COMPUTATIONAL STATUTORY REASONING

Statutory reasoning is the task of determining how laws apply to a legal case. This is a basic skill for lawyers, and in its computational form, a fundamental task for legal Artificial Intelligence (AI) systems. In this thesis, I take a first step towards solving computational statutory reasoning. First, I define computational statutory reasoning, in the context of legal practice, and AI more broadly (Chapter 1). I detail why statutory reasoning is important for legal AI, and how solving it will require addressing multiple problems in general AI research. In Chapter 2, I review research in reasoning, knowledge representation and legal AI that is relevant to the work in this thesis. My first contribution is the StAtutory Reasoning Assessment dataset (SARA), a benchmark dataset for computational statutory reasoning (Chapter 3). With the ability to measure performance on statutory reasoning, I show how a symbolic system can solve the SARA dataset, while state-of-the-art Machine Reading (MR) struggles. In Chapter 4, I connect statutory reasoning to established natural language processing tasks, in an attempt to diagnose MR errors. This yields more annotations on SARA, and a performance boost compared to my initial MR baselines. In Chapter 5, I return to the symbolic approach of Chapter 3. Revising the ontology used in Chapter 3, I introduce models for Information Extraction (IE) from SARA cases. The attained performance opens up new perspectives on how to solve statutory reasoning. I close by summarizing the contributions of this thesis, and expanding on possible future research (Chapter 6)