Producing NLP-based On-line Contentware

For its internal needs as well as for commercial purposes, CDC Group has produced several NLP-based on-line contentware applications for years. The development process of such applications is subject to numerous constraints such as quality of service, integration of new advances in NLP, direct reactions from users, continuous versioning, short delivery deadlines and cost control. Following this industrial and commercial experience, malleability of the applications, their openness towards foreign components, efficiency of applications and their ease of exploitation have appeared to be key points. In this paper, we describe TalLab, a powerful architecture for on-line contentware which fulfils these requirements.