📄 [Read the Paper Here](arxiv.org/abs/2412.11750)
📊 [Access the Dataset Here](gitlab.inria.fr/ariabi/cuban...)

Looking forward to presenting our work and engaging with the community at #VarDial2025!

#ALMAnaCH #INRIA
#NLP #VarDial2025 #COLING2025 #SpanishLanguage #AI #Research 5/5

- We introduce the dataset CubanSpVariety, which is, to the best of our knowledge, the first dataset for variety identification in Cuban. 🌴🌍 4/5

🔑 Our Contributions:

- We present a novel method leveraging training dynamics to identify hard-to-classify examples (common examples across varieties). This method can improve variety identification dataset annotation. 3/5

In this paper, we address the problem of the high level of variety overlap in Spanish. This issue is often overlooked, yet it can directly impact models' performance in culturally sensitive tasks like hate speech detection. 2/5