From Leibniz Characteristica Universalis to Google Translate: Machine Translation as Challenge and Opportunity for International Law

The recent renewed interest in the application of Artificial Intelligence to the legal domain has centered around two possible applications: a) Automated decision making, reviving in varying degrees to old dream of the “robojudge” and can either be used as a tool to assist traditional human adjudication, or embedded in electronic devices ensure “legal compliance by design", and b) Prediction of the likely outcome of cases before they are litigated – with at least some of the examples coming from the field of international law, such as the attempt to predict decisions of the European Court of Human Rights. What has received much less attention, even though it is arguably one of the success stories in machine learning, has been automated translation. For international law in particular, this is regrettable given the plethora of questions that the multilingual nature of many international treaties and international courts raises. Indeed, in the example of the prediction of ECHR decisions above, the multilingual nature of the decisions of the court was actively negated in the training of the model, distorting as a result the training set and potentially introducing biases into the model. By the same token, many of the problems that have been identified for machine learning for automated decision making also arise in the field of machine translation, including the opacity (“black boxing”) of the algorithm and the danger of biased translation (Google Translate is still systematically changing the gender of translations when they do not fit with stereotypes it picked up from data generated by its users). A with other fields of AI in law, automated legal translation promises benefits for the administration of justice, and in particular greater access to justice. In theory, especially more technical examples of international norm setting, such as ISO standards, can be translated with minimal costs into a much wider range of languages than hitherto possible. International collaboration between institutions such as Interpol can be facilitated as requests and queries can be translated at speed and without complex procedures to set up multiple simultaneous translators. Access to machine translation also changes the landscape of the legal protection of minority languages under international law, or recent immigrants accessing the legal services of their new country. But with these opportunities also come risks. In this presentation, we explore these risks from a legal perspective.