In order to design complex, effective and adaptable NLP systems a methodolo
gy able to satisfy two traditionally conflicting requirements in software e
ngineering, i.e (linguistic) expressiveness and robustness, is necessary. B
y combining NLP methodologies and Language Engineering (LE) methods with So
ftware Engineering (SE) criteria, we propose a software infrastructure able
to optimize the design and development of complex IE applications. The bas
ic idea is to embed within the software infrastructure itself a suitable li
nguistic description and make available at a computational level relevant p
ortions of the linguistic abstraction required by a variety of applications
.