A Fine-Grained Annotated Multi-Dialectal Arabic Corpus

Proceedings of Recent Advances in Natural Language Processing, pages 198–204, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_023 198 A Fine-Grained Annotated Multi-Dialectal Arabic Corpus Anis Charfi Information Systems Carnegie Mellon University in Qatar acharfi@andrew.cmu.edu Wajdi Zaghouani College of Humanities and Social Sciences Hamad Bin Khalifa University Qatar wzaghouani@hbku.edu.qa Syed Hassan Mehdi Information Systems Carnegie Mellon University in Qatar smehdi@andrew.cmu.edu Esraa Mohamed Information Systems Carnegie Mellon University in Qatar emohamad@andrew.cmu.edu Abstract We present ARAP-Tweet 2.0, a corpus of 5 million dialectal Arabic tweets and 50 million words of about 3000 Twitter users from 17 Arab countries. Compared to the first version, the new corpus has significant improvements in terms of the data volume and the annotation quality. It is fully balanced with respect to dialect, gender, and three age groups: under 25 years, between 25 and 34, and 35 years and above. This paper describes the pro- cess of creating the corpus starting from gathering the dialectal phrases to find the users, to annotating their accounts and re- trieving their tweets. We also report on the evaluation of the annotation quality using the inter-annotator agreement mea- sures which were applied to the whole cor- pus and not just a subset. The obtained re- sults were substantial with average Cohens Kappa values of 0.99, 0.92, and 0.88 for the annotation of gender, dialect, and age respectively. We also discuss some chal- lenges encountered when developing this corpus. 1 Introduction As the popularity of natural language process- ing (NLP) based tools increases, there is a rising need for language resources such as annotated cor- pora to develop NLP tools. In this regard, Ara- bic lags behind other languages due to several rea- sons. First, Arabic is complex at various levels of linguistic representation (phonology, orthogra- phy, morphology, and syntax). Even though na- tive Arabic speakers prefer using Modern Stan- dard Arabic (MSA) for official communications, we are witnessing with the popularity of social media a rise in the use of dialectal Arabic for in- formal online interactions such as those found in blogs, forums, chats, tweets, posts, etc. (Mubarak and Darwish, 2014; Bouamor et al., 2018). The content written by the users in social media sites can reveal some characteristics and attributes about them, which is the main focus of author pro- filing research. However, the lack of Arabic lan- guage resources limits that research on author pro- filing for the Arabic language in particular dialec- tal Arabic. We present in this paper ARAP-Tweet 2.0, which is a large-scale manually-annotated multi- dialectal Arabic corpus. A first version of this corpus was presented in (Zaghouani and Charfi, 2018a,b) and it was extended significantly in terms of data volume, number of users, and annotation quality. ARAP-Tweet 2.0 covers dialects from 17 Arab countries and 15 regions. The number of users per region is 198, including a total of about 3000 users and approximately 5 million tweets. All users’ accounts were manually annotated with respect to the dialect, gender, and age. Thereby, we distinguished three age groups: …

Content truncated to 500 words