Anonymisation of judicial decisions with machine learning

Slovenian Constitution determines that court proceedings are public. This means that court hearings (except when there are some special reasons, for instance involved minors, government secrets, etc.) shall be public, and judgments shall be pronounced publicly. Therefore public has the right to know the decisions of judiciary branch of power. However, publishing court decisions is also believed to create a push towards unification of jurisprudence. These are main reasons why Slovenian ministry of justice wants to publish all court decisions on the Internet. However, Slovenian Constitution also protects personal data, so court decisions should be published on the Internet in anonymous form. Before public release, all personal data or other data from which persons involved in trial could be identified, should be removed from court decision In a presentation we will present a tool Tacita, which helps in this anonymisation process. Tool was developed at Jožef Stefan Institute and uses machine learning to predict which part of a court decision should be removed (anonymised) with a high probability. Tacita is not working completely automatic, but helps in otherwise time-consuming manual process of anonymisation. We will also show how the tool has been developed and which tools for analyzing natural language has been used.