TinyFabulist Framework Advances English-Romanian Literary Translation
A new framework called TinyFabulist Translation Framework (TF2) has been introduced to improve English-to-Romanian literary translation using small open models. The framework builds on DS-TF1-EN-3M, the largest collection of synthetic English fables, to generate 15,000 high-quality Romanian references via a large language model. A two-stage fine-tuning process is applied to a 12-billion-parameter open-weight model: instruction tuning for narrative style and adapter compression for efficient deployment. Evaluation uses a five-dimension LLM-based rubric covering accuracy, fluency, coherence, style, and cultural appropriateness. This work addresses the challenge of literary translation for low-resource languages like Romanian, aiming to make high-quality translation tools more accessible.
Key facts
- TinyFabulist Translation Framework (TF2) targets English-to-Romanian literary translation
- Uses DS-TF1-EN-3M, the largest synthetic English fable collection
- Generates 15,000 Romanian references from TF1 pool via a high-performing LLM
- Fine-tunes a 12B-parameter open-weight model in two stages: instruction tuning and adapter compression
- Evaluation rubric has five dimensions: accuracy, fluency, coherence, style, cultural appropriateness
- Addresses low-resource language translation with open models
Entities
—