legal-nlp

Introduces LegalPincite, a large-scale legal information retrieval dataset built from CJEU judgments, featuring masked queries, full corpora, and paragraph-level citation annotations to enable multi-level retrieval evaluation.

This paper surveys and benchmarks NLP-based automatic deception detection in legal contexts, comparing fine-tuned transformers and seven LLMs with various prompting strategies across seven datasets. Results show domain sensitivity, with fine-tuned models excelling in general domains and few-shot LLMs competitive in low-resource legal settings.

This paper introduces BLAD, a curated multilingual dataset of 1,484 Bangladeshi legal acts spanning 1799 to 2025, with structured metadata for temporal and cross-lingual legal NLP research.

Introduces DECODEM, benchmark datasets for evaluating automated extraction of corporate governance variables from legal documents using large language models, showing high accuracy for many provisions.

This paper investigates whether Chain-of-Thought (CoT) prompting benefits patent claim generation, finding that implicit CoT (where reasoning is internal) consistently outperforms explicit CoT, which can introduce a verbalization bottleneck that compromises output quality through abstraction of details, disruption of patterns, and error propagation.

PRecG proposes a novel pipeline for legal precedent retrieval by decomposing documents into rhetorical segments, building knowledge graphs for each segment, and learning hierarchical representations using graph neural networks. Experiments on Indian legal data show effectiveness over baselines.

This paper presents an automated pipeline that uses the DeepSeek V3 model to decompose Italian tax-court judgments into individual legal issues structured in XML following the IRAC framework, and includes a hallucination-detection filter using the Linkoln parser to validate citations, validated by expert annotators.

Proposes a tree-of-thoughts inspired extractive-abstractive approach for legal case judgement summarization using LLMs, with experiments on DeepSeek and LLama showing improved summaries over extractive or abstractive methods alone.

Introduces LAUKIN, a dataset of clause pairs from Australia, UK, and India contracts labeled for legal equivalence, and evaluates 12 models achieving 65.11% macro-F1, establishing a challenging benchmark.

HKJudge is the first sentence-level expert-annotated legal discourse corpus for Hong Kong criminal judgments, featuring a two-tier discourse schema and benchmark evaluations of BERT-based and LLM models.

This paper introduces EURO-5K, a sentence-level dataset for extracting reporting obligations from EU legislation, and benchmarks discriminative and generative transformer models under full fine-tuning and parameter-efficient QLoRA. Results show that legal pretraining primarily benefits models with limited adaptation capacity, and all approaches converge around 3K samples.

This paper introduces a KAN-enhanced BiGRU architecture for classifying and summarizing multilingual legal documents from Bangladesh, achieving modest accuracy and ROUGE scores and demonstrating that the KAN block improves classification accuracy over the baseline BiGRU.

Introduces UA-Legal-Bench, a five-task benchmark for evaluating large language models on Ukrainian legal reasoning, built from the Unified State Register of Court Decisions. Evaluates 11 LLMs, revealing task-dependent few-shot effects and the misleading nature of accuracy on imbalanced legal tasks.

This paper presents a fully automated pipeline that transforms court decisions into legal commentaries by extracting, clustering, and summarizing paragraph-level chunks using LLMs, evaluated on German civil code cases.

This paper investigates temporal concept drift in legal judgment prediction by fine-tuning transformer models on Ukrainian court decisions from three epochs defined by geopolitical disruptions. Findings show severe forward degradation, asymmetry in backward transfer, and that chronological continual learning effectively mitigates forgetting while domain pretraining reduces degradation magnitude.

This paper introduces LP-Eval, a rubric and dataset for evaluating legal proposition generation by large language models, with annotations by legal experts. Results show that rubric-guided LLM evaluations align more closely with expert assessments than direct scoring.

The paper introduces IMLJD, a computational dataset designed for analyzing Indian matrimonial litigation, supporting natural language processing and legal analytics research.

This paper constructs the first large-scale citation graph from 100.7 million Ukrainian court decisions, extracting over 500 million citation links. It demonstrates that the citation structure can automatically recover legal domain boundaries and predict legislative importance with near-perfect accuracy, and releases the pipeline and data as open resources.

This paper compares a domain-trained small language model (Olava Extract) against frontier LLMs for structured contract extraction, showing that the specialized model achieves higher F1 scores and dramatically lower cost.