Scaling translation and monolingual fine-tuning analyses across language pairs

Determine whether the observed translation trends and the ability of monolingual fine-tuning to recover the performance of parallel fine-tuning persist across a large number of language pairs, and identify the other factors that explain when monolingual fine-tuning works well and when it does not.

Background

The translation experiments examine a relatively small number of language pairs, with the principal fine-tuning comparison centered on Xhosa and selected additional languages. It therefore remains unresolved how broadly the reported transfer from monolingual fine-tuning to translation applies, and which properties of languages, models, data, or translation directions determine its success.

References

Future work could scale up this experiment to investigate whether these trends hold across a large number of language pairs, and to what degree monolingual fine-tuning can recover the performance of parallel fine-tuning across many language pairs (and what other factors explain when this works well and when it does not).

— The Interlingua Hypothesis: LLMs Translate via a Latent Task-agnostic Feature Space  (2609.00515 - Brinton et al., 1 Sep 2026) in Section Limitations

Conclusive evidence requires human evaluation on additional pairs, which we leave for future work. How DDP-stratified patterns generalize to other typologically distant pairs, to more closely related pairs, and to non-English source languages remains an important open question.

— Discourse Dependency: A Continuous Criterion for Translation Difficulty  (2609.04959 - Kim et al., 4 Sep 2026) in Limitations, subsection “Single language pair”

Within the resolution of this benchmark the two translated variants therefore appear equivalent, and the separation visible in these results is between the translated corpora and the untranslated one rather than between languages. These experiments do not establish why.

— Reinforcement Learning for improving Large Language Models' Catalan text simplification capabilities  (2609.04823 - Domingo et al., 4 Sep 2026) in Appendix B, iDEM Statistical Significance