DRILL Shared Task 2025: The Challenge of Deep Retrieval in the Expansive Legal Landscape

Abstract

Abstract: This paper provides an overview of the DRILL Shared Task, a Vietnamese legal information retrieval challenge organized under the Vietnamese Language and Speech Processing workshop.
Over the two-month competition period, more than 50 teams participated, contributing a total of
1,255 submissions to the leaderboard. While most teams adopted a standard retrieve-then-rerank
pipeline complemented by a final fine-grained processing stage, the top-performing teams distinguished themselves by constructing learning-to-rank models enriched with diverse features, including those derived from large language models (LLMs). These carefully engineered methods delivered strong results, outperforming baseline systems. However, our error analysis reveals that current
systems struggle with questions involving commonsense knowledge, extremely long context, and
temporal relations, suggesting avenues for future work.
Keywords: Vietnamese NLP, Legal Information Retrieval, Article Retrieval, Shared Task, Deep
Learning.