Data-Driven Learning
What if learners found the rule themselves?
Data-driven learning is an approach, named by Tim Johns, in which learners examine real language data from a corpus, often concordance lines, to work out for themselves how words and patterns are used.
A learner asks whether it is depend of or depend on. Instead of answering, the teacher hands out twelve lines from a corpus, each with depend in the centre, and asks: what comes next? This is data-driven learning (DDL), a term Tim Johns began using around 1990 for corpus work with international students at the University of Birmingham.
Johns saw the computer not as a stand-in teacher but as a source of evidence that learners question for themselves. Learners read concordance lines, spot a pattern, form a hypothesis and test it on more data. In a 1997 chapter he summed up the goal in a much-quoted slogan: “every student a Sherlock Holmes”. In SLA terms, DDL turns noticing into a task rather than a hope. It does not even need computers: Boulton (2010) found that lower-level learners did well with printed concordance materials.
The evidence has grown steadily. Boulton and Cobb’s (2017) meta-analysis of 64 studies found large average effects, for example d = 0.95 when DDL groups were compared with control groups. A later meta-analysis of vocabulary studies by Lee, Warschauer and Lee (2019) found medium-sized gains, both straight after teaching and at delayed post-tests. In their data, effects varied with learner proficiency, training and the type of corpus used, while Boulton and Cobb flagged long-term retention as under-researched. So DDL is no longer a fringe idea, but how it is set up matters: the data, the task and the training learners receive.
In actionTwelve lines, cut to five
… whether we go | depends on | the weather … … prices will | depend on | demand … … it all | depends on | your point of view … … millions of people | depend on | the river … … the result | depended on | one vote …
Invented lines in concordance format. Learners notice that depend is followed by on, and that what follows can be a person, a thing or a whole idea.
In the classroom
- 1
Print eight to twelve concordance lines for a confusing pair (say and tell, make and do) and ask learners to state a rule they can then test on a fresh set of lines.
- 2
Run three steps: learners identify the pattern, sort the examples into groups, then write a rule in their own words and check it against new data.
- 3
Model one search in a free tool such as SkELL, then ask learners to settle their own writing doubts, such as a collocation they are unsure of, before asking you.
Myth and reality
DDL needs a computer lab, advanced learners and a background in linguistics.
Printed concordance lines work: Boulton (2010) used paper-based materials with lower-level learners. Choosing clear examples and designing a guided task matter more than the technology.
Reflect
When a learner asks you a usage question, when would handing over the evidence teach more than handing over the answer?
Worth a minute before your next lesson, or a conversation in the staffroom.
SourcesChecked against Crossref and library records
Johns used the term from around 1990, and it reached print in Classroom Concordancing (ELR Journal 4, 1991), which he edited with Philip King. This 1994 chapter is an early, widely cited account of the approach.
Johns, T. (1994). From printout to handout: Grammar and vocabulary teaching in the context of data-driven learning. In T. Odlin (Ed.), Perspectives on pedagogical grammar (pp. 293–313). Cambridge University Press.
Boulton, A., & Cobb, T. (2017). Corpus use in language learning: A meta-analysis. Language Learning, 67(2), 348–393.
The key meta-analysis: 64 studies, large average effects, and a candid list of what research still lacks.
Lee, H., Warschauer, M., & Lee, J. H. (2019). The effects of corpus use on second language vocabulary learning: A multilevel meta-analysis. Applied Linguistics, 40(5), 721–753.
Focuses on vocabulary and shows which learners, tasks and corpora are linked to bigger gains.
Boulton, A. (2010). Data-driven learning: Taking the computer out of the equation. Language Learning, 60(3), 534–572.
The classroom-friendly case for paper-based DDL with lower-level learners, backed by an experiment.
Boulton, A., & Vyatkina, N. (2021). Thirty years of data-driven learning: Taking stock and charting new directions over time. Language Learning & Technology, 25(3), 66–89.
An open-access map of three decades of DDL research and where the field is heading.
ConnectionsSee it in the graph
- Concordancenot yet writtenThe line-by-line corpus display that DDL puts in learners’ hands.
- The Noticing HypothesisDDL tasks are designed to make learners notice patterns in the input.
- Focus on FormAnother way of drawing learners’ attention to form, usually during meaning-focused work.
- Semantic Prosodynot yet writtenConcordance lines reveal the hidden attitude a word tends to carry.
Know a teacher who’d like this?
Send them the card. It’s made for Facebook, Zalo, LinkedIn and Instagram, and links back here.