Abstract
The present study was a nonequivalent pretest-posttest control group design investigating the effects of corpus-based instruction on grammatical accuracy and written application of past simple forms of habitual expressions. An experimental group was exposed to two weeks of corpus-based instruction as compared to control group that had been exposed to a conventional grammar instruction. Quantitative and qualitative analysis of written tests, along with comparisons of Ethiopian Students Corpus to the British National Corpus was carried out. Common learner mistakes, including subject omission, misuse of used to, and overextension of would to single past occurrences were identified. The two groups showed improvement on the post-test, but the group taught through corpus-based instruction demonstrated significantly stronger improvement and a smaller number of errors: - overgeneralisation and subject-verb agreement issues were observed. This showed that the corpus-based instruction can be used to make accurate and contextually appropriate command of past simple habitual constructions in writing despite the small technological drawbacks.