Better Literary Translation: A Multi-Aspect Data Generation and LLM Training Approach
Researchers have developed a novel framework for generating high-quality data to train LLMs for literary translation. This approach uses specialized LLMs to create translation references and preference data, focusing on distinct quality dimensions. The resulting models, LitMT-8B and LitMT-14B, show competitive performance on benchmarks and generalize well to new literary works. AI
IMPACT This research introduces a method to improve LLM performance on nuanced tasks like literary translation, potentially enabling more sophisticated cross-cultural communication tools.