Rui0828/Learning-From-Mistakes-Prompting

LoResMT@ACL 2024: Learning-From-Mistakes Prompting for Indigenous Language Translation – A feedback-driven approach to enhance low-resource translation. # Learning-From-Mistakes Prompting for Indigenous Language Translation This repository contains the codebase for the method described in our paper: **"Learning-From-Mistakes Prompting for Indigenous Language Translation,"** presented at **LoResMT 2024**. Our method, *Learning-From-Mistakes Prompting (LFM)*, is a feedback-driven framework that enhances low-resource machine translation by iteratively refining translations based on error analysis and targeted adjustments. This significantly improves translation performance for indigenous language datasets. For more details, please refer to the paper: - LoResMT@ACL 2024: View on ACL Anthology - arXiv: Read on arXiv By making this code publicly available, we hope to facilitate further research and encourage collaboration in the field of low-resource language machine translation. --- ## Table of Contents 1. Introduction 2. Installation 3. Datasets 4. Usage 5. Results 6. Citation 7. License --- ## Introduction Machine translation for indigenous languages faces unique challenges such as limited datasets, complex linguistic structures, and diverse dialects. This repository provides the code to reproduce the results of our paper, proposing three key methods: 1. **KNN-Prompting with Retrieved Prompting Context (RPC):** Enhances context by leveraging similar examples. 2. **Chain-of-Thought (CoT) Prompting:** Guides the model through a structured reasoning process before producing translations, enabling more accurate linguistic outputs. 3. **Learning-From-Mistakes (LFM) Prompting:** Iteratively refines translations using feedback-driven error corrections. Together, these methods bridge gaps in low-resource translation, making machine translation more accessible and effective for indigenous languages. --- ## Datasets Our framework is designed to be adaptable to any **low-resource language** where a **parallel corpus** can be obtained. A parallel corpus includes aligned linguistic data between a high-resource language …