With the development of language corpora, linguists have been able to identify how often specific words, phrases, and expressions are used, and in which contexts. However, applications of corpora in the wider domain of language teaching have remained limited. This article presents an approach to utilizing corpora, combining principles from communicative language teaching and data-driven learning, and based on a two-year action-research-based trial of the approach with a monolingual class at a Japanese university. Beginning with learner training in their L1, students learned to analyse concordance lines and were given the opportunity to recontextualize and conceptualize any identified language patterns and features through realistic communicative activities. While challenges remain in relation to the effectiveness of learner training and the preparation of suitable concordance lines, students were found to have increased their fluency and accuracy, and to have made improvements to their speaking confidence and complexity.