Extracting Normative Relationships from Business Contracts

The normative concepts offer a principled basis for engineering flexible multiagent systems for business and other cross-organizational settings. However, producing suitable specifications is nontrivial: the difficulty is an obstacle to the adoption of multiagent systems in industry. This paper considers normative relationships of six main types, namely, commitments (both practical and dialectical), authorizations, powers, prohibitions, and sanctions. It applies natural language processing and machine learning to extract these relationships from business contracts, establishing that they are realistic and their encoding can assist modelers, thereby lowering a barrier to adoption. A ten-fold cross-validation over more than 800 sentences randomly drawn from a corpus of real-life contracts (and manually labeled) yields promising results for the viability of this approach.