded in separate areas and integrated in ventromedial prefrontal cortex . The Journal of Neuroscience 35 ( 36 ), 12593 { 12605 ( 2015 )

18. Lim , M.Y. , Dias , J. , Aylett , R. , Paiva , A. : Creating adaptive a ective autonomous npcs . Autonomous Agents and Multi-Agent Systems 24(2) , 287 { 311 ( 2012 )

19. Marsella , S.C. , Gratch , J.: Ema: A process model of appraisal dynamics . Cognitive Systems Research 10 ( 1 ), 70 { 90 ( 2009 )

20. Martins , A.T. , Fa sca , L.M. , Esteves , F. , Muresan , A. , Reis , A. : Atypical moral judgment following traumatic brain injury . Judgment and Decision Making 7 ( 4 ), 478 ( 2012 )

21. Milders , M. , Ietswaart , M. , Crawford , J.R. , Currie , D. : Social behavior following traumatic brain injury and its association with emotion recognition, understanding of intentions, and cognitive exibility . Journal of the International Neuropsychological Society 14 ( 02 ), 318 { 326 ( 2008 )

22. Moretti , L. , Dragone , D. , Di Pellegrino , G.: Reward and social valuation de cits following ventromedial prefrontal damage . Journal of Cognitive Neuroscience 21 ( 1 ), 128 { 140 ( 2009 )

23. Oatley , K. : A taxonomy of the emotions of literary response and a theory of identi cation in ctional narrative . Poetics 23 ( 1 ), 53 { 74 ( 1995 )

24. Olson , E.: Tragedy and theory of drama , vol. 22 . Detroit: Wayne State University Press ( 1961 )

25. Peinado , F. , Cavazza , M. , Pizzi , D. : Revisiting character-based a ective storytelling under a narrative bdi framework . In: Interactive Storytelling , pp. 83 { 88 . Springer ( 2008 )

26. Riva , G.: Virtual reality in psychotherapy: review . Cyberpsychology & behavior 8 ( 3 ), 220 { 230 ( 2005 )

27. Smith , G.M.: Film structure and the emotion system . Cambridge University Press ( 2003 )

28. Zagalo , N. , Torres , A. , Branco , V. : Emotional spectrum developed by virtual storytelling . In: Virtual Storytelling. Using Virtual Reality Technologies for Storytelling , pp. 105 { 114 . Springer ( 2005 )