ES 2012 - 4th International Workshop EMOTION SENTIMENT & SOCIAL SIGNALS

The fourth instalment of the workshop series on Corpora for Research on Emotion (ES? ) held at LREC aims at further cross-fertilisation between the highly related communities of emotion and affect processing based on acoustics of the speech signal, and linguistic analysis of spoken and written text, i.e., the field of sentiment analysis including figurative languages such as irony, sarcasm, satire, metaphor, parody, etc. At the same time, the workshop opens up for the emerging field of behavioural and social signal processing including signals such as laughs, smiles, sighs, hesitations, consents, etc. Besides data from human-system interaction, dyadic and human-to-human data, its labelling and suited models as well as benchmark analysis and evaluation results on suited and relevant corpora are invited. By this, we aim at bridging between these larger and highly connected fields: Emotion and sentiment are part of social communication, and social signals are highly relevant in helping to better understand affective behaviour and its context. For example, understanding of a subject's personality is needed to make better sense of observed emotional patterns. At the same time, non-linguistic behaviour such as laughter and linguistic analysis can give further insight into the state or personality trait of the subject.
All these fields further share a unique trait: Genuine emotion, sentiment and social signals are hard to collect, ambiguous to annotate, and tricky to distribute due to privacy reasons. In addition, the few available corpora suffer from a number of issues owing to the peculiarity of these young and emerging fields: As in no related task, different forms of modelling exist, and ground truth is never solid due to the often highly different perception of the mostly very few annotators. Due to data sparseness, cross-validation without strict partitioning including development sets and without strict separation of speakers and subjects throughout partitioning are frequently seen.