Robert Pool in his recent blog touched upon an interesting question: bottom-up versus top-down approach in Academic Writing. The traditional bottom-up method looks at the narrow micro-textual features failing to provide an information for broader details, such as relation between lexical choices and rhetorical moves of the authors. The guest of the blog, James Garner, illustrates the combination of both methods by means of corpus data, such as COCA (Corpus of Contemporary American English) and MICUSP. For example, students are able to investigate not only the frequency information, but also the phraseology of the items.
As a corpus linguist, myself, I think that on-line corpora can be beneficial supplement not only for writing, but also in other fields, such as history, journalism, literary and gender studies.