Corpus linguistics has contributed considerably to the arsenal of tools available for language description, with methodologies applicable to virtually any area of language study. Yet like most academic disciplines, corpus linguistics is generally limited to researchers and other professionals; in the fields of foreign/second language teaching and learning, this involves considerable mediation upstream before the results filter down to the end users - teachers and learners. An alternative would be for corpus data to feature directly in the learning process, with learners acting as "researchers" to explore the language themselves. This is commonly referred to as "data-driven learning" (DDL), an idea particularly associated with the seminal collection of papers edited by Johns and King (1991). The last 20 years has seen substantial research interest in DDL, with countless scholarly articles arguing its various merits, though it is frequently remarked that empirical research is disturbingly lacking. However, Boulton (forthcoming) lists over 70 studies which seek to evaluate some aspect of L2 corpus consultation. This suggests that the research exists, but is fragmented and lacking in visibility and accessibility to a wider community. The aim of the present paper is to bring together the DDL research in ESP, where Gavioli (2005) argues convincingly that the learners represent a major population who stand to gain from DDL. An overview is presented of the tools, corpora, methodologies and research questions covered in 20 empirical DDL studies for ESP. The overwhelming majority show that learners are able to work with corpora in many contexts and for many purposes, and respond positively to the approach. In terms of learning outcomes, the results are generally promising, though not perhaps as conclusive as one might hope. Reasons for this are discussed, but it is argued that there are grounds for cautious optimism regarding corpus use in ESP in a wide variety of contexts.
