On the automatic analysis of rules governing online communities

Abstract

The automatic translation of rules or legal text from natural language into formal language has gained interest in the natural language processing domain, especially in the field of law and AI. Our research goal is to be able to automatically extract, from rules in natural language, the necessary elements that define these rules, such as the action in question, its modality (duty, right, privilege,..), the first person the rule addresses, the second person affected by the rule, and the condition (if the rule was a conditional rule). As a first step toward identifying these elements, we start by identifying the semantic subjects, verbs, and objects in sentences of online normative texts. This paper presents the SVO+ model that achieves this, and our evaluation illustrates the model’s high precision when tested with the terms of use from websites like Facebook and Twitter.