Ekman Theory Of Emotion

[Chadwick Bosse]

Thankyou for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Emotions have always been a source of reflection and interest from human beings: they have studied the philosophical, scientific, artistic and literary perspectives on emotions for centuries1,2,3. The numerous theories that emerged within various disciplines attempt to explain the origin, function, and other aspects of emotions, such as the relationship between the behaviour of the subject feeling the emotion, and the surrounding environment. Emotion theories span across categorical emotional models, dimensional emotional models, as well as other, more recent ones.

Dimensional emotional models define emotions according to predefined measures, which typically include valence, arousal, and control. Acknowledged dimensional theories can be distinguished into circumplex models12,13, and vector models14. While according to the circumplex model the dimensions of arousal and valence are distributed in a circular pattern, in the vector model the emotional direction is determined by an underlying arousal and a binary choice of valence (positive or negative).

In this line, along with existent literature on the matter [e.g., 24],our research focus is whether emergent or undefined emotions can be described with traditional models. For this reason, we have used both a corpus (Entity-Level Tweets Emotional Analysis) annotated with Ekman's theory emotions and a corpus (The Dictionary of Obscure Sorrows) annotated with newly created emotions, to establish the human subject agreement rate when annotating sentences using classical emotional labels.

Among the various well-known emotional models that define "basic" emotions understood as universal and possibly innate, the six emotions defined as "basic" by Ekman1,4 were selected for this investigation.

In order to test the hypothesis, we have administered a standard questionnaire to a community of people with the same social and cultural background, using a cognitive method, giving particular importance to alexithymia, described as a characteristic or personality trait characterised by a processing disorder affective-emotional that causes difficulties in identifying and describing feelings and emotions25,26,27,28.

The term alexithymia was introduced by Sifneos in 197329 and can be literally defined as "without words for emotions". It is a personality trait that involves the emotional sphere of the subject, and it is related to a large number of psychological disorders (e.g., depression30 and schizophrenia31) as well as physical diseases (e.g., psychosomatic illness32). The alexithymic subjects present both cognitive (e.g., difficulty in recognizing, describing and distinguishing emotions) and affective (e.g., emotionalizing) deficits33,34.

While real-life emotional situations reflect a dynamic process that can never be completely reduced to a finite set of categories, yet quantifying the experiential states and the elements that compose them is necessary for research aimed at a rigorous analysis of cognition for designing computational systems.

Our work originally stems from artificial intelligence research, where extensive annotated corpora, formal models, and machine learning are used to detect or predict emotions from text, facial expressions, gestures, etc. Current AI research typically focuses on datasets annotated via the six basic emotions [e.g., 35,36] that are universally shared among humans (like happiness, sadness, disgust, fear, anger and surprise). However, inspired by the Dictionary of Obscure Sorrows (DOS) we wonder if Ekman's emotions may not be enough to explain all the emotions we experience in real life situations. In this sense, in the context of an international project (The SPICE project deals with data-driven cultural engagement for social inclusion and empathy development. Its knowledge graph infrastructure includes multiple emotion theories that have been represented in the SPICE Ontology Network to support the automated analysis and sharing of citizen interpretation about works of art), we are enhancing emotion-oriented datasets and lexical resources, in order to investigate everyday emotions with computational means. How can emotional situations be detected, interpreted and categorised, e.g., in text, conversation, or multimodal interaction? We are representing emotion-oriented resources as knowledge graphs, a common format to create interoperable datasets based on formal semantics. An example of interoperable knowledge graphs from either factual or linguistic resources is Framester37, which uses a formal cognitive frame semantics38.The results from the experiment we present here spot some limits of existing emotion models, and the need to integrate or extend them to represent realistic emotion situations. Starting from the results of this work, an integrated, flexible computational model for emotion situations is under construction.

So far, in our sample 53.5% of subjects are 42 years old or more, and 46.5% 41 or less years old. Informed consent form was obtained from all participants trough the online module. The study was conducted according to the guidelines of the Declaration of Helsinki. All experimental protocols were approved by the Ethics Committee of the Department of Cognitive Science, University of Messina (protocol code COSPECS_04_2023).

Due to the lack of an Italian translation of the PAQ, following a previous study by Becerra et al.44, the English PAQ items have been independently translated into Italian by all the authors of this paper, followed an agreement procedure.

In accordance with the recent literature45 which highlights an important difficulty in verbalising emotions in people with alexithymia, the PAQ was included to verify how well the different levels of alexithymia, in healthy subjects, can influence the personal perception and conceptualization of emotions in daily life.

Of the 20 sentences, 10 were taken from the literary resource The Dictionary of Obscure Sorrows (DOS)18, and 10 from the annotated dataset ELTEA1746, which contains tweets annotated according to Ekman's emotion theory.

ELTEA17 was chosen because its content is close to the entries of The Dictionary of Obscure Sorrows: emotions are not simply described or named, but an emotional context is provided, typically as a situation involving the subject experiencing the emotion; while DOS is a project that aims to create new terms to describe complex emotions and experiences that often don't have a specific name. For example, the term "onism" is used to indicate the sensation of feeling very small compared to the surrounding world, or "sonder" to refer to the awareness that everyone has their own story (Supplementary Table S1 shows the DOS entries that we used in our questionnaire).

Related to the first category (low level of alexithymia), in the annotation of the entries of the ELTEA dataset there is an agreement with the original annotation only in two cases out of ten, and with an agreement percentage higher than 50% only in one case (ELTEA3, for the emotion "Happiness"). In the DOS dataset, only in one case there is an agreement in the annotation of the participants greater than 50% (DOS6 for the emotion "Happiness"). In the other emotional experiences, the agreement is generally between 21.1 and 47.40%.

About the second group (medium level of alexithymia), on the ELTEA dataset, the results show a greater agreement with the original annotation. In fact, only in 4 out of 10 cases there is no agreement. However, in the cases where agreement does occur, it is between 14.80% and 39.50% in 5 out of 6 cases. In fact, the maximum agreement was reached only in the case of the emotion "Happiness" for ELTEA3 (as in the first group) with an agreement degree of 77.80%. Also for the DOS dataset, the second group shows a good agreement in the annotation. For 7 out of 10 emotional experiences the agreement is around 50%, while in the other cases the agreement is between 17.30 and 37%.

Finally, the third category of participants (high level of alexithymia), the results show an agreement with the annotation of ELTEA in 6 cases out of 10, and an agreement with the annotation of the emotions of DOS equal or greater than 50% in half of the cases. For the annotation of ELTEA, the degree of agreement varies considerably, from a minimum of 28.60% to a maximum of 92.90% (ELTEA3 recorded the highest degree of agreement, this data is confirmed also in the other two groups).

A further analysis was performed to compare the frequency in the use of canonical emotions to classify the proposed scenarios in the 3 groups of participants, according to the score provided for all 10 subscales of the PAQ test. Our results show:

This work falls within the research area aimed at investigating the models and formal expressions of emotion description to test their usefulness and efficacy, particularly in relation to people's ability to identify and express emotions themselves.

Starting from the study conducted by Paul Ekman4 on the physical expressions (face, body, etc.) of human emotional experience, that led him to the conclusion that happiness, sadness, anger, fear, surprise and disgust are innate human emotions, our study aims to test the abstraction of Ekman's categorical model through affective labelling.

As described within the scientific literature e.g. Ref.19, affective labelling is the act of describing emotions verbally, either orally or in written form. Talking about our feelings, or using emotional language to describe what upsets us, has mostly been studied for its effects on emotion regulation, as in attenuating our emotional experiences19,20,21. This last aspect goes beyond the research scope of this work, but could constitute a further development of the study.

3a8082e126