Centre for Information and Innovation Law (CIIR)

LEGALESE – Danish Language Processing (NLP) for Legal Texts

LEGALESE is a joint venture by the University of Copenhagen’s Faculty of Law and Department of Computer Science, KMD, and Ankestyrelsen. The aim of the project is to is to research and develop a product that facilitates efficient legal information retrieval and automated recommendation using cutting-edge natural language processing (NLP).

THE PROJECT IS OPEN
Project period: 1 September 2020 – 1 September 2024

Central research questions and scope

The aim of legalese is to develop nlp solutions for legal information retrieval. Existing state-of-the art in NLP is based on self-attentive transformer models, but these only work optimally on shorter texts (up to 4096 tokens). Real world legal texts however are much longer. The project will research how this challenge can best be overcome. It does so by testing several NLP models in an empirical setting.

Another aim is to establish to which extent bias detection and removal (debiasing) are needed to ensure legal compliance and legitimacy in the legal information retrieval process in the setting of Danish public administration. The projectanalyses the implications of the currently existing legal framework, namely the GDPR, the EU Non-Discrimination Law Directives and the upcoming AI Act.. The research will be conducted in ongoing collaboration with the Department of Computer Science.

For more information: Sebastian Felix Schwemer, e-mail: sebastian.felix.schwemer@jur.ku.dk

Goals and future research

The aim of legalese is to develop a NLP model for danish language that can be used to drive an automated legal information retrieval system at Ankestyrelsen.

The work carried out in LEGALESE, is likely to spur new research on how NLP can be used in a responsible way to operate information retrieval systems in other parts of the public and private sector in Denmark and abroad.

Debiasing in natural language processing (NLP) – WP2

Bias detection and removal (debiasing) are essential for the success of such NLP system. WP2 will work on the related cutting-edge legal issues with a view to establish the standard for NLP implementations in terms of both privacy and bias as well as research on how debiasing may be implemented. In particular, WP2 analyses the implications of the application of the currently existing legal framework, namely the GDPR and the EU Non-Discrimination Law Directives, to legal information retrieval and explores the positive and negative consequences that newly proposed regulatory solutions, such as the EU Artificial Intelligence Act, may have on prevention and mitigation of bias in this context. The research will be conducted in ongoing collaboration with the Department of Computer Science.

For more information: Sebastian Felix Schwemer, e-mail: sebastian.felix.schwemer@jur.ku.dk