The Speaking-Hours Gap: Why University Language Centres Can't Staff Their Way to Fluency

Every university language centre knows this arithmetic, even if nobody writes it down.

A conversation class works at ten to fifteen seats — beyond that, each student's share of actual speaking time collapses into minutes. An instructor can teach perhaps twenty contact hours a week. Meanwhile the population that needs spoken fluency keeps growing: students facing a B1 or B2 accreditation requirement to graduate, outbound Erasmus cohorts with a mobility date on the calendar, incoming international students, doctoral candidates heading to conferences, lecturers moving into English-medium instruction.

Divide one number by the other and you get the speaking-hours gap: the difference between the spoken practice your students need and the spoken practice any realistic staffing budget can supply. You cannot close it by hiring, because the bottleneck isn't budget alone — it's that speaking practice has, until now, required one qualified human per conversation.

What actually builds speaking ability

The research consensus on speaking is unfashionably simple: learners get better at speaking by speaking — frequently, with feedback, slightly above their comfort level. Three conditions matter:

1. Volume. Regular production beats occasional immersion. Ten minutes daily outperforms a ninety-minute session every two weeks.
2. Feedback that lands. Corrections at the moment of the error, in context — not a generic worksheet the following week.
3. Accountability for the fix. An error corrected once isn't learned. It's learned when the student produces the correct form later, unprompted, in a new context.

Group classes deliver condition 2 well and conditions 1 and 3 barely — again, structurally, not through any fault of teaching. This is precisely the shape of problem that real-time conversational AI happens to fit.

What AI conversation practice does well — and what it doesn't

Honesty first, because this market has been oversold to.

A live AI conversation partner is genuinely good at volume and patience: it is available at 23:40 before the oral exam, it never tires of the same B1 mistake, and it costs a fraction of a private conversation hour. Done properly, it is also good at the accountability loop — logging every error, generating drills from the student's own mistakes rather than a generic syllabus, and re-testing old errors weeks later in new contexts to confirm they stayed fixed.

What it should not do is replace your instructors or your assessment. Certification judgement, curriculum design, cultural nuance, the motivation that comes from a human who knows your name — those remain the language centre's craft. The right mental model is a practice room, not a substitute teacher: students arrive at conversation class having already done their speaking reps, and instructors spend class time on what humans do best.

What this looks like in practice

LinguaLive's current individual-learner product is a live voice tutor: learners hold free-form spoken conversations in built-in scenarios such as an interview or hospital visit and receive corrections as they speak. Around those conversations, the learner can use:

• a typed-response level assessment whose result is saved for the learner;
• a mistake ledger per student — every error logged and categorised;
• personalised drills generated from those errors, not from a textbook;
• spaced re-testing that retires an error only when the learner proves it's fixed.

These are learner-facing features, not an institutional reporting system. LinguaLive does not currently ship rubric-graded speaking assessments, teacher dashboards or reports, teacher-visible grading, class-code or cohort administration, or institutional exports.

For a language centre, those missing workflows should become explicit pre-pilot requirements: define an inspectable speaking rubric and human oversight for any baseline/endline measure; specify exactly what teachers can see; agree a cohort export format; and implement and verify the provisioning, access-control, retention, and reporting paths before enrolment. The current product cannot yet produce stakeholder-ready cohort or pre/post evidence.

The budget question, answered the boring way

AI products have earned a reputation for unpredictable costs. LinguaLive already enforces a hard daily speaking cap per learner on its servers. Any eligible future institutional agreement should price against agreed caps, state whether overages exist, and explain what happens when a participant reaches the limit; those terms belong in the signed pilot specification.

Where to start

LinguaLive is not currently offering live institutional pilots while provider eligibility and legal clearance for the exact use remain unresolved. If that gate is cleared, a credible pilot would start with a defined cohort, fixed commercial terms, and a dated list of pre-pilot deliverables — including whatever class provisioning, teacher reporting, assessment rubric, and evidence export the evaluation requires. None of those institutional workflows should be treated as available until it has been implemented and verified.

If your centre has a cohort with a speaking deadline — a B2 exam window, a mobility departure, an EMI transition — that's the right cohort to pilot with.

Going deeper: Seven questions every institution should ask an AI language-tool vendor (including us) · LinguaLive for schools, universities & academies