Intermediate-task training gains and target language data scale in zero-shot XTREME-R 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: To what extent does the scale of target language data (e.g., 100 vs. 1000 samples) interact with the gains from English intermediate-task training for zero-shot performance on XTREME-R, measured by accuracy improvements per language?

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