Testing the Assumptions of Active Learning for Translation Tasks with Few Samples
Lorenzo Flores, Cesare Spinoso di-Piano, Ori Ernst, David Ifeoluwa Adelani, Jackie Chi Kit Cheung · 2026
Read Testing the Assumptions of Active Learning for Translation Tasks with Few Samples by Lorenzo Flores, Cesare Spinoso di-Piano, Ori Ernst, David Ifeoluwa Adelani, Jackie Chi Kit Cheung (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-08977.txt.
Description
Flores and four colleagues test the assumptions behind active learning for translation when only a handful of examples can be labeled. Active learning picks samples to annotate by optimizing informativeness and diversity. Yet recent work found it no better than random sampling on language generation with 100 to 500 samples. They show neither informativeness nor diversity tracks test performance in this setting. Instead, the order of training samples and their interplay with pretraining data matter more, so future methods for very few samples must account for those factors.
Catalog record
| CREATOR | Lorenzo Flores, Cesare Spinoso di-Piano, Ori Ernst, David Ifeoluwa Adelani, Jackie Chi Kit Cheung |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42791 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.08977