Tag
A research paper from Meta demonstrates that byte-level language models, initially inferior to token-based models, can outperform them as computational resources increase, shown through distilled 1B models trained on up to 1 trillion bytes.
This paper investigates weakly-supervised ByT5 fine-tuning for stress-aware sentence-level Filipino grapheme-to-phoneme conversion, achieving significant improvements in phoneme and character error rates on a manually-corrected test set.
TabletCraft is an open-source system enabling bidirectional Akkadian-English neural machine translation with cuneiform rendering, allowing users to both read ancient tablets and compose new messages in cuneiform. Accepted to the C3NLP workshop at ACL 2026, it reports first published quantitative results for English-to-Akkadian translation.
Presents a neural machine translation system for the severely under-resourced Tangkhul–English language pair, achieving strong BLEU, chrF++, BERTScore, and COMET scores using fine-tuned ByT5-large and mT5-small models.