Semantic Role Labeling for Automatic Software Test Cases Generation

Abstract

Decision table testing is a technique to develop test cases from descriptions of conditions and actions in software specification documents. We propose, experiment and evaluate a semantic role labeling technique of conditions and actions for automatic software test cases generation. Our approach uses natural language processing to select sentences from the specification based on syntactic similarity, and then to determine conditions and actions through dependency and case analysis. We got experiment results that precision reached from 0.901 to 0.988, recall reached from 0.946 to 0.974 for different style of descriptions, and the workload was reduced to one-sixth of manual work. Our results on case studies show the effectiveness of our technique.