Yuzhen Shi, Huanghai Liu, Yiran HU, Song Gaojie, Xu Xinran, Yubo Ma, Tianyi Tang, Li Zhang, Qingjing Chen, Feng Di, Wenbo Lv, Weiheng Wu, Kexin Yang, Sen Yang, Wei Wang, Rongyao Shi, Qiu Yuanyang, Yuemeng Qi, Zhang Jingwen, Sui Xiaoyu, Yifan Chen, Zhang Yi, An Yang, Bowen Yu, Dayiheng Liu, Junyang Lin, Weixing Shen, Bing Zhao, Charles L. A. Clarke, HU Wei. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

As the scale and capabilities of Large Language Models (LLMs) increase, their applications in knowledge-intensive fields such as legal domain have garnered widespread attention. However, it remains doubtful whether these LLMs make judgments based on domain knowledge for reasoning. If LLMs base their judgments solely on specific words or patterns, rather than on the underlying logic of the language, the “LLM-as-judges” paradigm poses substantial risks in the real-world applications. To address this question, we propose a method of legal knowledge injection attacks for robustness testing, thereby inferring whether LLMs have learned legal knowledge and reasoning logic. In this paper, we propose J&H: an evaluation framework for detecting the robustness of LLMs under knowledge injection attacks in the legal domain. The aim of the framework is to explore whether LLMs perform deductive reasoning when accomplishing legal tasks. To further this aim, we have attacked each part of the reasoning logic underlying these tasks (major premise, minor premise, and conclusion generation). We have collected mistakes that legal experts might make in judicial decisions in the real world, such as typos, legal synonyms, inaccurate external legal statutes retrieval. However, in real legal practice, legal experts tend to overlook these mistakes and make judgments based on logic. However, when faced with these errors, LLMs are likely to be misled by typographical errors and may not utilize logic in their judgments. We conducted knowledge injection attacks on existing general and domain-specific LLMs. Current LLMs are not robust against the attacks employed in our experiments. In addition we propose and compare several methods to enhance the knowledge robustness of LLMs. All code can be found at the link.

Haitao Li, Junjie Chen, Jingli Yang, Qingyao Ai, Wei Jia, Youfeng Liu, Kai Lin, Yueyue Wu, Guozhi Yuan, Yiran Hu, Wuyue Wang, Yiqun Liu, Minlie Huang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Recently, a number of repository-level code generation benchmarks-such as CoderEval, DevEval, RepoEval, RepoBench, and Long-Code-Arena-have emerged to evaluate the capabilities of large language models (LLMs) beyond standalone benchmarks like HumanEval and MBPP.Thus, a natural question is, would LLMs have similar performance in real world coding tasks as their performance in these benchmarks?Unfortunately, one cannot answer this question, since these benchmarks consist of short completions, synthetic examples, or focus on limited scale repositories, failing to represent real-world coding tasks.To address these challenges, we create RE-POCOD, a Python code-generation benchmark containing complex tasks with realistic dependencies in real-world large projects and appropriate metrics for evaluating source code.It includes 980 whole-function generation tasks from 11 popular projects, 50.8 % of which require repository-level context.REPOCOD includes 314 developer-written test cases per instance for better evaluation.We evaluate ten LLMs on REPOCOD and find that none achieves more than 30% pass@1 on REPOCOD, indicating the necessity of building stronger LLMs that can help developers in real-world software development.In addition, we found that retrieval-augmented generation achieves better results than using target function dependencies as context.

Prompt compression for Retrieval-Augmented Generation (RAG) often fails by treating all retrieved information uniformly. This undifferentiated approach neglects the critical distinction between foundational core knowledge and illustrative practical knowledge, a failure especially damaging in hierarchical domains like law where essential principles can be discarded for redundant details, diminishing information gain.

In recent years, Large Language Models (LLMs) have been widely applied to legal tasks.To enhance their understanding of legal texts and improve reasoning accuracy, a promising approach is to incorporate legal theories.One of the most widely adopted theories is the Four-Element Theory (FET), which defines the crime constitution through four elements: Subject, Object, Subjective Aspect, and Objective Aspect.While recent work has explored prompting LLMs to follow FET, our evaluation demonstrates that LLM-generated four-elements are often incomplete and less representative, limiting their effectiveness in legal reasoning.To address these issues, we present JUREX-4E, an expert-annotated fourelement knowledge base covering 155 criminal charges.The annotations follow a progressive hierarchical framework grounded in legal source validity and incorporate diverse interpretive methods to ensure precision and authority.We evaluate JUREX-4E on the Similar Charge Disambiguation task and apply it to Legal Case Retrieval.Experimental results validate the high quality of JUREX-4E and its substantial impact on downstream legal tasks, underscoring its potential for advancing legal AI applications.

Junkai Liu, Yujie Tong, Hui Huang, Bowen Zheng, Yiran Hu, Peicheng Wu, Chuan Xiao, Makoto Onizuka, Muyun Yang, Shuyuan Zheng. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Despite significant advancements in large language models (LLMs) that enhance robot agents' understanding and execution of natural language (NL) commands, ensuring the agents adhere to user-specified constraints remains challenging, particularly for complex commands and long-horizon tasks. To address this challenge, we present three key insights, equivalence voting, constrained decoding, and domain-specific fine-tuning, which significantly enhance LLM planners' capability in handling complex tasks. Equivalence voting ensures consistency by generating and sampling multiple Linear Temporal Logic (LTL) formulas from NL commands, grouping equivalent LTL formulas, and selecting the majority group of formulas as the final LTL formula. Constrained decoding then uses the generated LTL formula to enforce the autoregressive inference of plans, ensuring the generated plans conform to the LTL. Domain-specific fine-tuning customizes LLMs to produce safe and efficient plans within specific task domains. Our approach, Safe Efficient LLM Planner (SELP), combines these insights to create LLM planners to generate plans adhering to user commands with high confidence. We demonstrate the effective-ness and generalizability of SELP across different robot agents and tasks, including drone navigation and robot manipulation. For drone navigation tasks, SELP outperforms state-of-the-art planners by 10.8% in safety rate (i.e., finishing tasks conforming to NL commands) and by 19.8% in plan efficiency. For robot manipulation tasks, SELP achieves 20.4% improvement in safety rate. Our datasets for evaluating NL-to-LTL and robot task planning will be released in github.com/lt-asset/selp.

Retrieval-augmented generation (RAG) has proven highly effective in improving large language models (LLMs) across various domains. However, there is no benchmark specifically designed to assess the effectiveness of RAG in the legal domain, which restricts progress in this area. To fill this gap, we propose LexRAG, the first benchmark to evaluate RAG systems for multi-turn legal consultations. LexRAG consists of 1,013 multi-turn dialogue samples and 17,228 candidate legal articles. Each sample is annotated by legal experts and consists of five rounds of progressive questioning. LexRAG includes two key tasks: (1) Conversational knowledge retrieval, requiring accurate retrieval of relevant legal articles based on multi-turn context. (2) Response generation, focusing on producing legally sound answers. To ensure reliable reproducibility, we develop LexiT, a legal RAG toolkit that provides a comprehensive implementation of RAG system components tailored for the legal domain. Additionally, we introduce an LLM-as-a-judge evaluation pipeline to enable detailed and effective assessment. Through experimental analysis of various LLMs and retrieval methods, we reveal the key limitations of existing RAG systems in handling legal consultation conversations. LexRAG establishes a new benchmark for the practical application of RAG systems in the legal domain, with its code and data available at https://github.com/CSHaitao/LexRAG.

This paper introduces JuDGE (Judgment Document Generation Evaluation), a novel benchmark for evaluating the performance of judgment document generation in the Chinese legal system. We define the task as generating a complete legal judgment document from the given factual description of the case. To facilitate this benchmark, we construct a comprehensive dataset consisting of factual descriptions from real legal cases, paired with their corresponding full judgment documents, which serve as the ground truth for evaluating the quality of generated documents. This dataset is further augmented by two external legal corpora that provide additional legal knowledge for the task: one comprising statutes and regulations, and the other consisting of a large collection of past judgment documents. In collaboration with legal professionals, we establish a comprehensive automated evaluation framework to assess the quality of generated judgment documents across various dimensions. We evaluate various baseline approaches, including few-shot in-context learning, fine-tuning, and a multi-source retrieval-augmented generation (RAG) approach, using both general and legal-domain LLMs. The experimental results demonstrate that, while RAG approaches can effectively improve performance in this task, there is still substantial room for further improvement. All the codes and datasets are available at: https://github.com/oneal2000/JuDGE

In legal case retrieval, existing work has shown that human-mediated conversational search can improve users’ search experience. One of the key problems for a practical conversational search system is how to ask high-quality clarifying questions to initiate conversations with users and understand their search intents. Previous works demonstrated that human-annotated external domain knowledge (such as event schemas) can improve the legal utility of clarifying questions generated by large language models. However, these methods are restricted to specific law systems or languages and cannot be generalized to others. To this end, we propose to generate context and domain-specific questions with LLMs without external annotations or knowledge by extracting information from top-retrieved documents given the current conversation context. Specifically, we construct a conversational legal case retrieval system CARQ that iteratively selects neighbor candidate case documents from the retrieved list at each conversation step to ask clarifying questions. We pretrain CARQ to capture the differences between legal cases and employ the reward augmented maximum likelihood to optimize the system directly for retrieval metrics. Extensive automated and human evaluations on three widely adopted legal case retrieval datasets demonstrate the superior effectiveness of our approach as compared with the state-of-the-art baselines.

Legal case retrieval is an important task in information retrieval that aims to retrieve relevant cases for given query cases. Conversational search paradigms have been shown to improve the search experience in legal case retrieval. However, there are two challenges in applying conversational search to legal scenarios. Firstly, legal search conversations often focus on different parts of legal case documents, but existing models struggle to capture the complex structural information and extract accurate relevance signals. Secondly, collecting large-scale conversational search datasets is costly, making it difficult to build reliable conversational legal case retrieval models. To address these challenges, we propose a Structure-Aware Matching Model (SAMM) for conversational legal case retrieval. SAMM extracts matching signals between conversational utterances and segments of the legal cases to incorporate structural information. We decouple the conversational search task into three subtasks and design pre-training tasks to overcome the lack of training data. Additionally, we create ConvLegal, the largest conversational legal case retrieval dataset to the best of our knowledge, for better evaluation of different methods. We train and evaluate SAMM and baselines on both a public dataset (CLCR) and ConvLegal. Experimental results demonstrate that SAMM outperforms existing models in legal case retrieval and conversational search.

Similar case retrieval (SCR) is a representative legal AI application that plays a pivotal role in promoting judicial fairness. However, existing SCR datasets only focus on the fact description section when judging the similarity between cases, ignoring other valuable sections (e.g., the court's opinion) that can provide insightful reasoning process behind. Furthermore, the case similarities are typically measured solely by the textual semantics of the fact descriptions, which may fail to capture the full complexity of legal cases from the perspective of legal knowledge. In this work, we present MUSER, a similar case retrieval dataset based on multi-view similarity measurement and comprehensive legal element with sentence-level legal element annotations. Specifically, we select three perspectives (legal fact, dispute focus, and law statutory) and build a comprehensive and structured label schema of legal elements for each of them, to enable accurate and knowledgeable evaluation of case similarities. The constructed dataset originates from Chinese civil cases and contains 100 query cases and 4,024 candidate cases. We implement several text classification algorithms for legal element prediction and various retrieval methods for retrieving similar cases on MUSER. The experimental results indicate that incorporating legal elements can benefit the performance of SCR models, but further efforts are still required to address the remaining challenges posed by MUSER. The source code and dataset are released at https://github.com/THUlawtech/MUSER.

Legal case retrieval is a special IR task aiming to retrieve supporting cases for a given query case. Existing works have shown that conversational search paradigm can improve users' search experience in legal case retrieval. One of the keys to a practical conversational search system is how to ask high-quality clarifying questions to initiate conversations with users and understand their search intents. Recently, Large Language Models, such as ChatGPT and GPT-4, have shown superior ability in both open-domain QA and conversations with human. Thus it is natural to believe that they could be applied to legal conversational search as well. However, our preliminary study has shown that generating clarifying questions in legal conversational search with SOTA LLMs (e.g., GPT-4) often suffers from several problems such as duplication and low-utility contents. To address these problems, we propose LeClari, which leverages legal event schema as external knowledge to instruct LLMs to generate effective clarifying questions for legal conversational search. LeClari is constructed with a prompt module and a novel legal event selection module. The former defines a prompt with legal events for clarifying question generation and the latter selects potential event types by modeling the relationships of legal event types, conversational context, and candidate cases. We also propose ranking-oriented rewards and employ the reward augmented maximum likelihood (RAML) method to optimize LeClari directly based on the final retrieval performance of the conversational legal search system. Empirical results over two widely adopted legal case retrieval datasets demonstrate the effectiveness of our approach as compared with the state-of-the-art baselines.