On the Need for Context Processing in Affective Computing ○ Michal Ptaszynski, Rafal Rzepka, Kenji Araki Graduate School of Information Science and Technology Hokkaido University Abstract: Research on Emotions within Artificial Intelligence and related fields has flourished rapidly through several years. Unfortunately, in much research the contextuality of emotions is disregarded. In this paper we argue, that recognizing emotions without recognizing their context is incomplete and cannot be sufficient for real-world applications. We present logical underpinnings of this claim and describe some consequences of disregarding the context of emotions. We also present our approach, in which the context of emotions is considered and describe some of the first experiments performed in this matter. The paper is finalized with a discussion on future development and applications of context processing within Affective Computing. 1. Introduction Affective Computing (AC) [1] is a new field within Artificial Intelligence research. It is focused on developing machines capable to understand user emotions and adapt its behavior according to these emotions. Research on emotion recognition, the main stream in the field, has a nearly fifteen-year-long history of attempts to the task of recognizing emotions. In the most popular methods the emotions are recognized from: facial expressions [2], voice [3], language [4] or biometric data [5]. Affective Computing have contributed greatly to the creation of more human-like interfaces and user-friendly systems [6,7,8]. However, in much research questions like "How to use the recognized information?" or "Is the expressed emotion appropriate for the context it appears in?" are often disregarded. Based only on behavioral approaches, methods for emotion recognition ignore the context of emotional expression. Therefore, although achieving good results in laboratory conditions, such methods are often inapplicable in real world tasks. For example, a system for recognition of emotions from facial expressions, assigning "sadness" when user is crying would be critically mistaken if the user was e.g. cutting an onion in the kitchen. In this paper we will emphasize that recognizing emotions while disregarding their contexts causes methods to be insufficient for real-world applications. We will present the reasoning for this statement and describe some examples of how neglecting the context of emotion could cause a fallacy in system performance. We will also present our approach, where we focus both on the expression of emotion and the context it appears in. We will also briefly describe some of the first experiments performed in this matter. We will conclude this paper with a discussion on some future directions and applications of context processing to Affective Computing. 2. Emotions and Intentionality View on emotion phenomena has evolved in time. In Middle Ages, emotions were considered as biological disturbances, passive states with no relation to rational thinking or cognition [9,10]. This approach has been proved wrong in neurobiology where it was showed that emotions and rationality are not separable entities but stem from each other as equally important processes in decision-making [11,12,13], and cognitive processes [14,15,16,17,18]. It was thus reassured that emotions are conscious and intentional mental phenomena [19,20]. Oxford English Dictionary defines intentionality as "the distinguishing property of mental phenomena of being necessarily directed upon an object, whether real or imaginary" [29]. In other words, intentional phenomena are always "about something". As a property of emotional processes, the idea of intentionality implies that emotions necessarily need to be assigned a [formal/intentional] object [18,21,23]. The linguistic-pragmatic reality proves this. When people express emotions they often express them in terms of specifying their objects. For example, we are afraid/proud of something , or happy about something 1 , etc. A function of all specific emotion objects forms a formal object of emotion. The formal objects of emotions have been defined as "axiological properties 2 which individuate emotions, make them intelligible and give them correctness conditions" [24,25,26]. Moreover, Solomon, in his theory of emotions as "engagements with the world" argues that emotions are not only intentional, but they are conscious choices and strategies by which people manage the world. The targets of those strategies are formal objects. Moreover, emotions and their formal objects are necessarily in a causal relation [27]. Formal objects, as sets of axioms defining the emotions, can be further reformulated as properties determining the context of emotions. 3. Emotions and Contextuality The idea of contextuality with an application in logics, as proposed by Gershenson, assumes that "concepts are determined […] by the context they are used in." Gershenson gives a relative notion of a context as follows. "A context consists of the set of circumstances and conditions which surround and determine an idea, theory, proposition, or concept. These circumstances and conditions can be spatial, temporal, situational, personal, social, cultural, ecological, etc." Gershenson, 2002 [28] 1 It has been argued that "moods" are not about anything specific and therefore are not intentional, which would introduce an inconsistency in the definition of emotion and emotion-related processes. Solomon solves this problem by noticing that the objects of moods are not unspecified, but rather moods take as their objects the whole world [21,22]. 2 Or "properties derived from axioms [= here, specific objects of emotions]". TG4-3 26th Fuzzy System Symposium (Hiroshima, September 13-15, 2010) 920