Abstract:

This paper discusses a number of challenges that we had to face when creating GameX, a 3D game-with-a-purpose to collect Arabic coreference annotations. The game necessitated hiding composite annotation tasks, coreference annotation, with orthogonal game elements without compromising the quality of the data that was gathered. Additionally, the balance between graphical richness and gaming performance presented further difficulties. Furthermore, one category of issues had to do with Arabic input words that are broken down into separate morphological units during preprocessing, because in Arabic, as in other Semitic languages as well as languages such as Italian, Romanian, Spanish and Turkish, entity mentions can be morphologically incorporated in a word. When separated they might have a negative effect on the text’s readability. The last two issues have been resolved, although, an answer is still waiting for the first challenge.