Learning analytics is the measurement, collection, analysis and reporting of data about learners and their contexts in order to understand and improve learning and the conditions in which it happens. The field crystallised at the first Learning Analytics and Knowledge (LAK) conference in 2011, organised by figures including George Siemens, and was institutionalised by the Society for Learning Analytics Research (SoLAR), whose 2011 framing of the term as "the measurement, collection, analysis and reporting of data about learners and their contexts, for purposes of understanding and optimising learning and the environments in which it occurs" became the field's reference definition. The label sits next to educational data mining, a closely related strand more focused on automated pattern discovery; the two overlap heavily in methods while differing in emphasis.
The raw material is the digital trace learners leave behind. A learning management system logs every login, page view, forum post, quiz attempt and submission deadline; language apps record which items a learner reviewed, how long a translation took, how many speaking attempts a sentence needed. Analysts run this through four broad lenses: descriptive (what happened), diagnostic (why), predictive (what is likely next), and prescriptive (what to do about it). Methods range from simple aggregation to machine-learning models that flag students at risk of dropping out. The output most learners and teachers actually see is the dashboard, a visual summary of progress, engagement and standing relative to a cohort, intended to make patterns legible at a glance and to prompt timely intervention.
Language education is data-rich in ways that suit analytics. Adaptive apps already lean on it: spaced-repetition schedulers and the difficulty-tuning behind adaptive testing are learning analytics in miniature, reading per-item performance to decide what comes next. Trace data on which grammar points a class repeatedly fails, or which reading passages get abandoned, can feed formative assessment and reshape teaching. There is a natural adjacency to data-driven learning, where corpus evidence informs what learners study, and to online assessment platforms that capture fine-grained response data. As AI in language teaching tools proliferate, the analytics they generate increasingly drive the personalisation they promise.
The sharpest scholarly critique is Neil Selwyn's "What's the Problem with Learning Analytics?" (Journal of Learning Analytics, 2019, 6(3), 11–19), which reads the field sociotechnically and warns that analytics can entrench the status quo, disadvantage already-vulnerable groups, and subordinate education to a profit-driven data economy. Several concrete problems recur. Privacy and consent are fraught: under regimes such as the GDPR, genuinely informed, voluntary consent is hard to secure at scale, lengthy privacy notices produce only superficial agreement, and the right to withdraw or erase data collides with models that depend on long-term records. The metrics are often reductive proxies standing in for things they cannot directly measure; time-on-task, a workhorse of predictive models, is estimated by crude heuristics that cannot tell focused study from a tab left open, so engagement and learning get quietly conflated. Predictive models can carry algorithmic bias, systematically disadvantaging students by gender, socio-economic status or background when trained on skewed data. And the whole enterprise risks false precision: education runs on ambiguity and judgement that clean dashboards flatten, and the appearance of objectivity can license a quiet drift toward surveillance, where being watched reshapes how learners behave.
Read dashboards as prompts for a conversation, not verdicts. A red flag on a struggling student should trigger a teacher's inquiry, not an automated penalty, since the model knows only what the system logged. Be candid with learners about what is collected, why, and who sees it, and offer real choice rather than a buried checkbox. Triangulate any analytic signal against classroom evidence before acting; a low engagement score may reflect offline study the platform never saw. Favour metrics that map onto genuine learning outcomes over those that merely measure clicks, and treat an at-risk prediction as a hypothesis to test with the student, never a label to apply to them. The professional skill is knowing what the numbers omit.