Automatically Extracting Meaning from Legal Texts : Opportunities and Challenges

This paper examines impressive new applications of legal text analytics in automated contract review, litigation support, conceptual legal information retrieval, and legal question answering against the backdrop of some pressing technological constraints. First, artificial intelligence (Al) programs cannot read legal texts like lawyers can. Using statistical methods, Al can only extract some semantic information from legal texts. For example, it can use the extracted meanings to improve retrieval and ranking, but it cannot yet extract legal rules in logical form from statutory texts. Second, machine learning (ML) may yield answers, but it cannot explain its answers to legal questions or reason robustly about how different circumstances would affect its answers. Third, extending the capabilities of legal text analytics requires manual annotation to create more training sets of legal documents for purposes of supervised ML. To some extent, the limitations are temporary. The questions they raise are the subjects of current research concerning the feasibility of drawing inferences from information that: (1) is implicit or distributed across documents such as contracts; (2) captures substantive strengths or weaknesses of a legal scenario; (3) requires manual annotation to teach a computer to identify; or (4) should play a role in explaining the inferences. The paper closes with some practical strategies for dealing with these limitations. It addresses the kinds of legal-process engineering and research the legal community should undertake and underwrite to address these issues and to increase the ability of text-analytic techniques to extract semantic information, draw legal inferences, and explain them