2026
EMNLP
EMNLP 2026
GMU WMT 2026 CreoleMT System Description
Abstract
AbstractWe present GMU’s submissions to the WMT26 Creole Language Translation Shared Task, exploring LLM fine-tuning and few-shot prompting for English-Creole language translation. Our experiments show that LLM-based systems outperform NLLB-based systems, while few-shot prompting is particularly effective for the lowest-resource language pairs. We also demonstrate that extracting incidental parallel sentences from archival grammar books improves translation performance.