Fine-tuning XLM-R on English vs. Mixed-Language Intermediate Tasks for Zero-Shot Cross-Lingual Reasoning Performance

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 fine-tuning XLM-R on English intermediate tasks compare to fine-tuning on mixed-language intermediate tasks (e.g., 50% English, 50% non-English) in zero-shot cross-lingual reasoning performance on XTREME-R, measured by accuracy and inference latency?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.2/10.