In an attempt to address the gap in machine learning methods for legal document analysis within humanities and social sciences communications, the interest theory of rights and Hohfeldian taxonomy are presented as a means of capturing the essential features of such texts. Using this legal theory, a stratified representation of knowledge is enabled, making it ideal for machine learning abstraction. In order to test how such legal dimensions could be identified in this domain, a novel heuristic was implemented based on philosophy and language models. Sentence Bidirectional Encoder Representations from Transformers were used to classify Hohfeldian relations to an accuracy of 92.5% when tested on religious discrimination policy texts in the United Kingdom.
This paper contributes significantly to the advancement of legal reasoning tools and the use of machine learning to analyse human and social sciences texts. It demonstrates that despite the subtleties of differences between Hohfeldian legal relations, language models can be trained to accurately identify them. Furthermore, the paper proposes a new approach to measure human interactions via a universal ethical heuristic.
Artificial intelligence (AI) has been increasingly used in public interests such as government departments, courts, and NGOs offering legal services to improve efficiency. Big data natural language processing (NLP) enables a systematic analysis of documents for new findings. Legal documents are characterised as texts that describe power interactions with consideration to outcomes of such interactions.
Rights and duties form the basis of these interactions, essentiating the importance of capturing these principal legal dimensions for machine learning. As such, the interest theory of rights and Hohfeldian taxonomy were proposed. This paper details how these dimensions may be identified using a philosophical heuristic and language models.
In addition, this article raises several ethical considerations to take into account when making use of machine learning in legal document analysis. Indeed, a systematic understanding of the principal dimensions of legal documents is critical to ensure that ML performs according to legal expectations. Ultimately, the implications of such technology and how it may be used ethically are to be further considered by experts in this field.
