Enterprise terminologies are now getting close to a panel of semantic representations coming from various disciplines and practices, and that share the relationship between lexical and conceptual levels within data structures. As long as these representations are built up from texts that reflect usage, their role goes far beyond drawing a simple inventory of terms and definitions. From now on, these structures are indicators of either convergences or misfunctions (where), of vocabularies as well as professional competences (who) or even temporal evolutions (when). By doing so, document based semantic representations are detectors useful to improve the company communication and knowledge management. To better measure these challenges, it is worth underlying the impact of corpora selection the exact nature of data structures and the kind of available techniques and software for their design from texts.