A multilabel dataset for distinguishing Bosnian, Croatian, Montenegrin, and Serbian

<p>This dataset contains files used in the VarDial 2024 Shared Task on Distinguishing Between Similar Languages - Multiple Labels for the Bosnian - Croatian - Montenegrin - Serbian (BCMS) subtask.</p> <p>The starting point for this dataset is the one published by Rupnik et al. (2023). It contains geolocated data from the BCMS linguistic area collected from Twitter (rebranded as X).<br>Each instance contains the full tweet production of a single user, which was manually annotated for the user's country.<br>The original annotation was single-label, and it was produced by a single annotator. In the version of the data produced here, the test and dev sets were reannotated by multiple annotators, in a multi-label setting. For the details on the reannotation process, please see Miletić and Miletić (2024). We have also excluded retweets from the original data, as these represent reproduced content from a different user account and may not be representative of the language use of the user themselves.</p> <p>For details on the shared task, we refer you to Chifu et al. (2024).</p>