Arabic dialects annotation using an online game

Modern Standard Arabic is the written standard across the Arab world; but there is an increasing use of Arabic dialects in social media, so this is appropriate as a source of a corpus for research on classifying Arabic dialect texts using machine learning algorithms. An important first step is annotation of the text corpus with correct dialect tags. We collected tweets from Twitter and comments from Facebook and online newspapers, aiming for representative samples of five groups of Arabic dialects: Gulf, Iraqi, Egyptian, Levantine, and North African. Then, we explored an approach to crowdsourcing corpus annotation. The task of annotation was developed as an online game, where players can test their dialect classification skills and get a score of their knowledge. This approach has so far achieved 24K annotated documents containing 587K tokens; 16,179 tagged as a dialect and 7,821 as Modern Standard Arabic.