Using the interest theory of rights and Hohfeldian taxonomy to address a gap in machine learning methods for legal document analysis

Abstract Rights and duties are essential features of legal documents. Machine learning algorithms have been increasingly applied to extract information from such texts. Currently, their main focus is on named entity recognition, sentiment analysis, the classification court cases predict outcome. In this paper it argued that until texts captured, analysis can remain bottle-necked by very technology being used assess them. As such, use theory identify most pertinent dimensions proposed. Specifically, interest rights, first-order Hohfeldian taxonomy relations. These principal allow for a stratified representation knowledge, making them ideal abstractions needed machine learning. This study considers how may be identified. To do so implements novel heuristic based in philosophy coupled with language models. relations ‘rights-duties’ vs. ‘privileges-no-rights’ determined identifiable. Classification each type relation accuracies 92.5% found using Sentence Bidirectional Encoder Representations Transformers. Testing carried out religious discrimination policy United Kingdom.