CATALA translates law into code for more reliable administration

Transforming legislative texts into computer programmes is essential if the law is to be applied on a large scale, for example in calculating taxes or distributing welfare benefits. However, this task is complex due to the often ambiguous and contextual nature of legal texts. CATALA was designed to meet this challenge: a programming language specially developed to transcribe the law faithfully and in a way that is easy to understand, ensuring that legal rules are applied accurately.

Denis Merigoux, project leader for software infrastructures in support of public policies at Inria, came up with the idea for this project while working on a renovation of the M language and its compiler alongside the French Public Finance Department (DGFIP), to make tax calculations even more reliable.

The scientist then decided to collaborate with legal experts and a sociologist to bring his project to fruition: Liane Huttner, lecturer in law at the Université Paris-Saclay, Sarah Lawsky, professor of tax law, doctorate in formal logic and holder of an international chair at Inria in the Prosecco group, which has studied the structuring of legislative texts in formal language, and finally Marie Alauzen, specialist in issues of administrative modernisation of the State. Their aim was to combine their skills to design a programming language that would preserve the legislator's intentions while being able to be integrated into modern software architectures.

"The aim is not to formalise or put into code all the law, because that would make no sense, but we are interested in the law that is already executed automatically, such as the calculation of social benefits, tax or unemployment", explains Denis Merigoux, before adding "We are really getting lawyers and computer scientists to work together, who exchange information live on the code that the computer scientist is writing so that they can make changes to it in real time". Thanks to this approach, it is possible to guarantee that tax and social security rules are applied accurately and faithfully to the law, thereby reducing the risk of error and enhancing the transparency of automated administrative decisions.

CATALA, which involves several Inria engineers and researchers (involved in the AVoCat exploratory action), is particularly relevant for public administrations and bodies responsible for applying tax and social policies. By replacing legacy IT systems for calculating tax and social security benefits, this language provides exhaustive and explainable assurance that the administration's algorithms apply the law as interpreted by the administration's legal departments, without distortion or approximation.

Two proofs of concept for government departments have already been produced: the first for the Caisse Nationale des Allocations Familiales, which is currently carrying out studies on the future of the system for calculating all social benefits in France, and the second for the Direction Générale des Finances Publiques, for calculating income tax.