Aumentare la trasparenza normativa con similarità semantica e Akoma Ntoso

Abstract

Enhancing regulatory transparency through semantic similarity and Akoma Ntoso
Regulatory transparency depends on the possibility of reconstructing and verifying the links between sources, rather than on the mere availability of texts. The transposition of EU directives is an emblematic case, because European obligations are often reformulated and distributed across multiple national provisions, making the manual reconstruction of correspondences costly and difficult to replicate. This work proposes a pipeline to support comparative analysis between EU directives and the Italian legislative decrees that implement them, while keeping an explicit connection between computational results and the legal text. The workflow normalizes the sources into a structured representation. Italian decrees are acquired from Normattiva in Akoma Ntoso XML format, while EU directives are downloaded from EUR-Lex in the “OJ” HTML version and converted into a consistent XML structure with coherent metadata and segmentation into articles and paragraphs. Semantic similarity (Sentence-BERT, cosine similarity) and a complementary lexical overlap indicator are calculated on these units; the candidates are then ranked and provided through inspectable outputs and visualizations (heatmaps and graphs) designed for human review.