Although there is not an accurate and well-agreed upon definition for emotions; Scherer illustrated emotions in a insightful profound way describing how emotions cognitively emerge in the following: Emotions are episodes of synchronized recruitment of mental resources allowing an organism to adapt to or cope with an event subjectively appraised as being highly relevant to its needs and goals (Scherer, 2004). Many theories discuss how emotions are elicited in our cognitive system. Discrete Emotions theories (Ekman, 1992; LeDoux, 1996; Öhman & Wiens, 2004) argue that emotions are a set of sensory-motor programs; each of these programs consists of a brain circuit linking eliciting cognitions and somatic responses into a single neural system. While Dimension theories (e.g., Russell, 2003) argue that emotions are cognitive labels to physiological activation, which are characterized in terms of broad bipolar dimensions such as valence and arousal (e.g. I feel negative arousal in a context where I’ve been wronged, therefore I must be angry) Invalid source specified.. In the 1980s Appraisal Theories became a major perspective for emotions studies.

It essential to review the history of proposed theories when studying emotions models. The work of (Moors, 2009) reviews many theories mentioned here.

It argues that a stimulus activates bodily response, the experience and feed back of this response produces the emotional response and experience. The intensity and type of emotion relies on the intensity and type of bodily response. Nevertheless, this theory doesn’t describe how bodily responses elicit and map to emotions. This theory was criticized Cannon (1927) on the basis that not all physiological changes cause emotions and that we still experience feelings in the absence of bodily responses.

This theory considered Cannon’s criticism and stated that a stimulus causes a physiological arousal. This arousal is cognitively mapped to the corresponding emotional experience and its type while the intensity of the arousal defines the intensity of the emotion. In this theory, emotional experience is equated to feeling just like in James’.

Studies proved that unconscious cognition is involved in emotion elicitation. Arnold (1960) coined the term appraisal to denote the cognitive processes preceding emotion elicitation. Appraisal theories agree with Schachter in the fact that arousal precedes emotions however they differ in the placing the cognitive component. Appraisal theories state that cognitive processes precede the bodily physical response as these processes map stimuli to emotions whereas conscious attribution of the cause to the emotion follows. Stimuli-emotion mapping could not be extracted due to the different patterns of causes that evoke different emotions in people. Thus appraisal theories define set of variables. Values of these variables form an appraisal pattern for a specific emotion. The goal relevance to the organism is one variable that affects emotion intensity, along with the stimulus. Different theories consider different set of variables. Appraisal theories propose multi-mode for the functioning and processing of appraisal. This model relies on rule-based mechanisms that compute variables and combine them to elicit appropriate emotions, and associative mechanisms (e.g. Clore & Ortony, 2000;) that retrieve previous pervious appraisal patterns.

These theories consider associative methods in emotion elicitations as those theories relate to semantic network models and memories (e.g., Berkowitz, 1990; Bower, 1981; Lang, 1985; Leventhal, 1980, 1984). When an emotional experience is evoked, the stimuli, actions, goals and responses are saved in the memory in a specific schema network for each emotion. Every time a stimulus is encountered, the schema of this stimulus or a similar one gets activated. Schemas might get activated if any of their nodes are activated. For example the activation of a response of a sad face could activate the corresponding sad emotion. These theories make use of the rule-based mechanisms in order to compute the values of stimuli affecting the elicited emotion just like appraisal theories do.

Emotions generation involves processes of stimulus evaluation, which is the focus of appraisal theories, and then translating the evaluation into emotion and consequences, which is the focus of affect program theory (e.g., Ekman, 1992, 2007; Izard, 1977; Panksepp, 1982, 1998, 2000; Tomkins, 1962). The theory states that every emotion has a specific neural circuit serving an evolutionary survival function. For example fear emotion serves survival purpose. Ekman (1992) argues that a neural circuit is triggered after prior appraisal just like the multi-modal. When a circuit is triggered it generate emotional experience, action tendencies and other emotion-related activities. The activation of a neural circuit should exceed the circuit’s threshold or else it won’t get activated. This theory is committed for the view that “basic emotions are the blocks for building emotional life” (Moors, 2009). There is a list of evidences for this theory from which we mention: - Neurological evidence for specialized emotion neural circuits (Panksepp, 1998, 2000) - Existence of emotion specific responses, which are also universal (Ekman, 1972) It is note-worthy that this theory is compatible with the theories previously mentioned.

Barrett (2006b) builds her theory on Russell’s (2003) who argues against the notion that basic emotions are building blocks of emotional life, and he states that emotional variables of valence and arousal are the building blocks. These variables are properties of stimuli, neurophysiological states and of conscious experience. The combination of these variables is called “affective quality” which causes a state called “core affect”. Specific emotions are in this case the categorization of core affect, which are not naturally given rather they socially emerge. Barrett agrees with Russell that emotions are not natural and that stimuli elicit core affect. Barrett states that categorization of core affect is a form of perception that helps shaping the emotional experience. This categorization is affected by previous conceptual knowledge and thus she uses the term conceptual act to denote it. Category representation knowledge is perceptual because it has sensory features, embodied as its activation simulates previous instances of the same category, and situated according to contexts. Emotion categories are essential part of emotional experience in this theory, while in appraisal theories emotion category is a consequent of emotional experience. Like theories of James (1890) and Schachter (1964), this theory equate emotion with emotional experience and hence could be also called feeling theory.

An appraisal is the process of continuously evaluating events resulting in emotions. The event doesn’t cause emotions; the subjective evaluation of an event is responsible for causing and eliciting emotions. In this context, emotions are called “appraisal structures”. Most theories consider these emotions appraisal structure is a set of appraisal component (Kuppens, Van Mechelen, Smits, & De Boeck, 2003). Common appraisal components include Invalid source specified., for example, appraising events as relevant to a goal, evaluating resources for coping with an event, making attributions of causality and responsibility, judging an event’s similarity with a motive or goal, and assessing whether an action falls short of personal and moral standards (Lazarus, 1991; Roseman, 2001; Scherer, 2001). For example, the appraisal structure of anger involves (1) appraising an event as relevant to a goal; (2) appraising the event as incongruent with the goal; (3) judging a threat to one‘s social- or self-esteem; and (4) blaming someone for the threat Invalid source specified..

The appraisal evaluation is not conscious and controlled (Roseman & Smith, 2001). To the contrary, appraisal processes are presumed to be non-conscious and automatic (Smith &Kirby, 2001). Appraisals are fast to adapt with the quickly changing environment (Lazarus, 2001). Appraisal theories resemble the discrete emotion theories in linking elicitation (in terms of appraisal) and body physiological response, but they differ in proposing a richer mapping between appraisal and elicitation on the one hand, and response on the other hand to express human emotions effectivelyInvalid source specified.

Emotions are important to reach a believable and effective interaction with agents. Over long time, the emotional reactions have to comply with the personality of the agent or else the agent would be perceived as randomly reacting. Modeling personality is important aspect to have in long-term companions. The difference between emotions, moods, and personalities is the time periods; emotions last for minutes while moods last for days and weeks but personality lasts for years or lifetime. OCC defines a personality by setting thresholds for emotions so that if the intensity is not higher than the emotion’s threshold it won’t evoke this emotion. This way the same event would be differently reacted to according to the personalities or the set of thresholds of emotions related to this event.

Appraisal theories view interest as an emotion, not as a judgment (Aitken, 1974), Invalid source specified.. Two appraisal components relate to interest appraisal structure, the first one is “novelty check” of the event whether it is new, sudden, unfamiliar, ambiguous, complex, unexpected, or not understood (Scherer 1999). The second appraisal component is coping potential, which appraises the likelihood of the ambiguous and the unfamiliar event to become clear and understood.

Many theories model emotions and related reasoning. One main and well-known model is the OCC theory (Ortony et al., 1988). It is widely used because the authors developed a computational model for emotions Invalid source specified. making it easy to implement in computer science because of its appraisal theory that is clear in describing the theory. OCC defines a group of emotions in 22 categories organized in a specific hierarchy. This model has five main phases to process an event starting with classifying the emotional categories for an event, object or action. The intensities of categories are quantified and used to define the emotional values that will be mapped and expressed by facial expressions or body movements or any other form possible for the agent. Elliott’s Affective Reasoner was the first application to use the OCC theory Invalid source specified.. However, appraisal applications that followed focused more on comprehensive and general techniques for appraisal variables according to knowledge and goals. For example, Affective Reasoner used hand-written rules, while the subsequent EM system (Neal Reilly, 1996) employed general reasoning mechanisms appraisal that functioned over domain-specific knowledge structures Invalid source specified.. Other appraisal theories proposed by Richard Lazarus (1991), Nico Frijda (1987) and Klaus Scherer (2001) are broader and more complete theories that considered emotions continuously influence and are influenced by cognition. One downside of Ortony, Clore and Collins’ model is that it focused on the cognitive appraisal structure of emotions only without balancing with the overall emotion process, this is concretely manifested in computational models using OCCInvalid source specified.. Nevertheless, every theory has its own limitations. For example, it is obvious that irrelevant factors have an effect on the emotional response, yet this is not included in the appraisal process. Appraisal usually depends on the occurrence of certain patterns of action and events in order to generate emotions without considering other factors like the weather (Clore & Palmer, 2009) as they also discuss empirical points and findings that argue against this link between cognitive representations and emotions. Embodied cognition recent work argues that emotions are generated constantly as the dynamics of the environment are constantly changing. Therefore Invalid source specified. Propose that emotions are best seen as embodied, dynamic processes instead of an appraisal of cognitive representations. The work of Invalid source specified. summarizes the current and previous work on emotions models, spots recent criticisms of appraisal models and propose promising relevant work and solutions.

Current work on emotion has revealed that emotions affect our judgments in general. However this doesn’t imply that emotions necessarily affect judgment, rather they act as a source of embodied information evaluating events and objects or whatever is in focus (Clore & Palmer, 2009). Niedenthal’s experiments showed that positive feelings and moods denote positive attributes, events and judgments; and negative feelings signify negative attributes, events and judgments (Niedenthal, 2007). The Affective Montage Principle proposes that affect is a reaction to whatever is currently in mind(Clore, Wyer, Dienes, Gasper, & Gohm, 2001). The Russian filmmaker Kuleshov experimented in this context testing how people perceive the emotional state of an expressionless face of a famous actor followed by either a soup bowl image or a woman in a coffin. Viewer perceived the face as being hungry after seeing the soup bowl and sad after seeing the coffin. This principle describes the way our affect emerges. A believable agent emotional system should consider that effect of emotions is closely related and affecting the attention and goals of the agent. In a series of experiments, participants showed the effect of moods on their task performance as following: - Testing memory: participants were able to reconstruct a more accurate model of the object they were asked to remember more than the other group. Classification of the two groups was according to who was thinking of a happy or a sad event. The happy-event thinkers performed better. - Testing local/global perception: the happy-event thinkers perceived the global view, while the sad-event thinkers focused on the local view. - Testing local/global perception with “chronic egalitarians” (Huntsinger, Sinclair, Dunn, & Clore, 2007) the experiment showed that that happy mood perceived local view while sad mood promoted global view perception. According to (Clore & Palmer, 2009) the reason for this effect is not anatomical but rather psychological. Happy and positive affect appears to promote currently activated thoughts and tendencies, while negative and sad affect inhibits currently activated thoughts and tendencies. One interesting affect is of angry mood, in contrary to what is expected, anger promoted what is active in the brain just like the happy mood does due to the fact that anger awards positive values on one’s own perspective. Another compatible general interpretation of the happy/sad affect is proposed by (Clore & Storbeck, 2006) where they link happy processing to cognition and sad processing to perception. Moreover, a lot of literature is available discussing the same affect. Artificial intelligent agents could exhibit mood and affect influences even without having biological basis.

EMA (Gratch, 2004; Gratch 2003; Marsella, 2006) is a computational model that follows the cognitive-motivational-emotive psychological theory (Lazarus, 1991). In this model the appraisal process assesses certain variable to judge the situation and elicit the related emotion. It depends on the perspective of who is the subject experiencing the event and what is the desirability of this event. Who is responsible for causing the event, the temporal status changeability and controllability of the event along with its likelihood all contribute to eliciting the related emotion while its intensity is calculated according to OCC theory. “Marsella and Gratch also tackle the question of emotion as a moment-by-moment adjustment to the changing demands of a responsive environment. They use their computational appraisal model, EMA, to carefully analyze a naturalistic emotional episode. They illustrate how emotions in their model unfold over time as a function of dynamics in the world and the unfolding time-course of cognitive processes. This approach achieves a form of sequential patterning similar to Scherer’s (2001) Sequential Checking Theory, but in a vary different manner than realized by Mariner et. al. By emphasizing a sharp distinction between the construction of an internal representation of the situation (which may be slow and sequential) and appraisal (which is rapid, parallel and continuously shapes this unfolding construction), they argue that there really is no necessary sequence to appraisal. Rather, this order “emerges” from the agent’s task demands, the underlying dynamics the environment and the sequential nature of some cognitive processes. “Invalid source specified.

After appraising a situation or an event, the agent has to respond to the appraisal by taking or planning for certain actions or changing the agent’s beliefs and hopes. The EMA model includes many strategies to use for coping. The agent can act or plan for an act by itself or by using help and support. In case it can’t induce the appropriate action to do, it can try to reappraise negative events trying to find something positive about them. if there isn’t any positive interpretation for the event then the agent has three choices: either to accept the undesirable and unchangeable threat or to deny the negative appraisal trying to minimize it or to mentally disengage by reducing the desirability of the unachievable goal resulting in reducing the emotional intensity of the event. One last option the agent can still do is to hope and wait for a change in the environment without taking any part in it.

MAMID models the cognitive appraisal process to dynamically generate emotions in response to incoming stimuli, affecting in planning and selecting goals of the current individual. MAMIDInvalid source specified. stage processes are: sensory pre-processing, to extract from sensory input data task-relevant cues; in attention, those cues are filtered according to what is related to the agent’s attention and focus; situation assessment, using individual cues in the assessment; expectation generation, employing current situation to activate possible future states; affect appraiser, eliciting the affective state from a variety of external and internal elicitors; goal selection, selecting crucial goals to pursue; and action selection, deciding the best actions to achieve the desired goal. Situations, expectations and goals are called Mental Constructs that have many attributes like familiarity, novelty, salience, threat level, valence, etc to determine their output. These mental constructs intermediate the process of transforming sensory data into actions towards achieving goals. Belief nets are used to represent long-term memory including previous negative or positive experiences that enables the process of each module or stage. Belief nets are causal relations graphs that represent “If-Then” rules.

The Affect Appraisal module is essential in this model and it incorporates two appraisal theories: multiple-levels (Leventhal & Scherer, 1987; Sloman, 2003; Smith & Kirby, 2001) and multiple-stage (Scherer, 2001). The multi-level theory generates two assessments in turn; a low-resolution assessment in terms of a valence, and a higher-resolution assessment, in terms of four of the basic emotions: anxiety/fear, anger, sadness, happiness Invalid source specified.. The multi-stage account uses both universal elicitors (e.g., novelty, threat level, pleasantness, unexpectedness), to generate the valence, and more cognitively-complex and individual elicitors (e.g., individual history, expectation- and goal-congruence), to generate a categorical assessment using an expanded appraisal. The resulting affective states are used in the rules selecting the agent’s goals and actions; they influence the speed and capacity of MAMID architecture modules; they influence ranking like determining whether a specific cue or situation is processed, or specific goal selected. The last two effects have been a particular focus of this model, and aim to implement emotional effects to cognitive processes. MAMID architecture uses the model suggested by (Hudlicka, 2002; 1998) that maps specific states/traits profiles onto specific architecture parameter values. These parameters control the way the architecture modules process data and elicit emotions. An example is mentioned in the work Invalid source specified.: for a high trait and state anxiety and fear individual, reduced attention and working memory are mapped and reflected onto a limited working memory capacity of the architecture modules resulting in reducing the number of processed data like cues, situations and expectations. Threat bias is modeled by higher ranking of threatening cues and by higher ranking of threatening situations and expectations. Trait-linked structural differences in LTM are supported by allowing selection of alternative LTM clusters, reflecting distinct personality traits (e.g., selection of clusters with greater proportion of threat- and self-related schemas to represent individuals with high trait-anxiety (high neuroticism). Traits also influence the dynamic characteristics of the emotional responses (like maximum intensities). In this architecture it is easy to model different trait profiles and integrate existing profile like “obsessive-compulsive”. Moreover it can easily integrate conflicting emotions and traits for artificial agents just like in humans. This method is a psychotherapy treatment through VR. Below is an example of actions controlled by MAMID in a “Fear of public speaking” application. This example is used in building virtual characters and avatars to treat patients with social phobia. These avatars will evoke the undesired symptoms in the patient (like being negative and aggressive toward the speaking patient). Once these symptoms are identified, the patient is treated wit the appropriate therapeutic interventions (e.g., cognitive restructuring, systematic desensitization).

As we have seen before, emotions are personal cognitive appraisals that follow personal interests, goals and perspectives. Parkinson(Parkinson, 1996) however, connects emotions to social relationships. According to his article, humans are affected by people around them and responding to this affection in emotions. Some emotions are related directly to other people like anger, envy, jealousy, hate and shame. Those emotions are elicited by appraisal of current interpersonal status of the relationship with others. Even when emotions are not directly related to others they could still connected be connected to social aspects like when a person is afraid or anxious about what others think of him. Moreover, cultural aspects affect emotions significance. Appraisal theories assess personal goals and plans of the person in order to elicit appropriate emotion. Personal goals are culturally supplemented such as wealth and fame. This is also noticeable when looking at different cultures and how they experience different emotions in response to similar situations. Culture extends its effect on the display and the way of expression of emotions also. Supporting the previous discussion about importance of one agent to have an understanding of other agents’ emotional models and situations, we see that emotions are looked at like a contagious affect. Each other’s emotions and moods socially affect people. Non-verbal communication affect others and hence result in non-verbal reactions and hence emotions. Emotions affection shows for example in couples (Gottman, 1979) where communication of negative or positive emotions from one party affects the other party. This effect seems to be automatic rather than being reasoned and thought about. Again this point reconfirms the necessity of one agent to automatically consider other agent’s emotions to react accordingly. Interpersonal Function of Emotions Emotions expression conveys the value of an appraised object or event. The reason behind that is probably to achieve interpersonal effect by making such emotional claims (Parkinson, 1996). Chapman (1983) examined children watching cartoon; a child laughed more when he was alone and two children laughed more than a group. This study shows the interpersonal effect on facilitating or inhibiting emotional expression. Emotions are Communicative Parkinson categorizes the interpersonal messages of each core theme emotion that appraisal theories characterize in the below table.

Emotion+++++++++Core relational theme++++++++++++++++++++++Communicative Agenda

Anger+++++++++++Other-blame++++++++++++++++++++++++++++++Take me seriously and give me the respect I deserve!

Fear++++++++++++Danger+++++++++++++++++++++++++++++++++++Help/protect me!

Guilt+++++++++++Self-blame+++++++++++++++++++++++++++++++Forgive me!

Happiness+++++++Success+++++++++++++++++++++++++++++++++ Let’s celebrate!

Hope+++++++++++ Potential for success++++++++++++++++++++Let’s keep on trying!

Love+++++++++++ Desiring or participating in affection+++Be/stay my special ally! Let’s be intimate friends

Sadness+++++++++++Irrevocable loss++++++++++++++++++++++++++++Comfort/reassure me!

Pride+++++++++++Enhancement of ego identity by taking credit for an achievement+++++++++++Adjust your opinion of me upwards in accordance with my achievement!

This categorization is promising if applied in emotional model of artificial agents. If an agent is able to comprehend the message of another agent implied via the latter’s emotional expression, then the interaction would reach a high level of sophistication between agents on the one hand and agents and humans on the other hand.

Öhman, , A., and S. Wiens. “The concept of an evolved fear module and cognitive theories of anxiety.” In Feelings and Emotions, by N. Frijda & A. H. Fischer (Eds.) A. Manstead, 58-80. Cambridge: Cambridge University Press, 2004.

Aitken, P. P. “Judgments of pleasingness and interestingness as functions of visual complexity.” Journal of Experimental Psychology, 1974: 103, 240–244. Bartneck, Christoph. “Characters, Integrating the OCC Model of Emotions in Embodied.” 2002.

Barrett, L. F. (2006b). Solving the emotion paradox: Categorization and the experience of emotion. Personality and Social Psychology Review , 10, 20-46.

Ekman, P. “An argument for basic emotions.” In Cognition and Emotion, 6, 169-2000. 1992.

Elliott, C. “The affective reasoner: A process model of emotions in a multi-agent system (Ph.D Dissertation No. 32). .” Northwestern, IL: Northwestern University Institute for the Learning Scienceso, 1992.

Gratch, Jonathan, Stacy Marsella, and Paolo Petta. “Modeling the Cognitive Antecedents and Consequences of Emotion.” Hudlicka, Eva. “A Computational Model of Emotion and Personality: Applications to Psychotherapy Research and Practice.” In Proceedings of the 10th Annual CyberTherapy Conference: A Decade of Virtual Reality, Basel, Switzerland. 2005.

Hudlicka, Eva. “Modeling Emotion in Symbolic Cognitive Architectures.” AAAI Fall Symposium Series, TR FS-98-03. Menlo Park, CA: AAAI Press., 1998.

Hudlicka, Eva. “This time with feeling: Integrated Model of Trait and State Effects on Cognition and Behavior.” In Applied Artificial Intelligence,, 16:1-31. 2002.

Kuppens, P., I. Van Mechelen, and D. J. M. Smits. “The appraisal basis of anger: Specificity, necessity, and sufficiency of components.” In Emotion, 254–269. 2003.

Lazarus, R. S. Emotion and adaptation. New York: Oxford University Press, 1991.

Lazarus, R. S. “Relational meaning and discrete emotions.” In Appraisal processes in emotion: Theory, methods, research , by A. Schorr, & T. Johnstone (Eds.) K. R. Scherer, 37–67. New York: Oxford University Press, 2001.

LeDoux, J. The Emotional Brain: The Mysterious Underpinnings of Emotional Life. New York, NY: Simon & Schuster, 1996.

Roseman, I. J. “A model of appraisal in the emotion system: Integrating theory, research, and applications.” In Appraisal processes in emotion: Theory, methods, research, by A. Schorr, & T. Johnstone (Eds.) K. R. Scherer, 68–91. New York: Oxford University Press, 2001.

Roseman, I. J., and C. A. Smith. “Appraisal theory: Overview, assumptions, varieties, controversies.” In Appraisal processes in emotion: Theory, methods, research, by A. Schorr, & T. Johnstone (Eds.) K. R. Scherer, 3– 19. New York: Oxford University Press., 2001.

Scherer, K. R. “Appraisal considered as a process of multilevel sequential checking.” In Appraisal processes in emotion: Theory, methods, research, by A. Schorr, & T. Johnstone (Eds.) K. R. Scherer, 92–120. New York: Oxford University Press, 2001.

Scherer, K. R. “On the sequential nature of appraisal processes: Indirect evidence from a recognition task.” In Cognition and Emotion, 13, 763–793. 1999.

Silvia, P. J. “Cognitive appraisals and interest in visual art: Exploring an appraisal theory of aesthetic emotions.” In Empirical Studies of the Arts, 23, 119-133. 2005.

Smith, C. A., and L. D. Kirby. “Toward delivering on the promise of appraisal theory.” In Appraisal processes in emotion: Theory, methods, research, by A. Schorr, & T. Johnstone (Eds.) K. R. Scherer, 121–138. New York: Oxford University Press, 2001.