New article out in Law and Artificial Intelligence

Identifying case law and other legal resources that substantiate legal propositions is a fundamental aspect of legal research and decision-making. Existing legal information retrieval systems assist this task by recommending relevant legal documents. However, these documents are often lengthy, and users are primarily interested in accessing and referencing specific, directly relevant sections. Augmenting recommendation at the document level with suggestions at the paragraph level, here referred to as ‘pincite recommendations’, could significantly enhance efficiency. This paper presents, tests, and proposes an approach for such a pincite recommendation model. Using the case law of the Court of Justice of the European Union as a test case, we demonstrate that a language embeddings model can predict citations with a high degree of accuracy, providing users precise and pertinent pincites for legal propositions.   
Article available here: