Figurative language

Looking at Dr. Sheldon Cooper one could figure out the difficulty for a machine of dealing with figurative language, even more whether we work with emotive language. This is the main aim of SemEval-2015 Task 11: Sentiment Analysis of Figurative Language in Twitter.

http://alt.qcri.org/semeval2015/task11/

We have helped Dr. Paolo Rosso to collect a large Twitter dataset and Dr. Tony Veale’s team annotated tweets using CrowdFlower with 7 annotations each. The sentiment score for each is a weighted mean of annotator scores, where the weights are calculated as a function of annotator reliability (which in turn is a measure of how well an annotator does on the gold-standard tweets that are interlaced into the task).

Training data for this task (8000 figurative tweets annotated with sentiment scores in the range -5…+5) is now available.

Happy researching!