This paper describes a five-year research project aimed at developing a corpus-based language support tool able to respond to the English writing needs of Japanese software engineers who do not speak English natively. Our research was based on recent developments in corpus and text linguistics. Since foreign readers often complain that English text produced by Japanese authors is difficult to understand because it is poorly organized and incoherent, we focused on the possibility of designing a writing tool that would provide discourse-level as well as sentence-level assistance. We collected a total of 539 sample English abstracts from four well-known technical journals and tagged them with linguistic and rhetorical information. Using this tagged corpus, an initial prototype was developed on a Unix-based workstation and a second one on the Web. The Web-based prototype was then evaluated in terms of its usability by engineers in Ricoh's Software Research and Development Group. They evaluated the final product positively. However, they expressed uncertainty about its ability to address their weaknesses in using transition words effectively as cohesive devices. In spite of unexpected difficulties, product improvement continues.