https://nlp.unibo.it/tag/legal-summarization/Legal SummarizationHugo Blox Builder (https://hugoblox.com)en-usTue, 13 Oct 2026 00:00:00 +0000https://nlp.unibo.it/media/logo.svghttps://nlp.unibo.it/tag/legal-summarization/https://nlp.unibo.it/proposals/legal/summarization/Tue, 13 Oct 2026 00:00:00 +0000https://nlp.unibo.it/proposals/legal/summarization/<h2 id="context">Context</h2> <p>Legal summarization condenses long, technical judicial documents into accurate overviews of their key content. A faithful summary contains no hallucinated facts, so it can support legal analysis and research. Its quality is hard to assess: standard lexical metrics, which count shared words, miss factual consistency and argumentative and legal correctness.</p> <h2 id="objective">Objective</h2> <p>A thesis may improve the generation of legal summaries, their automatic evaluation, or both. It may use knowledge made by human experts, such as guidelines, annotations, and official summaries, given in the prompt or through retrieval-augmented generation (RAG).</p> <figure class="lt-viz lt-svg" style="--lt-svg-min: 640px"> <svg viewBox="0 0 800 280" role="img" aria-labelledby="legalsum-title" xmlns="http://www.w3.org/2000/svg"> <defs><marker id="legalsum-arrow" viewBox="0 0 10 10" refX="8" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path class="lt-s-head" d="M0 0 L10 5 L0 10 z"/></marker></defs> <title id="legalsum-title">The two directions of the proposal. 1, generation: a long judicial decision goes to a language model, which writes a summary. 2, evaluation: the summary is assessed for factual consistency and for argumentative and legal correctness, which standard lexical metrics miss. In both directions, expert knowledge, namely guidelines, annotations, and official summaries, enters the prompt or a retrieval-augmented generation system.</title> <text class="lt-s-sub" x="40" y="22">EXPERT KNOWLEDGE, IN THE PROMPT OR THROUGH RAG</text> <rect class="lt-s-box lt-s-pred" x="40" y="34" width="720" height="48" rx="12"/> <text class="lt-s-txt" x="400" y="63">guidelines, annotations, and official summaries</text> <path class="lt-s-arrow" marker-end="url(#legalsum-arrow)" d="M330 86 V136"/> <path class="lt-s-arrow" marker-end="url(#legalsum-arrow)" d="M665 86 V136"/> <rect class="lt-s-tok" x="40" y="140" width="160" height="100" rx="10"/> <text class="lt-s-txt lt-s-strong" x="120" y="182">Judicial decision</text> <text class="lt-s-txt lt-s-small" x="120" y="204">long, technical</text> <path class="lt-s-arrow" marker-end="url(#legalsum-arrow)" d="M204 190 H246"/> <rect class="lt-s-box lt-s-sel" x="250" y="140" width="160" height="100" rx="12"/> <text class="lt-s-txt lt-s-strong" x="330" y="182">1. Generation</text> <text class="lt-s-txt lt-s-small" x="330" y="204">a language model</text> <text class="lt-s-txt lt-s-small" x="330" y="222">writes a summary</text> <path class="lt-s-arrow" marker-end="url(#legalsum-arrow)" d="M414 190 H446"/> <rect class="lt-s-tok" x="450" y="162" width="100" height="56" rx="10"/> <text class="lt-s-txt lt-s-strong" x="500" y="195">Summary</text> <path class="lt-s-arrow" marker-end="url(#legalsum-arrow)" d="M554 190 H566"/> <rect class="lt-s-box lt-s-sel" x="570" y="140" width="190" height="100" rx="12"/> <text class="lt-s-txt lt-s-strong" x="665" y="170">2. Evaluation</text> <text class="lt-s-txt lt-s-small" x="665" y="192">factual consistency,</text> <text class="lt-s-txt lt-s-small" x="665" y="210">argumentative and</text> <text class="lt-s-txt lt-s-small" x="665" y="228">legal correctness</text> <text class="lt-s-txt lt-s-small" x="400" y="268">A thesis may work on generation, on evaluation, or on both.</text> </svg> <figcaption>The two directions of the proposal, built from its description: generating legal summaries and evaluating them, both with expert knowledge.</figcaption> </figure>