Norm Conflict Identification using Vector Space Offsets

Contracts formally represent agreements between parties and often involve the exchange of goods and services. In contracts, norms define the expected behaviors from parties using deontic statements, such as obligations, permissions, and prohibitions. However, norms may conflict when two or more apply to the same context but have different deontic statements, such as a permission to delay payment being present in the same contract as an obligation to pay in a fixed deadline. Contracts with conflicting norms may be invalidated in whole or in part, making conflict identification a major concern in contract writing. Conflict identification in contracts by humans is a time-consuming and error-prone task that would greatly benefit from automated aid. In order to automate such identification, we introduce an approach to identify potential conflicts between norms in contracts written in natural language that compares a latent (vector) representation of norms using an implicit offset that encodes normative conflicts. Experimental evaluation shows that our approach is substantially more accurate than the existing state of the art in an open dataset.