Behavioral Programming With a Subset of Natural Language: An Evaluation Study

Abstract

We consider the idea of behavioral programming by writing controlled natural language requirements, which are then compiled directly into executable code. Our motivation is to bridge the gap between system requirements and a final system. We claim that formal structured natural language requirements can serve as the means and the end to programming the behavior of reactive systems, if one utilizes a fully executable language that supports story-like sentences, such as live sequence charts (LSC). We use natural language processing methods combined with user interaction in order to understand English requirements, and translate them into LSCs, disambiguating and clarifying intentions with the user when necessary. We test a multi-modal user interface we term show & tell, where natural language is interspersed with play-in, and model-based co-reference resolution. Our approach is domain general and builds the underlying model of the system leading to direct execution. The paper presents novel evaluation studies that show its potential as an effective tool for natural language programming of reactive system behavior.