Tag
This paper empirically studies cross-lingual transfer in in-context learning across seven tasks, six models, and typologically diverse languages, showing that fine-tuning based expectations do not consistently apply and offering new heuristics for source language selection.