An Obligations Extraction System for Heterogeneous Legal Documents: Building and Evaluating Data and Model
- Resource Type
- Authors
- Iacono, Maria; Rossi, Laura; Dangelo, Paolo; Tesei, Andrea; De Mattei, Lorenzo
- Source
- Subject
- Computational Linguistics
linguistica
linguistique computationelle
Linguistics
linguistica computazionale
linguistique
Language
- Language
- English
A system that extracts obligations automatically from heterogeneous regulations could be of great help for a variety of stakeholders including financial institutions. In order to reach this goal, we propose a methodology to build a training set of regulations written in Italian coming from a set of different legal sources and a system based on a Transformer language model to solve this task. More importantly, we deep dive into the process of human and machine-learned annotations by carrying out both quantitative and manual evaluations of both of them.