One of my areas of interest and expertise is emotion detection. In this post, I look at the existing theories of emotion, in particular the assumptions and limitations they set on the detection of emotion in text. In the literature on emotion detection, no unified or generally accepted theory of emotion exists. However, there are six theories that have had a significant impact on the field of emotion detection research.  I briefly review each of the six theories, then look at the assumptions each theory contributes particularly to the natural language processing of emotions in text. For a more thorough discussion, see my dissertation (Chapter 2).

Theory 1: Darwin’s evolutionary theory

The first theory of emotion can be traced to Charles Darwin’s evolution theory of emotion in 1872 [1]. Darwin’s theory focuses on the nature of emotion expression, and it states that non-verbal communication, such as body language, movements, and facial expressions, are not only used to communicate meaning but have also been genetically retained because they were useful to ancestors. Darwin also suggested that emotional expressions are initially learned behaviors. The main emphasis of the theory though is the “survival value” of emotions, particularly their universal similarity across races and cultures (illustrated in Figure below).

Theory 2: James-Lange theory

In his 1884 article, “What is an Emotion?”, James argued that bodily changes must come first and that it would be impossible to have emotions without these bodily changes [2]. Similarly, Carl Lange, a Danish professor, emphasized the influence of vasomotor changes to emotional experiences [3]. Because the two scientists similarly emphasized that physiological arousal precedes emotions, their two theories were combined to form one theory, known as the James-Lange (J-L) theory. The theory states that physiological arousal occurs first, and when this arousal is perceived or interpreted, emotion is experienced (illustrated in Figure below). In other words, a stimulus triggers physiological changes in a person’s body, and a person’s brain interprets these physical changes into the appropriate emotion [4].

The Cannon-Bard theory argues against the J-L theory and states that physiological arousal, such as sweating and trembling, occurs simultaneously with emotions. The theory argues that the thalamus is a necessity for experiencing emotion. According to the theory, the thalamus sends messages to the cortex for an interpretation of the emotion, which then generates the subjective feeling of emotion, and simultaneously sends them to the sympathetic nervous system for the appropriate physiological responses, thus producing arousal at the same time [6, 7] (illustrated in Figure below).

Like the Cannon-Bard theory, the Schachter-Singer theory [8] acknowledges that the same pattern of physiological arousal can occur for different types of emotions. Similar to the J-L theory, the Schachter-Singer theory also states that physiological arousal occurs first and provides important feedback for interpretation; however, rather than simply perceiving or interpreting the arousal, Schachter-Singer’s theory suggests that a reason for the arousal must be identified before being able to experience an emotion (see Figure below).

The focus of the cognitive-appraisal theory is that thought and emotion are inseparable [9]. According to this theory, to experience an emotion and respond to it, one must think about the situation they are in. The theory is often believed to provide the missing link that explains the interpretation or perception in the J-L theory. Interpretation is thus explained by cognition, more particularly by appraisal, a term coined by Arnold [10] to represent sense judgments, which are “direct, immediate, nonreflective, nonintellectual, [and] automatic.” Theorists of this perspective, pointed out that depending on the significance for the individual, the appraisal of a situation will automatically trigger an emotion and physiological response as an appropriate response to the stimuli, which can either be immediate, imagined, or remembered [11] (illustrated in Figure below).

Social constructivists view that “emotions are not just remnants of our phylogenetic past, nor can they be explained in strictly physiological terms. Rather, they are social constructions, and they can be fully understood only on a social level of analysis.” [12]. The theory focuses on the systems of culturally specific rules that govern how, when, and by whom particular emotions are to be experienced and expressed [9] (see Figure below). From this perspective, emotions fulfill a social purpose by regulating interactions between individuals. Although the theory differs from Darwin’s and James’ theories, those who support Darwin’s theory have acknowledged the role of culture in regulating emotional displays [13].

Discussion on the assumptions each theory contributes to emotion detection in text

Each of the above six theories agree that emotions are triggered by a stimulus or event (external or internal) that is deemed important to the organism. The theories can be grouped into five main categories: evolutionary, physiological, neurological, cognitive-appraisal, and social constructivism [14]. Although the theories seem to contradict one another, they actually focus on different perspectives of emotions, which Cornelius [14] summarized as follows:

“Neurophysiologists are interested – almost by definition – in the neural organization of emotion, Darwinians are interested in the evolutionary organization of emotion, Jamesians [those following the James-Lange theory] are interested in the bodily organization of emotion (for want of a better term), cognitive-[appraisal] emotion theorists are interested in the psychological organization of emotion, and social constructivists are interested in the social-psychological and sociological organization of emotion.”

With these theories, a natural language processing researcher in emotion detection has to determine which theories and their assumptions provide the most suitable basis for describing emotions in the context of their research.

With the Darwin’s theory, it can be assumed that there are universally recognized facial expressions present when experiencing emotions. It can also be assumed that there are body movements or reactions that are primary in every living being [15]. From this perspective, and considering a textual environment, evidence of facial expressions would be present in texts in descriptions of the face, i.e., “I have sad face” or “she is smiling at me,” or in the use of symbols, e.g., emoticons. Otherwise, there would be no evidence of facial expressions in a textual environment, and thus the emotion could not be identified in accordance with the theory. Hence, the Darwinian Theory is more suited to studies that have access to visual signals.

From the perspective of the J-L theory, it can be assumed that an emotion experience is accompanied by unique patterns of physiological activities, i.e., changes in the autonomic nervous system. In the textual environment, expressive and descriptive words may provide evidence of physiological arousal, as it is not possible to use equipment to measure physiological signals. For example, descriptive phrases such as “my palms are sweating” in reference to nervousness or “I am finding it hard to breathe” in reference to panic provide self-physiological activity descriptions; however, the theory is limited to only subjective feelings which is the component of emotion that the J-L theory focused on [16].

Moreover, based on the neurophysiological perspective, which was represented by the Cannon-Bard theory in the review above, it can be assumed that there are activities in the nervous system that cause some of the emotion experiences and the accompanying physiological arousals. In particular, the emotion experience can occur without an awareness of bodily changes. That is, people can react to the emotional significance of a stimulus before fully understanding the stimulus [17]. The theory also assumes that there are neural circuits that have developed evolutionarily [18]. From a text perspective, this theory makes it more difficult to determine how the nervous system’s activities make assessments of a stimuli because it would be overly complicated to use circuit models as a means to differentiate one emotion from another in text.

From a cognitive-appraisal perspective, which includes the Schachter-Singer theory, it can be assumed that each emotion experience has its own corresponding and unique pattern of appraisal, thought, and mental activity. More specifically, following from the Schachter-Singer theory, it can be assumed that an emotion has been labeled and recognized by an author of a piece of text. As cognitive-appraisals arise from personal conceptions of a situation, identifying an emotion experience becomes a complex task to perform because the uniqueness of each pattern makes it difficult to evaluate it across different people [19]. This challenge is particularly difficult in a textual environment because detailed information is not often available.

Furthermore, from the social constructive perspective, it can be assumed that there are cultural and social factors in play during emotion experiences and expressions. Social processes and cultural norms play significant roles in specifying when emotions are felt and how emotions are expressed [19]. Some emotions are directed towards other people and arise from interactions with them. Although it is apparent that social and cultural norms do affect emotion expression and that there is a need to study emotions in a social context, it is a challenging task in a textual environment because the textual environment might not offer enough background information to obtain accurate results when adopting this perspective.

In summary, each theory presents a different perspective which can guide but also has implications regarding the approach one can use to detect emotions in mediums such as text, audio, video, body or facial movements, etc.

References:

[1] C. Darwin, The expression of the emotions in man and animals, Vol. 526, (University of Chicago press, Chicago, US, 1965). (Originally published in 1872).
[2] W. James, “What is an emotion?,” Mind 9, 188–205 (1884).
[3] C. G. Lange, “The mechanism of the emotions,” in The Emotions,D. Dunlap, ed. (Williams & Wilkins, Baltimore, MD, USA, 1885), pp. 33–92.
[4] W. James, The Principles of psychology (Dover, New York, NY, 1890).[5] J. Walsh, “Theories of emotion,” (12/15/2013), Khan Academy, https://www.
khanacademy.org/video/theories-of-emotion (visited on 2017-03-12).
[6] W. Cannon, Bodily Changes in Pain, Hunger, Fear and Rage: An Account of Recent Researches Into the Function of Emotional Excitement (Appleton-Century, New York, NY, 1929).
[7] P. Bard, “On emotional expression after decortication with some remarks on certain theoretical views: Part I,” Psychological Review 41, 309–329 (1934).
[8] S. Schachter and J. Singer, “Cognitive, social, and physiological determinants of emotional state,” Psychological review 69, 379–399 (1962).
[9] R. R. Cornelius, “Theoretical approaches to emotion,” in Proceedings of the ISCA Workshop on Speech and Emotion (ISCA, 2000), pp. 3–10.
[10] M. B. Arnold, Emotion and personality (Columbia University Press, New York, NY, 1960).
[11] P. C. Ellsworth and K. R. Scherer, “Appraisal processes in emotion,” in Handbook of affective sciences, K. Scherer and H. Goldsmith, eds. (Erlbaum, Mahwah, NJ, 2003), pp. 572–595.
[12] J. Averill, “A Constructivist View of Emotion,” in In Emotion: Theory, Research and Experience, R. Plutchik and H. Kellerman, eds. (Academic Press, New York: NY, 1980), pp. 305–339.
[13] P. Ekman, W. V. Friesen, and P. Ellsworth, Emotion in the Human Face: Guidelines for Research and an Integration of Findings: Guidelines for Research and an Integration of Findings (Pergamon Press Inc., Elmsford, NY, 1972).
[14] R. R. Cornelius, The science of emotion: Research and tradition in the psychology of emotions (Prentice-Hall, Inc, Englewood Cliffs, NJ, 1996).
[15] N. H. Frijda, The emotions (Cambridge University Press, New York, NY, 1986).
[16] K. R. Scherer, “What are emotions? And how can they be measured?,” Social science information 44, 695–729 (2005).
[17] J. E. LeDoux, “Emotion, memory and the brain,” Scientific American 270, 50–57 (1994).
[18] K. R. Scherer, “Psychological models of emotion,” The neuropsychology of emotion 137, 137–162 (2000).
[19] R. A. Calvo and S. D’Mello, “Affect detection: An interdisciplinary review of models, methods, and their applications,” IEEE Transactions on affective computing 1, 18–37 (2010).