This paper assumes that text linguistics is essentially probabilistic in accounting for language use or performance. It proposes as a basic form of test the computational variant of the Shannon Game. The research paradigm it wishes to pursue is that of building always better Shannon-Game-playing systems.