Recently, Deep Learning has been widely used for tackling challenging NLP tasks such as text classification, question answering, etc. Similarly, the application of Deep Neural Networks in the legal analytics has increased significantly although there is room for more. Current existing methodologies tailored to legal analytics differ on three main perspectives: (i) Text feature representation, (ii) Neural Network structure, and (iii) the performance and outcomes.

At the Luxembourg Institute of Science and Technology, research has been done to develop publicly available word embeddings oriented to legal text trained on large corpora comprised of legislation from UK, EU, Canada, Australia, USA, and Japan among other legal documents. Word embeddings are low-dimensional dense vectors as word feature representations. The legal word embeddings produced in this research were trained using the WORD2VEC model sharing them for public use. The Law2Vec models have been published at this link.


Main contributor(s): Dimitrios Kampas