Organizers
Carlo Strapparava and Rada Mihalcea

Web page: http://www.cs.unt.edu/~rada/affectivetext

Motivation
This task is intended as an exploration of the connection between lexical semantics and emotions. All words can potentially convey affective meaning. Every word, even those that are apparently neutral, can evoke pleasant or painful experiences due to their semantic relation with emotional concepts or categories. While some words have emotional meaning with respect to an individual story, for many others the affective power is part of the collective imagination (e.g. words such as "mum", "ghost", "war").

This latter group of words are particularly interesting, because their affective meaning is part of common sense knowledge and can be detected in the linguistic usage. For this reason, we believe it is important to study the use of words in textual productions, and possibly their co-occurrence with words in which the affective meaning is explicit. Several previous studies in linguistics and psychology have considered research issues related to the affective lexicon. For example Ortony et al. [Ortony et al., 1987] distinguishes between words directly referring to emotional states (e.g. "fear", "cheerful") and those having only an indirect reference that depends on the context (e.g. words that indicate possible emotional causes such as "killer" or emotional responses such as "cry").

The automatic detection of emotion in texts is becoming increasingly important from an applicative point of view. Consider for example the tasks of opinion mining and market analysis, affective computing, natural language processing for user-interfaces (e.g. e-learning environments, such as educational/edutainment games). Possible beneficial effects of emotions on memory, attention, and in general on fostering creativity are also well-known in psychology. Finally, news web sites are already very popular and automatic classification of news along emotive dimensions could be useful and interesting.

These characteristics make the news headlines particularly suitable for use in an automatic emotion recognition setting, as the affective/emotional features (if present) are guaranteed to appear in these short sentences.

The structure of the task is as follows:

• Corpus: News headlines, extracted from news web sites (such as Google news, CNN) and/or newspapers. In the case of web sites, we can easily collect a few thousand titles in a short amount of time.
• Objective: Provided a set of predefined emotion labels (e.g. joy, fear, surprise), classify the titles with the appropriate emotion label and with a valence indication (positive/negative).

  The emotion annotation and the valence labeling will be regarded as two separate subtasks, and therefore a team can choose to participate in only one or both annotation tasks.

The task will be carried out in an unsupervised setting, and consequently no training will be provided. The reason behind this decision is that we want to emphasize the study of emotion lexical semantics, and avoid biasing the participants toward simple "text categorization" approaches. Nonetheless supervised systems will be not precluded from the participation, and in such cases the participating teams will be allowed to create their own supervised training sets.

The task organizers will provide in advance the set of emotion labels and a development corpus. The timeline of the task will follow the general Semeval-2007 timeframe as follows:

• [three weeks before the deadline]: a development corpus of approximately 250 headlines annotated for emotions will be provided
• [one week before the deadline]: the data set of approx. 2,000 news headlines will be provided
• [deadline]: the participants submit the headlines annotated with emotion labels and/or a valence indication (positive/negative); as done in previous Senseval evaluation, the participating teams are free to choose their own submission deadline, as long as it falls within the window of time imposed by Semeval-2007.