Reasoning Transfer for an Extremely Low-resource and Endangered Language: Bridging Languages through Sample-Efficient Language Understanding

Khanh-Tung Tran, Barry O'Sullivan, Hoang D. Nguyen

Accepted to AAAI-26

Overview

This repository contains code and resources for the paper Reasoning Alignment for an Extremely Low-resource and Endangered Language: Separating Reasoning and Language Understanding.

Structure

• ./data: evaluation data, including our contributed dataset LC2024.
• ./src/train: training code, including for the baseline Native-CoT Training and English-Pivoted CoT Training.
• ./src/evaluation: evaluation code, based on the SkyThought repo.
• ./appendix.pdf: technical appendix.

Training

1. Install dependencies:

   cd ./src/train
   pip install -r requirements.txt

2. Run the main script:

   bash scripts/lang_adapt/run_sft.sh # change the paths accordingly

Evaluation

1. Install dependencies:

   cd ./src/evaluation
   pip install -r requirements.txt

2. Run evaluation:

   cd ./src/evaluation/skythought/skythought_evals
   python eval.py --model ${YOUR_MODEL_HERE}$ --evals=aime,irish_aime,LC2024 --tp=1 --output_file=results.txt --temperatures 0.6 --n 64

For more information, refer to the original document from the SkyThought repo.

Citation

TBU