Research map / Large language models / Instruction and fine-tuning
Improving Low-Resource Translation with Dictionary-Guided Fine-Tuning and RL: A Spanish-to-Wayuunaiki Study
Manuel Mosquera, Melissa Verónica Robles, Johan R. Portela, Rubén Manrique
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SFT with synthetic dictionary-call examples | Qwen2.5-0.5B-Instruct · Spanish-Wayuunaiki test | avg BLEU: 0.06 -> 3.08 | Improved |
| add GRPO RL stage with BLEU reward after SFT (ablation) | Qwen2.5-0.5B-Instruct · Spanish-Wayuunaiki test | avg BLEU: 3.08 -> 3.42 | Improved |
| give model access to dictionary tool (ablation) | Qwen2.5-0.5B-Instruct · Spanish-Wayuunaiki test | avg BLEU: about 6% relative gain | Improved |
| use CharacTer reward instead of BLEU in RL (ablation) | Qwen2.5-0.5B-Instruct · Spanish-Wayuunaiki test | avg BLEU: 3.42 -> 2.76 | Hurt |
| train RL for 1400 steps instead of 400 (BLEU reward) (ablation) | Qwen2.5-0.5B-Instruct · Spanish-Wayuunaiki test | avg BLEU: 3.16 -> 3.42 | Improved |
| scale from Qwen2.5-0.5B to Qwen2.5-7B (ablation) | Qwen2.5 · Spanish-Wayuunaiki test | avg BLEU: 3.42 -> 4.45 | Improved |
| apply RL to NLLB-3B translation model (ablation) | NLLB-3B · Spanish-Wayuunaiki test | avg BLEU: 0.93 -> 0.93 | No clear effect |
| apply the method to Llama-3.2-1B-Instruct (ablation) | Llama-3.2-1B-Instruct · Spanish-Wayuunaiki test | avg BLEU: 3.15 -> 3.48 (SFT to RL) | Improved |
| tell base model a dictionary exists without examples (ablation) | Qwen2.5-0.5B-Instruct · Spanish-Wayuunaiki test | avg BLEU: 0.83 -> 0.06 | Hurt |
Closest papers by meaning
- Teaching Large Language Models to Translate with Comparison (AAAI 2024 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.88)
- Mufu: Multilingual Fused Learning for Low-Resource Translation with LLM (ICLR 2025 · similarity 0.87)
- Can LLMs Really Learn to Translate a Low-Resource Language from One Grammar Book? (ICLR 2025 · similarity 0.87)
- Meta-Learning Preferences for Multilingual LLM Alignment (NeurIPS 2026 · similarity 0.87)
- When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
