EigenDialectos

End-to-end pipeline that classifies Spanish text into eight dialect varieties (Peninsular, Andalusian, Canarian, Rioplatense, Mexican, Caribbean, Chilean, Andean–Bolivian). Combines multi-task contrastive pre-training, angular margins (ArcFace) and linear-eigenmode spectral fingerprints over the embeddings. In progress.

https://github.com/joseluissaorin/eigendialectos