A Deep Learning Approach for Norm Conflict Identification

Regulations are often applied to social members in a society in order to minimize conflicting behaviors [9]. Such regulations, also known as social norms, define expected behaviors for society members [7] and help ensure that individuals act in socially acceptable behavior. Besides regulating entire societies, social norms are also used to regulate interactions in smaller groups, and are often present in social relationships involving agreements over products and services. A common way to formalize sets of norms applied to a certain agreement is through contracts [8]. In human societies, contracts are semi-structured documents written in natural language, which are used in almost every existing formal agreement. Contracts define the parties involved in the agreement, their relations, and the behavior expected of each party within clauses. When written in natural language, contracts may use imprecise and possibly vague language to define parties, obligations and objects of its clauses, leading to inconsistencies. Such inconsistencies may create, in the long run, unforeseen legal problems for one or more of the involved parties. To identify and solve such conflicts and inconsistencies, the contract maker needs to read the entire contract and identify each conflicting pair of norms. As contracts may have a large number norms, the identification of norm conflicts by human beings takes substantial effort and tends to be error-prone. We address the problem of identifying and quantifying potential normative conflicts between natural language contract clauses [1]. Our main contributions consist of an approach based on deep learning to address the problem of identifying potential normative conflicts between natural language contract clauses, as well as the corpus containing normative conflicts [2] using to train the classifiers involved. We process raw text from contracts and identify their norms.