Nheengatu, the Amazonian Língua Geral, has only 6,000–10,000 speakers and is co-official in São Gabriel da Cachoeira, Brazil. Despite this status, the language remains poorly served by evaluated Portuguese–Nheengatu machine translation (MT) resources, limiting access to technology and information for its speakers.
A new paper on arXiv addresses this gap by presenting an open MT system and a reproducible corpus pipeline. The authors describe their work as the largest contribution of its kind for this language pair, though the abstract does not specify whether this refers to the corpus, the system, or the evaluation.
The focus on reproducibility is notable: the pipeline allows others to rebuild the corpus and retrain the system, which could help sustain future work on Nheengatu MT. The paper does not provide evaluation results in the abstract, so the actual quality of the translations remains to be seen in the full text.