This workshop was about computational models of (a) emotion generation, and (b) the effects of emotion on cognition, but from a particular angle that emphasizes the interplay between emotion psychology and computational modeling. In short, the workshop was about computational modeling of the interplay between affect and cognition, i.e., computational modeling of affective processes (CMAP).
The workshop’s aim was to better understand the process of computational modeling of emotion, so that the resulting models are traceable (testable), comparable with other models, and so that the psychological grounding of the models is clear (including where the grounding breaks down). These things are needed for two main reasons. First, the diversity in computational models of emotion will continue to grow, and diversity is good if we understand what the exact “diversity” is, and how the models relate to each other in detail. This will help us to better decide what kind of emotion model is needed for what context. Second, if computational models of emotion are viewed as mechanisms-oriented instantiations of psychological theories of emotion, then these should be able to tell us something about emotion in the human species and other animals and generate testable hypotheses, but only if we know exactly how the model is built, and how it is grounded. The SEM2011 workshop aimed to advance the discussion of these two aspects.
The workshop was visited by 25 participants from 7 countries, and has a good mix of senior and junior researchers and psychologically oriented and computation oriented individuals. The workshop consisted of plenary sessions of invited speakers and submitted paper presentations, but the better part of it was filled with lively discussion around the main workshop topics including psychological grounding, model validation, standardization of modeling process, and “super models” (models that include different affective phenomena, such as the interplay between mood and emotion). We thank all of the participants for their active attitude!
A major outcome of the workshop is the realization among all participants that computational models of emotion are psychological experimental tools to gain insight in emotion, and that many of the details needed for developing these models simply are not available in emotion psychology. Hence, instead of asking for these details from psychologists, the models should bring insight. Further, the workshop identified major challenges that need to be addressed to better understand the process of computational modeling of emotion. These include: what are the benefits and pitfalls of taking either a domain specific or an abstract approach; what are issues in using games-like scenarios as complex stimuli to evaluate models and elicit emotions; what are the issues in using a common decomposition of the appraisal process for the evaluation of models, understanding appraisal theories, or for providing a common implementation framework for theories. Can we agree on this decomposition; what are the different types of goals, and how do these relate to utility, desire, desirability, (social) values, and appraisal; can we identify a protocol for the development of a CMAP (why do you need the model, why select a particular appraisal theoretical basis, ways of validating, etc…) with associated questions/assumptions that might be needed for each step towards implementation; what is the function of the model and what are the designer’s goals; what are best practices towards computational modeling of affective processes; what is the relation and interplay between naïve (normal peoples’) and scientific models of emotion. Finally, with respect to the intrapersonal nature of emotion we agreed on the following point of view, that seemed to best represent both the psychologists’ views as well as the computational modelers’ interpretations. Emotion refers to the descriptive label for particular patterns of joint activity of multimodal constituents (e.g., valence, expression, action, arousal, appraisal). The experience of emotion refers to the (conscious) detection (global availability) of such patterns. The intrapersonal function of emotion is a complex feedback signal for a(n) (synthetic) organism to adapt its behavior; each emotion has a different kind of signaling function, some are about anticipated events (fear) while others are about past events (anger).
The Lorentz center was great. Of course it gave important financial support, but on top of that it facilitated everything with friendly, helpful and very available personnel. Further we would like to thank the TU Delft (Catholijn Jonker) and the VU (Jan Treur) for their additional financial support to cover extra travelling costs. Finally, we sincerely thank all the participants for the lively discussions filled with emotion (in all thinkable forms and quality).