A UNIFIED 5-language training corpus (en / fr / de / es / ko, byte-vocab V=256) that extends anima-corpus-5lang-unified (v1) by ADDING the register/act/emotion/genre/code-switch slices that the KOSMOS-grounded enrichment analysis ranked as the biggest gaps. v1 is left intact; v2 = v1 surfaces + an enrichment surface.

what v2 adds on top of v1

v1 had two surfaces: encyclopedic wiki (50%) + persona × SNS roleplay (50%). The enrichment… See the full description on the dataset page: https://huggingface.co/datasets/dancinlab/anima-corpus-5lang-unified-v2.