Information Technology Reference
In-Depth Information
From Requirements to Models: Feedback Generation as
a Result of Formalization
Leonid Kof and Birgit Penzenstadler
Fakultat f ur Informatik, Technische Universitat Munchen,
Boltzmannstr. 3, D-85748, Garching bei Munchen, Germany
{
kof,penzenst
}
@informatik.tu-muenchen.de
Abstract.
Natural language is the main presentation means in industrial require-
ments documents. In addition, communication between the different stakeholders
is often insufficient, therefore requirements documents are frequently incomplete
and inconsistent. This causes problems during modeling or programming.
The aim of the presented paper is to make deficiencies in behavior specifi-
cations apparent in the early project stage. The basic idea is to model the re-
quired system behavior and to generate feedback for human analysts, based on
the deficiencies of the resulting models. The presented feedback generation was
evaluated in an experiment. It was found that it can address genuine problems of
requirements documents.
Keywords:
Requirements Engineering, Model Extraction, Feedback Generation.
1
Requirements Documents Suffer from Missing Information
At the beginning of a software project, the requirements of different stakeholders are
usually gathered in a document. The majority of these documents are written in natural
language, as the survey by Mich et al. shows [1]. Diversity of stakeholders and insuf-
ficient communication results in imprecise, incomplete, and inconsistent requirements
documents, because precision, completeness and consistency are extremely difficult to
achieve using mere natural language as the main presentation means.
In software development, the later an error is found, the more expensive its correc-
tion. Thus, it is one of the goals of requirements analysis to find and to correct the
deficiencies of requirements documents. A practical way to detect errors in require-
ments documents is to convert informal specifications to system models. In this case,
errors in documents would lead to inconsistencies or omissions in models, and, due to
the more formal nature of models, inconsistencies and omissions are easier to detect in
models than in textual documents.
Although there exist a number of automatic approaches that analyze specifications
written in natural language and provide a model, the existing approaches go in one
direction only: they transform a textual specification into a formal model. However,
in the case that the specification exhibits some deficiencies, they either heuristically
compensate these deficiencies or fail silently.
This work was supported by the German Research Council (DFG), Grant BR 887/26-1.
