Impact of Intermediate-Task Benchmarks on Cross-Lingual LLM Generalization in XTREME-R

Intermediate-task training---fine-tuning a pretrained model on an intermediate task before fine-tuning again on the target task---often improves model performance substantially on language understanding tasks in monolingual English settings. We investigate whether English intermediate-task training is still helpful on non-English target tasks. Using nine intermediate language-understanding tasks, we evaluate intermediate-task transfer in a zero-shot cross-lingual setting on the XTREME benchmark. We see large improvements from intermediate training on the BUCC and Tatoeba sentence retrieval tas

Research goal: How does the choice of English intermediate-task benchmark (e.g., SQuAD, GLUE, SuperGLUE, MMLU) influence the cross-lingual generalization of LLMs on XTREME-R, measured by per-language accuracy improvements when fine-tuning with different batch sizes?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 7.8/10.