LegalSim-PT: Building a Dataset for Legal Document Simplification in Portuguese Leveraging Linguistic Metrics

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, medium

Summary

LegalSim-PT is introduced as the first large-scale Portuguese dataset specifically designed for legal document simplification. This dataset addresses the growing interest in document-level simplification, which requires maintaining fluency, conciseness, and coherence across entire texts, often integrating summarization techniques. Developed by Arthur Scalercio, Maria José Finatto, and Aline Paes, and presented at PROPOR 2026, LegalSim-PT was constructed using data augmentation strategies combined with readability, semantic similarity, and diversity metrics to select high-quality document pairs, reducing the need for extensive manual evaluation. The creators performed a detailed analysis of the dataset's surface features, compared it to existing simplification corpora, and validated its quality through automatic metrics, linguistic indicators, and human evaluations. Additionally, two baseline models were fine-tuned on LegalSim-PT, demonstrating improved performance in document-level simplification.

Key takeaway

For NLP Engineers working on Portuguese language processing, LegalSim-PT offers a critical resource for advancing document simplification, especially in the legal domain. Your efforts to develop or improve text simplification models can directly benefit from fine-tuning on this specialized dataset, potentially leading to more accurate and contextually appropriate outputs. Consider integrating LegalSim-PT into your model training pipelines to enhance performance beyond sentence-level simplification.

Key insights

LegalSim-PT is the first large-scale Portuguese legal document simplification dataset, created using linguistic metrics and data augmentation.

Principles

• Document simplification requires preserving fluency, conciseness, and coherence.
• Linguistic metrics can reduce reliance on manual evaluation for dataset creation.

Method

LegalSim-PT dataset creation involved data augmentation, followed by selection of document pairs using readability, semantic similarity, and diversity metrics. Quality was assessed via automatic metrics, linguistic indicators, and human evaluations.

In practice

• Fine-tune models on LegalSim-PT for improved Portuguese legal text simplification.
• Apply data augmentation with linguistic metrics for dataset creation in other languages.

Topics

• Legal Document Simplification
• Portuguese Language Processing
• LegalSim-PT Dataset
• Linguistic Metrics
• Data Augmentation

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Auditing the Evaluators: How Far Can Automatic Evaluation Go in Assessing Portuguese Financial Texts? — Automatic and LLM-as-a-judge metrics struggle with Portuguese financial texts, requiring domain-specific evaluation.
• Discovery of Legal Patterns in Civil Petitions via LLM-Based Fact Extraction and Density Clustering — LLM-based fact extraction significantly improves legal document clustering by reducing noise and enhancing semantic coherence.
• PorTEXTO: A European Portuguese Benchmark for Visual Text Extraction — Specialized multilingual data significantly improves European Portuguese OCR performance over model size or resolution.
• Concepts and measures of bureaucratic constraints in European Union laws from hand-coding to machine-learning — Models learn from manually annotated text to analyze EU law constraints.
• RoLegalGEC: Legal Domain Grammatical Error Detection and Correction Dataset for Romanian — RoLegalGEC provides the first Romanian legal grammatical error dataset and evaluates neural models for its use.
• TransLaw: A Large-Scale Dataset and Multi-Agent Benchmark Simulating Professional Translation of Hong Kong Case Law — Multi-agent LLM systems can achieve high-quality, cost-effective legal translation by simulating human workflows.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.