Automated Classification of Legal Cross References Based on Semantic Intent

Abstract

[Context and motivation] To elaborate legal compliance requirements, analysts need to read and interpret the relevant legal provisions. An important complexity while performing this task is that the information pertaining to a compliance requirement may be scattered across several provisions that are related via cross references. [Question/Problem] Prior research highlights the importance of determining and accounting for the semantics of cross references in legal texts during requirements elaboration, with taxonomies having been already proposed for this purpose. Little work nevertheless exists on automating the classification of cross references based on their semantic intent. Such automation is beneficial both for handling large and complex legal texts, and also for providing guidance to analysts. [Principal ideas/results] We develop an approach for automated classification of legal cross references based on their semantic intent. Our approach draws on a qualitative study indicating that, in most cases, the text segments appearing before and after a cross reference contain cues about the cross reference's intent. [Contributions] We report on the results of our qualitative study, which include an enhanced semantic taxonomy for cross references and a set of natural language patterns associated with the intent types in this taxonomy. Using the patterns, we build an automated classifier for cross references. We evaluate the accuracy of this classifier through case studies. Our results indicate that our classifier yields an average accuracy F-measure of $$\approx 84\,\%$$i¾?84%.