Technology Category
Natural Language Processing: (Large) Language ModelsMachine Learning: Large Multimodal Models (LMMs)Planning, Routing, and Scheduling: Planning with Language Models
Small Open Models Achieve Near Parity with Large Models in Low Resource Literary Translation at a Fraction of the Cost
🤖 AI Summary
To address the suboptimal performance of open-source small language models in literary translation for low-resource languages, this paper proposes the TINYFABULIST TRANSLATION FRAMEWORK (TF2), an end-to-end, reproducible research framework. Methodologically, we introduce a high-quality synthetically generated English–Romanian parallel corpus; release TF2-12B, a 12-billion-parameter instruction-tuned model; and employ LLM-assisted reference translation generation, two-stage fine-tuning (instruction tuning followed by adapter-based compression), and LLM-powered multidimensional automatic evaluation—integrating BLEU with five human-aligned dimensions. Experimental results demonstrate that TF2-12B achieves fluency and adequacy on par with state-of-the-art proprietary large models, while offering key advantages: full openness, significantly lower training cost, and efficient deployment. Complementing the framework, we publicly release the dataset, model weights, and integrated toolchain to comprehensively support open research in literary translation.
Technology Category
Application Category
📝 Abstract
Literary translation has recently gained attention as a distinct and complex task in machine translation research. However, the translation by small open models remains an open problem. We contribute to this ongoing research by introducing TINYFABULIST TRANSLATION FRAMEWORK (TF2), a unified framework for dataset creation, fine tuning, and evaluation in English-Romanian literary translations, centred on the creation and open release of both a compact, fine tuned language model (TF2-12B) and large scale synthetic parallel datasets (DS-TF2-EN-RO-3M and DS-TF2-EN-RO-15K). Building on DS-TF1-EN-3M (TF1), the largest collection of synthetic English fables to date, we address the need for rich, high quality literary datasets in low resource languages such as Romanian. Our pipeline first generates 15k high quality Romanian references from the TF1 pool using a high performing LLM. We then apply a two stage fine tuning process to a 12B parameter open weight model: (i) instruction tuning to capture genre specific narrative style, and (ii) adapter compression for efficient deployment. Evaluation combines corpus level BLEU and a five dimension LLM based rubric (accuracy, fluency, coherence, style, cultural adaptation) to provide a nuanced assessment of translation quality. Results show that our fine tuned model achieves fluency and adequacy competitive with top performing large proprietary models, while being open, accessible, and significantly more cost effective. Alongside the fine tuned model and both datasets, we publicly release all scripts and evaluation prompts. TF2 thus provides an end-to-end, reproducible pipeline for research on cost efficient translation, cross lingual narrative generation, and the broad adoption of open models for culturally significant literary content in low resource settings.
Problem
Research questions and friction points this paper is trying to address.
Addressing literary translation challenges for low-resource languages like Romanian
Developing cost-effective small open models competitive with large proprietary systems
Creating high-quality synthetic datasets for English-Romanian literary translation tasks
Innovation
Methods, ideas, or system contributions that make the work stand out.
Unified framework for dataset creation and fine-tuning
Two-stage fine-tuning with instruction and adapter compression
Open release of model, datasets, and evaluation scripts
🔎 Similar Papers
No similar papers found.
