Papers by Alessandro Capotondi
Learning from Wrong Predictions in Low-Resource Neural Machine Translation (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing approaches to Neural Machine Translation use additional linguistic sources and software tools but these are often not available in less favoured languages. |
| Approach: | They propose a pre-training strategy that leverages the relationships and similarities that exist between unaligned sentences to increase the dataset size of endangered and low-resource languages. |
| Outcome: | The proposed approach increases the dataset size of endangered and low-resource languages by the square of the initial quantity, matching the typical size of high-resourced datasets such as WMT14 En-Fr. |