Automated Identification of Deontic Modalities in Software Engineering Contracts: A Domain Adaptation-Based Generative Approach

Contracts are legally binding agreements between parties that establish rights, obligations, and terms for their business relationship. They articulate the deontic modalities (Obligations, Permissions, Prohibitions, and Exclusions) that apply to those involved in the contractual agreement. Deontic modalities can be leveraged to effectively elicit Software Engineering (SE) requirements. In this paper, we propose a novel approach for identifying deontic modalities from contracts, employing a combination of text generation and domain adaptation techniques. Among the SOTA approaches we experimented with, the T5-large model yielded the best results, with an average precision and recall of 0.96 and 0.95, respectively. Comparing our approach with previous methods, the results show that our approach handles significant class imbalances in the training data and demonstrates good generalization ability on new datasets.