Low-Resource Methods for Hawaiian Machine Translation

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, short

Summary

Nolan Brophy and Winston Wu investigated low-resource machine translation for ʻŌlelo Hawaiʻi (Hawaiian), a critically endangered Polynesian language, in their 2026 paper. They compiled a corpus of publicly available Hawaiian-English bitext and evaluated neural sequence-to-sequence models and large language models. To combat data scarcity, the researchers employed backtranslation, multilingual training with parallel corpora from related languages, and dictionary entries. Experiments revealed that multilingual training substantially improved model performance, particularly when incorporating bitext from other Polynesian languages. Notably, fine-tuned large language models failed to surpass mBART, indicating that smaller, simpler models remain effective for low-resource machine translation tasks.

Key takeaway

For NLP engineers developing machine translation systems for endangered or low-resource languages, you should prioritize multilingual training, especially by incorporating parallel corpora from related languages. Your efforts might yield better results with established models like mBART rather than fine-tuning large language models, which did not outperform simpler alternatives in this context. Consider data augmentation techniques like backtranslation and dictionary integration to address data scarcity effectively.

Key insights

Multilingual training and simpler models like mBART are effective for low-resource machine translation of endangered languages.

Principles

• Multilingual training boosts low-resource MT.
• Related language bitext enhances performance.
• Simpler models can outperform LLMs.

Method

Researchers compiled Hawaiian-English bitext, applied backtranslation, multilingual training with related Polynesian languages, and dictionary entries to neural sequence-to-sequence models and LLMs.

In practice

• Prioritize multilingual training for MT.
• Explore mBART for low-resource tasks.
• Utilize dictionary entries for data.

Topics

• Low-Resource Machine Translation
• ʻŌlelo Hawaiʻi
• Multilingual Training
• Data Augmentation
• mBART
• Endangered Languages

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• I spent 4 days training an AI model to understand my country. It barely works — Low-resource NLP projects face significant challenges in data collection, translation, and accurate labeling, often yielding modest baselines.
• Ancient Greek to Modern Greek Machine Translation: A Novel Benchmark and Fine-Tuning Experiments on LLMs and NMT Models — A new AG-MG parallel corpus and benchmark significantly advance Ancient Greek to Modern Greek machine translation.
• Low-resource Language Discrimination Towards Chinese Dialects with Transfer learning and Data Augmentation — Transfer learning and data augmentation effectively overcome scarce annotation resources for low-resource Chinese dialect discrimination.
• Data Synthesis and Parameter-Efficient Fine-Tuning for Low-Resource NMT: A Case Study on Q'eqchi' Mayan — Synthetic data effectively bootstraps NMT structural learning for low-resource languages, but requires authentic data for semantic depth.
• AI-assisted cultural heritage dissemination: Comparing NMT and glossary-augmented LLM translation in rock art documents — Glossary-augmented LLMs significantly improve terminology accuracy in specialized translation without sacrificing overall quality.
• Linguistic Issues in Computational Linguistics for Non-English Languages — English-centric LLM design creates systematic linguistic weaknesses for non-English languages across multiple levels.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.