Cognitive Computational Models of Emotions and Affective Behaviors

Emotions are one of the important subconscious mechanisms that influence human behaviors, attentions, and decision making. The emotion process helps to determine how humans perceive their internal status and needs in order to form consciousness of an individual. Emotions have been studied from multidisciplinary perspectives and covered a wide range of empirical and psychological topics, such as understanding the emotional processes, creating cognitive and computational models of emotions, and applications in computational intelligence. This paper presents a comprehensive survey of cognitive and computational models of emotions resulted from multidisciplinary studies. It explores how cognitive models serve as the theoretical basis of computational models of emotions. The mechanisms underlying affective behaviors are examined as important elements in the design of these computational models. A comparative analysis of current approaches is elaborated based on recent advances towards a coherent cognitive computational model of emotions, which leads to the machine simulated emotions for cognitive robots and autonomous agent systems in cognitive informatics and cognitive computing. DOI: 10.4018/jssci.2012040103 42 International Journal of Software Science and Computational Intelligence, 4(2), 41-63, April-June 2012 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. 2006; Wang & Wang, 2006; Wang, Kinsner, & Zhang, 2009). The LRMB model explains the functional mechanisms and cognitive processes of the brain and the natural intelligence. The main cognitive processes at the perception layer of LRMB are emotion, motivation, and attitude. It is recognized that a crucial component of the future generation of computers, known as cognitive computers (Wang, 2009), is a perceptual engine, which mimics the natural intelligence such as emotions and motivations (Wang, 2010; Wang et al., 2009). It is observed that emotions influence human behavior in several ways. Emotions alter our processes of perception, attention, and decision making, enabling the development of emotionally driven responses (Damasio, 1994; Phelps, 2006; Wang, 2007a; Wang et al., 2006). Also, emotions help to determine the configuration of our facial expressions, body postures, and intonation of voice when interacting with others, revealing, via nonverbal behavior, our internal affective condition and attitudes towards situations, objects, and other individuals (LeDoux, 1989; Scherer, 2003). Because of the multiple facets and components underlying the process of human emotions, it can be approached from a diversity of perspectives. Moreover, due to the nature of this process and its applications, emotions are currently the focus of study in multiple disciplines such as psychology, neuroscience, philosophy, computer science, cognitive sciences, and cognitive informatics (Fellous & Arbib, 2005; Phelps, 2006; LeDoux, 1989; Wang, 2007a, 2007b, 2007c, 2011, 2012a, 2012b; Wang & Wang, 2006; Wang et al., 2006, 2009, 2011). This multidisciplinary inquiry has provided evidence that shows the significance of emotions not only to the rational behavior of individuals, but to achieve more believable and human-like behaviors in intelligent systems. In particular, fields such as psychology and neuroscience have contributed a number of theories and models that explain the diversity of the emotion process. These theories are focused on revealing the mechanisms underlying the process by which humans transform external stimuli into emotional perspectives. Similarly, in fields such as computer science, cognitive informatics, computational intelligence, and artificial intelligence, researchers are interested in the design of formal and computational models of emotions that help improve artificial intelligent systems used for cognitive robots (Wang, 2010), autonomous agents (Wang et al., 2009), and human-computer interactions (Wang, 2007b). In this dual approach, computational modeling technologies are used for testing and refining psychological, biological, and cognitive models, which are further used to support the design of computational models of emotions. The design of autonomous agents (AAs) aimed at embodying human-like behaviors has taken advantage of evidence from studies of human emotions. AAs have been endowed with mechanisms that simulate emotional processes in the architecture of Cognitive Computational Models of Emotions (C2MEs), which are biologically inspired models intended to describe human emotional functions such as the evaluation of emotionally relevant stimuli, the elicitation of emotions, and the generation of fast and deliberated emotional responses. In some cases, C2MEs focus on reproducing specific facets in this process, but in many others, they cover a more complete emotional cycle that goes from evaluation of stimuli to the generation of emotionally adjusted behaviors (Wang, 2007a). Affective behaviors are thus induced in AAs through the embodiment of C2MEs in their cognitive architectures. This type of behavior is an observable consequence of the verbal and non-verbal responses implemented by the agent, which reflect its internal condition, emotions, attitudes, and motivations. Moreover, the implementation of such affective behavior is what enables the attribution of particular emotion labels to the emotional state of the agent, such as happiness, anger, and embarrassment. In this context, the development of C2MEs should be ultimately designed to allow AAs to implement affective behavior. In order to achieve such International Journal of Software Science and Computational Intelligence, 4(2), 41-63, April-June 2012 43 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. objective, various approaches have been considered which have their basis on emotion theories and models elaborated in cognitive and computational sciences. In this paper, we review emotion theories and models originated in the fields of psychology and neuroscience that have extensively inspired the development of C2MEs. Particular instances of C2MEs are analyzed in order to investigate their internal architectures and cognitive functions. We explain how these computational models process perceived stimuli to translate them into cognitive and computational processes and explore the characteristics and mechanisms associated with the affective behavior induced by C2MEs. This paper is structured as follows. The next section reviews theoretical models of emotions from a perspective that reflects their main contributions to the design of C2MEs. Section 3 investigates the internal workings of some representative instances of C2MEs. Then, Section 4 examines the nature of the affective behavior developed by AAs. A comparative analysis and discussion about C2MEs is presented in Section 5. Finally, concluding remarks are given in Section 6. 2. COGNITIVE MODELS OF EMOTIONS Given the nature of C2MEs and their human-centered applications, the design of such models is not only based on computational technologies, but also on findings contributed by multiple disciplines concerned with the understanding of the human emotion processes. Most C2MEs have been designed under the influence of theoretical models elaborated in fields such as psychology, cognitive science, neuroscience, and cognitive informatics (Gebhard, 2005; Marsella & Gratch, 2009; Velásquez, 1998; Wang, 2007b). The following subsections review some of the most influential theories applied in the design of C2MEs, thus providing a deeper understanding of the nature and theoretical foundations of C2MEs. 2.1. Hierarchical Theories of Emotions In order to understand the domain of human emotions, many classifications have been proposed (Wang, 2007a). A widely accepted classification categorizes emotions as primary and secondary (or basic and non-basic), which derives from the assumption that there is only a small set of basic emotions (Lewis et al., 1989). Emotions may be classified into the classes of primary and secondary ones. Primary emotions are supposed to be innate, instinctive, and with an evolutionary basis. Particular instances are fear, anger, and happiness. The eliciting conditions of some primary emotions have been identified and corresponding facial expressions are uniquely recognized across people in various cultures (Lewis et al., 1989; Ekman, 1999). On the other hand, secondary emotions are learned through experience. Instances of this class of emotions are embarrassment, guilt, shame, and pride. This type of emotions is often considered as derived from combinations of primary emotions. The eliciting patterns of secondary emotions and related facial expressions are dependent on individuals’ culture and educational background (Lewis et al., 1989). Primary emotions induce reactions that are essential for the individual’s survival, while the secondary ones induce appropriate reactions in social situations in which many environmental factors are involved. Although there is not a commonly accepted list of primary and secondary emotions, Ekman (1999) proposes a set of six basic emotions known as anger, disgust, enjoyment, fear, sadness, and surprise. It has been suggested that other emotions may be considered as basic, such as contempt, shame, relief, and embarrassment. Damasio (2003) organizes emotions into three general categories: background emotions, primary emotions, and social emotions. The emotions in the first category are supposed to be generated by simple regulatory processes in terms of state of 44 International Journal of Software Science and Computational Intelligence, 4(2), 41-63, April-June 2012 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. being, although these have not much influence on the behavior of the individual. Damasio (2003) identified a nesting principle to explain how complex emotions are composed of simpler ones. This principle essentially suggests that background emotions are the basic building blocks of primary emotions, and that primary emotions are basic building blocks of the social ones. In a recent study, a hierarchical model of emotions is developed by Wang (2007a). It is found that human emotions at the perceptual layer may be classified into two opposite categories: pleasant and unpleasant. Various emotions in the two categories can be categorized at five levels according to their strengths of subjective feelings as shown in Table 1, where each level encompasses a pair of positive/negative or pleasant/unpleasant emotions. Table 1 indicates that the human emotional system is a binary system that interprets or perceives an external stimulus and/or internal status as pleasant or unpleasant (Wang, 2007). Although there are various emotional categories at different levels, the binary emotional system of the brain provides a set of pairwise universal solutions to express human feelings. For example, angry may be explained as a default solution or generic reaction for an emotional event when there is no better solution available; in opposite, delight is another default emotional reaction. In cognitive computational models the aforementioned classifications of emotions have been often adopted. For instance, since autonomous agents are usually intended to interact with humans, secondary emotions provide them a mechanism for showing social abilities and generating and expressing learned emotions supposed to arise on diverse social situations and events. Also, the blend of basic emotions is a mechanism widely employed as the origin of these secondary emotions, and the behavior of autonomous agents is often in accordance with the pattern behaviors established for primary emotions (Becker-Asano & Wachsmuth, 2010; Velásquez, 1998). 2.2. Appraisal Theories of Emotions Appraisal theories of emotions explain the elicitation and differentiation of emotions on the basis of the relationship between individuals and their environment (Frijda, Kuipers, & Schure, 1989; Ortony, Clore, & Collins, 1990; Scherer, 2001; Smith & Lazarus, 1990). Appraisal theories assume that emotions arise from the evaluation of situations, objects, and agents existing in the environment and which directly or indirectly impact the goals, plans, and beliefs of the individual. This evaluation of the individual-environment relationship is carried out using a Table 1. The hierarchy of emotions (Wang, 2007a)