Unsupervised and supervised text similarity systems for automated identification of national implementing measures of European directives

Abstract

The automated identification of national implementations (NIMs) of European directives by text similarity techniques has shown promising preliminary results. Previous works have proposed and utilized unsupervised lexical and semantic similarity techniques based on vector space models, latent semantic analysis and topic models. However, these techniques were evaluated on a small multilingual corpus of directives and NIMs. In this paper, we utilize word and paragraph embedding models learned by shallow neural networks from a multilingual legal corpus of European directives and national legislation (from Ireland, Luxembourg and Italy) to develop unsupervised semantic similarity systems to identify transpositions. We evaluate these models and compare their results with the previous unsupervised methods on a multilingual test corpus of 43 Directives and their corresponding NIMs. We also develop supervised machine learning models to identify transpositions and compare their performance with different feature sets.

Multilingual Legal Information Retrieval System for Mapping Recitals and Normative Provisions

“…Figure 3 presents the macro-average precision, recall and F-Score for the mappings between directive (sub-)articles and NIM provisions.These results indicate that the Luxembourg directive mappings consistently achieved a higher recall, precision and F-Score than Italian and English ones for all the similarity measures. This is consistent with the research presented in [12]. The best overall macro-average F-Score values are 0.8243, 0.7276 and 0.6712 for mappings between directive (sub-)articles and the NIM provisions of Luxembourg, Italy and Ireland respectively.…”

“…A multilingual parallel corpus of 43 directives and their corresponding NIMs from Luxembourg, Ireland and Italy (15,400 norms) was presented in [12]. This corpus only contains mappings between directive (sub-)articles and NIM provisions.…”

“…We removed common noisy words (using spaCy's default list of stopwords) as well as punctuation. We did not select words to retain based on their part-of-speech tag because in the case of short text similarity models (as we are comparing legal norms instead of documents), we need to utilize all the available linguistic features to achieve an acceptable magnitude of text similarity [12]. The text representation phase then encodes the linguistic (syntactic and semantic) features.…”

Sense and similarity: automating legal text comparison

“…Such national implementation of international law is particularly common in Europe where directives of the supranational European Union (EU) are being implemented through national measures within the organization's member states. While the EU currently tasks specialized consulting companies to manually monitor the faithful implementation of such directives, researchers have shown that text-based similarity measures can identify national implementing statutes with reasonable accuracy across different EU member states (Nanda et al, 2018).…”

Application of Computer Techniques and Systems in the Study of Law, Intellectual Analysis and Modeling of Legal Activity: A Systematic Review

“…Так, в 2007 г. в качестве онтолингвистического ресурса была обоснована характеристика знаний DALOS и система организации знаний (KOS) для обеспечения контроля и поддержки законодательного процесса в многоязычной среде европейских стран, обязанных имплементировать евродирективы в национальное право [1]. В 2019 г. были представлены результаты разработки и апробации моделей на данных, обеспечивающих контроль над имплементацией европейского права в национальные правовые системы; эти модели были реализованы с использованием неконтролируемых методов лексического и семантического сходства, основанных на моделях векторного пространства, скрытом семантическом анализе и тематических моделях [88].…”