Adapter-based fine-tuning for efficient zero-shot cross-lingual transfer in intermediate-task training

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: Does adapter-based fine-tuning during intermediate-task training improve zero-shot cross-lingual transfer efficiency (measured in inference speed and F1 score) compared to full fine-tuning for models of varying sizes (100M, 500M, 1B parameters) on PAWS-X and GLUE-X?

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