Comparative Impact of Intermediate-Task Training on Code-Switched Versus Monolingual Corpora for Zero-Shot Generalization in
Description
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: What is the comparative effect of intermediate-task training on code-switched versus monolingual non-English corpora on the zero-shot generalization capabilities of 10B+ parameter models across divergent language families in the XTREME suite?
Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.1/10.
Notes
Files
paper.pdf
Files
(78.8 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:55093326273c07748082355a6b2a0120
|
78.8 kB | Preview Download |