Reproduce “Learning-From-Mistakes Prompting for Indigenous Language Translation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
You-Cheng Liao, Chen-Jui Yu, Chi-Yi Lin, He-Feng Yun, Yen-Hsiang Wang, Hsiao-Min Li, Yao-Chung Fan
This research from National Chung Hsing University develops and evaluates a progressive, multi-stage prompting framework, including Learning-from-Mistakes (LFM) prompting, to enable Large Language Models (LLMs) to translate extremely low-resource indigenous languages from Chinese. The LFM strategy achieved the highest BLEU3 scores across several Taiwanese indigenous languages, demonstrating that LLMs can effectively learn to translate languages unseen during their pre-training through iterative error feedback and in-context learning.
Key insights
- LLMs can function as 'language compilers' for unseen indigenous languages by leveraging external tools like parallel examples and word-level dictionaries through sophisticated prompting.
- Progressive prompting strategies, particularly incorporating Chain-of-Thought and Learning-from-Mistakes, are crucial for enhancing LLM understanding of complex syntactic and semantic structures in target languages.
- Explicit error feedback mechanisms (LFM) significantly contribute to refining translation quality by guiding the LLM to correct its initial mistakes and internalize specific linguistic patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.