The aim of this thesis, part of a two-phase project, is to build a software retrieval
system where software components are represented by their natural language
descriptions. The focus is to improve recall, while precision will be
taken care of in the other part of the project. In contrast to the widely used
techniques, the thesis provides a new approach based on the systemic
functional theory to represent the software components. Basically, the notion
of the Theme-Rheme which is a subsystem of the language system, is
applied. It deals with the textual meaning of texts. Taking into account the
uncertainty related to natural language, incompleteness of users' queries and
the multiple-view classification of software components, a fuzzy
representation with respect to some measures is derived by the thematic
analysis of texts. The fuzzy representation reflects the hierarchical nature
of the documents and suggests the use of type-2 fuzzy sets. The matching 
between the user's query and the content of the software repository is
performed by means of an aggregated similarity measure (combination of themes
and rhemes). As a second solution, the fuzzy representation is translated into
a cascade of two neural networks. The first level in this cascade is a fuzzy
associative memory network (FAM) which maps rhemes to themes and the second
level consists of a fuzzy adaptive resonance theory network (Fuzzy ART) which
relates themes to classes. The approach was experimentally evaluated using
different versions of the prototype SyRS built in the framework of this
research. The results show that SyRS achieves the goal of this dissertation.