AIDoes adding languages to a model have to reduce performance?
A post explores the theoretical “curse of multilinguality,” focusing on the structure of models’ embedding spaces.
TLDR
A post describes declining performance as more languages are added to a model as the “curse of multilinguality.” It asks whether that observed degradation has to happen in theory, framing the question around embedding-space structure.
Does adding languages to a model have to reduce performance?
A post explores the theoretical “curse of multilinguality,” focusing on the structure of models’ embedding spaces.
TLDR
A post describes declining performance as more languages are added to a model as the “curse of multilinguality.” It asks whether that observed degradation has to happen in theory, framing the question around embedding-space structure.
