Sequential Task Integration for Cross-Lingual Alignment in XTREME-R: Efficiency Trade-offs

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 combination of multiple intermediate English tasks in a sequential training pipeline impact zero-shot cross-lingual alignment scores on XTREME-R subsets, compared to using a single intermediate task, and what is the trade-off in training efficiency (measured in FLOPs or throughput)?

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