Machine Learning vs. Knowlegde Engineering in Classification of Sentences in Dutch Law

The ultimate goal of the application of artificial intelligence in the field of legislation is to support automated modeling of sources of law. These models can help to provide fast, comprehensible and easy access to up-to-date legal knowledge to both civilians and professionals in the legal domain. A first step towards the translation of provisions into formal models is the classification of legal texts into different categories. Research in this field showed that both a machine learning (ML) approach as a knowledge engineering (KE) approach can be successfully applied to this task. This work presents a comparative study between pattern based KE techniques and a ML approach that uses support vector machines to classify sentences from Dutch law. The results suggest that at a sentence level ML techniques can achieve good results (over 90% accuracy), but that they can at best equal the results obtained by the more robust KE approach.