Robust Requirements Analysis in Complex Systems through Machine Learning - Trustworthy Eternal Systems via Evolving, Software Data and Knowledge - page 46

has been defined to search existing software functionalities in a specific domain
through user requirements expressed in natural language.
In the rest of the paper, Section 2 discusses the application of Human Lan-
guage Technologies in RE. Section 3 proposes the architecture of an automatic
system for Requirement Analysis. Section 4 presents the evaluation of the adopted
techniques for the Naval Combat Systems requirement analysis.
2 Language Technologies for Requirement Analysis
The robustness recently achieved by NLP technologies makes their applicability
in the support the analysis and design of system development very promising.
As an example, the reuse of existing technological components during the design
stages of new complex systems can be drastically increased whenever a seman-
tic search system from the targeted component repository is available. Such an
engine would be able to rely on conceptual notions in the user queries (e.g. func-
tions and norms), as they are originally extracted from technical specification
documents, and retrieve components suitable for the design needs and validate
them according to their compliancy or composability . The role of Human Lan-
guage Technologies (HLT) in this proactive support to the analyst is clear as
it favors the incremental design through reuse. HLT are crucial to support ro-
bust and accurate analysis of unstructured texts, and enrich them by semantic
meta-data or other kinds of information implicit in the texts. HLT allows ex-
tracting the interesting semantic phenomena and mapping them into structured
representation of a target domain. When a semantic meta-model is available,
for example in form of an existing ontology, HLT allows to locate concepts in
the text (irrespectively from the variable forms in which they appear in the free
text), mark them according to Knowledge Representation Languages (such as
RDF or OWL) thus unifying different shallow representations of the same con-
cepts. In this way semantic annotations of concepts in the text (i.e. automatic
semantic indexes) are obtained for the original document, making it more suit-
able for clustering, retrieval and browsing activities. In synthesis, HLT enables
to perform and simplify several advanced functionalities (e.g. semantic and not
keyword based search) that are possible over the text. The semantic annota-
tion task, just outlined above, has been largely studied by the NLP community
and it is known as Information Extraction (IE), i.e. “ The identification and ex-
traction of instances of a particular class of events or relationships in a natural
language text and their transformation into a structured representation (e.g. a
database). ” [8]. IE requires typically three stages. In the first, the target infor-
mation is abstracted and structured set of inter-related categories are designed.
These structures are called templates and the categories (roles) that need to
be filled with information are called slots . For example, if we want to extract
conceptual information about vessels from specifications, we may be interested
in the name but also in the type of ship or its maximum speed ,aswellasits
combat system equipment . Therefore, a Ship template can be defined as a con-
junctive combination of slots such as name , ship type , maximum speed or combat