ExANT - Extraktion von Anforderungen aus natürlichsprachlichem Text; Teilvorhaben Automatischer Test (AT)

Abstract

The approach developed during the ExANT study uses AI (Artificial Intelligence) to formalize requirements formulated in natural language text. It extracts information to assess the quality and complexity of requirements, to derive inputs for software verification and implementation, and to compare requirements automatically by means of metrics. The intended application fields are (critical) embedded systems. The open source NLP (Natural Language Processing) framework spaCy is used for tokenization, Part-of-Speech (PoS) tagging and dependency parsing based on a pre-trained English language model. For the following identification of text chunks a rule-based exploration approach was chosen in contrast to domain-specific training to ensure independence of a specific engineering domain and language - as far as possible. The text is broken down into chunks which are put into a normalised order. If this is not possible, a quality issue may be detected. The normalised sequence of chunks can be used for building test oracles, for the implementation of software tests, and for applying metrics so that requirements may be compared for similarity or be evaluated on quality.