On the Ethical Limits of Natural Language Processing on Legal Text

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3590–3599 August 1–6, 2021. ©2021 Association for Computational Linguistics 3590 On the Ethical Limits of Natural Language Processing on Legal Text Dimitrios Tsarapatsanis Law School University of York dimitrios.tsarapatsanis@york.ac.uk Nikolaos Aletras Computer Science Department University of Sheffield n.aletras@sheffield.ac.uk Abstract Natural language processing (NLP) methods for analyzing legal text offer legal scholars and practitioners a range of tools allowing to empirically analyze law on a large scale. However, researchers seem to struggle when it comes to identifying ethical limits to using NLP systems for acquiring genuine insights both about the law and the systems’ predictive capacity. In this paper we set out a number of ways in which to think systematically about such issues. We place emphasis on three cru- cial normative parameters which have, to the best of our knowledge, been underestimated by current debates: (a) the importance of aca- demic freedom, (b) the existence of a wide di- versity of legal and ethical norms domestically but even more so internationally and (c) the threat of moralism in research related to com- putational law. For each of these three parame- ters we provide specific recommendations for the legal NLP community. Our discussion is structured around the study of a real-life sce- nario that has prompted recent debate in the legal NLP research community. 1 Introduction Developing computational methods for analyzing legal text is an emerging area in natural language processing (NLP) with various applications such as legal topic classification (Nallapati and Manning, 2008), court opinion generation (Ye et al., 2018) and legal judgment prediction (Aletras et al., 2016; Luo et al., 2017; Zhong et al., 2018; Chalkidis et al., 2019). Legal NLP holds the promise of improving access to justice and offers to legal scholars the tools that allow for an empirical analysis of law on a large scale (Katz, 2012; Zhong et al., 2020). The development and use of legal text processing technologies also raise a series of ethical questions, on which we focus in this paper. For example, following the publication at EMNLP 2019 of a paper on automatic prison term prediction (Chen et al., 2019) using a dataset constructed from pub- lished and publicly available records of past cases of the Supreme People’s Court of China, a debate ensued about the ethical limits of legal NLP. More specifically, Leins et al. (2020) queried in a system- atic way whether papers such as that of Chen et al. (2019) should be published. Leins et al. (2020) invoked a number of arguments including consid- erations to do with the construction of the dataset (Bender and Friedman, 2018), and so-called ‘dual use’ arguments (Radford et al., 2019), i.e. the possi- bility of using a system developed for some purpose for another, potentially harmful, purpose. Follow- ing a rich discussion, Leins et al. (2020) asked whether it is ethically permissible that legal NLP should be used at all to predict items such as prison terms. We believe that the kind of …

Content truncated to 500 words