Every day, the judicial authorities receive countless legal documents that are not assigned to a clear topic or case category, there is a lack of important information about those involved, and content or scans are sometimes summarized in documents randomly. However, judges and clerks need documents in a certain structure and designation so that they can work efficiently. In order to enable processing at all and to enable an exchange between courts, but also to control the work of the courts, files still had to be reworked, i.e. anonymized in order to protect personal rights.
All of this costs time and money and is intensive manual work!
Up to now, models for recognizing categories, entities, parties in lawsuits and patterns in legal documents required a fixed, form-like structure. In fact, those models did not cover the entire range of possible files that occur in legal relations. Data that appears in legal texts not only in tabular form, but also as continuous text or with images included, has not been adequately recorded yet.
For the Austrian Federal Ministry of Justice (BMJ for short), the IT specialist EBCONT has therefore created and implemented an overall concept for the automated processing of legal documents in a number of use cases, which, after several partial implementations, will now be fully activated in autumn.
The groundbreaking automated anonymization of legal documents and especially of judgments developed by EBCONT is particularly worth mentioning, which will replace the current solution for Austrian courts in autumn 2021 in full expansion.
What was previously done manually and required a lot of time and resources, can now be done in a few moments after the successful implementation. Hence, it was also possible to digitize and anonymize over 1 million old or previously unprocessed judgments from all judicial authorities.
What made the project so challenging for EBCONT was that there were neither quality metrics nor 100% usable training data, namely for none of the intermediate information, but only data for “blackened” and “not blackened”, so the team had to create them through complex heuristics.
Both the BMJ and EBCONT expect notable improvements in data processing not only for judges, but for all those involved in legal proceedings, taking data protection standards into account.
Martin Hackl, Chief Digital Officer of the Federal Ministry of Justice: “It is now important that we establish the processes. People and machines make mistakes, so we still have to create appropriate workflows for automated processes to be accepted. “
Frederick Bednar, Data Analytics and Data Science Consultant at EBCONT can look back on many years of experience in the field of Natural Language Processing and semantic use cases: “The BMJ values our competence and has therefore fully relied on our extensive knowledge of both data science and machine learning. In the specific example of anonymization, we have gained valuable experience that a combination of several methods pledges to be the most effective solution, especially for highly complex domains. We are very pleased to have contributed to the full development of the anonymization of the BMJ’s legal documents. “