The COLIEE competition for legal text processing includes four challenge tasks—case law retrieval (Task 1), case law entailment (Task 2), statute law retrieval (Task 3), and statute law entailment (Task 4). These tasks are designed to address practical pressures in the legal information processing domain. In this paper, we present our approach to all four tasks: for case law retrieval, a hybrid method that combines traditional lexical techniques with large language models (LLMs); for case law entailment, a modular retrieval–inference pipeline that integrates lexical and dense retrieval with zero-shot and few-shot LLMs; for statute law retrieval, we cast the problem as a supervised learning task that guides the fine-tuning of LLMs; and for statute law entailment, we employ a domain-specific LLM to identify similarities among instances of case descriptions. For each approach to each task, we summarize previous work and provide an evaluation of the value of our extensions for each task.
Baek et al. (Fri,) studied this question.