G^2C-MT: Graph-Guided Context Selection for Document-Level Machine Translation
Researchers have developed a new method called G^2C-MT for document-level machine translation that models discourse dependencies using a lightweight graph. This approach represents paragraphs as nodes in a graph, with relationships based on semantic similarity, adjacency, and keyword overlap. A depth-biased random walk over this graph samples context paths to prompt large language models for translation, improving accuracy and robustness across various domains and LLMs. AI
IMPACT Introduces a novel graph-based approach to improve context selection in document-level machine translation, potentially enhancing LLM performance on complex texts.