Information Technology Reference
In-Depth Information
having a problem coding a solution - we are having a problem understanding what
solution to code ... If you focus on requirements and verification and validation,
the coding will take care of itself ”. Vagueness and ambiguity are the main phe-
nomena that make the natural language used to describe user requirement a
challenging task. Consider the complexity of a sentence when it contains clauses
and phrases that describe and relate several objects, conditions, events and/or
actions.
Natural Language Processing (NLP) approaches gained much interest in the
community of Software Engineering, as recent works in this direction suggest.
In [2] a similarity measure based on linguistic information is used for clustering
correlated software artifacts. In particular, authors explore the effects of mining
lexical information about different artifact element, such as Function Names, Pa-
rameter Names or Software Comments. In [3], an automatic approach to classify
affordances of web services according to the texts describing them is presented.
Regarding Requirement Analysis, Abbot [4] proposes a technique attempting
to guide the systematic procedure that compiles design models from textual re-
quirements. While it was able to produce static analysis and design modules,
it was nonetheless requiring high levels of user involvement for decision mak-
ing. Saeki et. Al. [5] illustrates a process of incrementally constructing software
modules from object-oriented specifications as obtained by interpreting text re-
quirements. Nouns were considered as classes and their corresponding verbs as
methods. These were automatically extracted from the raw textual descriptions
but lexical ambiguity problems and hand-coding were striking limitations in the
construction of fully reusable formal specifications. In the REVERE [6] system,
a summary of requirements from a natural language text is derived. The system
makes use of a lexicon to recognize suitable word senses in the texts. However, no
attempt to model the system at the functional level is carried out. In [7] natural
language analysis is suggested as a possible approach for automatically compile
formalized control mechanisms in the requirement specifications. An expressive
semantics-based point cuts within a requirement are detected and mapped into
the RDL semi-formal description language. The authors suggest that syntac-
tic and semantic analysis of natural language expressions can be made precise
enough to support the definition of a flexible composition mechanism for re-
quirements analysis. All those systems, while exploring the applicability of NLP,
propose traditional tools for the specific RE context. Most of the traditional
limitations of NLP are thus inherited by the above works, namely costly design
and development processes, complex maintenance of the large Knowledge Bases
necessary for full NL analysis as well as poor portability across domain, systems
and scenarios.
In this work we propose statistical learning methods embedded in a large scale
natural language processing system in support of RE. The adoption of advanced
technique of NLP combined with Machine Learning capabilities, i.e. Statistical
Information Extraction, is a crucial advance to improve applicability of this
technology on a large scale. Moreover, the effectiveness of acquired information
is evaluated in a Information Retrieval scenario, where a robust search engine
Search WWH ::




Custom Search