Information extraction from documents for automating software testing

Abstract

Natural-language-based tools can be valuable for automating software engineering, in particular for automating software testing. However, efforts to automate software engineering rarely include natural-language texts, instead of focusing on source code or on encoding knowledge in specialized formats. The specification information from text (SIFT) tool demonstrates the potential for incorporating existing texts directly into an automated testing system; it generates tests directly from information extracted from specification documents or user manuals. SIFT provides a general framework for using domain-specific parsing techniques, and has shown its utility in constructing tests for the OpenVMS operating system interface routines.