The paper examines two basic modes of text building, first proposed by Sinclair (1994). It argues that these text-building means are not just different, but are, rather, operational opposites. It suggests that this difference means that ESL learners who first acquire one mode of text building are, by this means, also equipped with the wherewithal to prevent implementation of the other. The result is an emergent systemic discourse-level difference from target norms in the learners' use of the later-acquired means of text building. This is evidenced in an assessment of a sample of expository essays by fairly advanced learners of English, whose first language is Cantonese. The data assessment also shows that the nontarget-like system is not replaced by a more target-like substitute, in spite of sustained opportunity for continued learning of the target norms. Rather, the resulting nontarget-like structure is retained as a permanent feature, contributing to the distinctness of the local variety, because it is well accommodated in natural communication (even in cross-cultural intercourse), for the simple reason that it does not cause miscommunication.