Search for: All records

Smiling and laughter are typically associated with amusement. If they occur under negative emotions, systems responding naively may confuse an uncomfortable smile or laugh with an amused state. We present a passive text and video elicitation task and collect spontaneous laughter and smiles in reaction to amusing and negative experiences, using standard, ubiquitous sensors (webcam and microphone), along with participant self-ratings. While we rely on a state-of-the-art smile recognizer, for laughter recognition our transfer learning architecture enhanced on modest data outperforms other models with up to 85% accuracy (F1 = 0.86), suggesting this technique as promising for improving affect models. Subsequently, we analyze and automatically predict laughter as amused vs. negative. However, contrasting with prior findings for acted data, for this spontaneously elicited dataset classifying laughter by emotional valence is not satisfactory. 

Abstract Arousal evoked by detecting a performance error may provide a mechanism by which error detection leads to either adaptive or maladaptive changes in attention and performance. By pairing EEG data acquisition with simultaneous measurements of pupil diameter, which is thought to reflect norepinephrinergic arousal, this study tested whether transient changes in EEG oscillations in the alpha frequency range (8–12 Hz) following performance mistakes may reflect error‐evoked arousal. In the inter‐trial interval following performance mistakes (approximately 8% of trials), pupil diameter increased and EEG alpha power decreased, compared to the inter‐trial interval following correct responses. Moreover when trials were binned based on pupil diameter on a within‐subjects basis, trials with greater pupil diameter were associated with lower EEG alpha power during the inter‐trial interval. This pattern of association suggests that error‐related alpha suppression, like pupil dilation, reflects arousal in response to error commission. Errors were also followed by worse next‐trial performance, implying that error‐evoked arousal may not always be beneficial for adaptive control.