The Limits of AI in Language Teaching
- 1 day ago
- 5 min read
How many times have we heard that 'evidence-based teaching' is the new thing - as if we had never had it before?
It's not new.
Back in 2007, the same debates were being held. In his book 'The Art and Science of Teaching' Robert Marzano argued that while evidence-based strategies inform instruction, their execution requires the teacher's judgment, intuition and adaptation.
Who would've thought? It's the TEACHER that really matters.
Here's a rough summary - we know that all about the pros and cons of implicit and explicit instruction, the role of formative assessment and the wonders of cognitive load theory. Great teachers know when to break the rules: when to push, when to pause and when to challenge. Great teachers read the room, sense student engagement and adjust in ways no formula can predict. Students do not just respond to instructional techniques; they respond to teachers they trust and respect. Teaching takes oratory, performance, coaching. A kind of genuine professional friendship.
So I worry about the impact of AI on teaching. I really do. Where's the place for the great teacher? I don't mind about the use of AI in lesson preparation, or in report-writing, and all the myriad of little admin things that take up so much time of a teacher's day - and their thoughts at the beginning and end of it, when they're at home with their families or their dogs or their gardens.
The current enthusiasm surrounding AI in education seems to rest on a narrow understanding of what teaching actually is. AI can generate explanations, quizzes, translations, summaries and even lesson plans within seconds. But there's something missing: AI has no pedagogical principles. It can scrape together subject content knowledge. But it possesses no pedagogical content knowledge, no lived experience of classrooms, and no understanding of the subtle decisions teachers make moment by moment in response to learners.
I think this matters enormously, particularly in language teaching. Even with so-called dead languages, such as Latin and ancient Greek.
Often discussions about AI in education reduce teaching to the transmission of information or the improving exam performance.: delivering facts efficiently, rehearsing recall, and training students to reproduce the kinds of responses rewarded by assessment criteria. AI can perform aspects of this model convincingly because much of it is pattern recognition and reproduction. It draws on standardized testing materials produced by external examinations and by teacher-made resources themselves, I suspect. But as a result it appears unable to see beyond them, or to step outside these sorts of materials. It has no imagination.
And it can't get into the classroom. It can't feel like a teacher feels.
Teachers do not simply teach students what to think. They teach students how to construct arguments, how to develop a line of reasoning and how to connect new learning to prior understanding. Through elicitation, prompting, modelling and carefully sequenced questioning, teachers help students see relationships between ideas that may otherwise remain invisible. They cultivate students' own vocabulary (I'm not thinking of Latin vocabulary here) as tools for thinking. They throw in a bit of 'grit', to test their true knowledge. They ask left-field questions, to see how they respond. They make jokes - ones that are shared responses to the situation. Students learn by observing through listening and trying it out for themselves how arguments are built, revised and challenged, and then repeated and refined in authentic intellectual contexts. This is the essence of teaching.
AI just tells you what you'd like to know. It's ever patient. It always listens to you. It never gets bored. AI never says 'Hurry! You've got to finish!', and it doesn't really mean it when it shows pleasure that you've achieved something. It's blandly nice.
The limitations of AI become particularly visible in ancient languages teaching. I have long argued for a broader and more sophisticated repertoire of language-learning activities and resources in subjects such as Latin (see my book Teaching Latin: contexts, theories, practices for example). If Latin learning is conceived primarily as a coding exercise - decoding grammatical structures and matching forms to meanings - then AI will undoubtedly accelerate and automate many tasks. In such a framework, language learning risks becoming little more than procedural problem-solving. Some people will like that: for them, the very essence of 'doing Latin' is the maths.
However, if we understand language learning as an opportunity to widen access, deepen interpretation and provide varied intellectual experiences for diverse learners, then the picture changes considerably. Many students do not learn languages effectively through purely analytical or 'coding' approaches. They benefit from richer contextualisation, narrative engagement, oral work, scaffolded vocabulary development and opportunities to connect language to culture, literature and personal meaning. Does AI care what you think? While AI may be useful, as I have mentioned above, in producing rich pedagogically-aligned texts or other types of language-teaching resources, it could also take the humanity out of the language and out of the learning experience.
One of the defining characteristics of AI-generated educational material is its lack of sharpness. There is a sloppiness to much of its output: generic lesson plans, vague questioning, superficial differentiation, and broad claims detached from the realities of particular classrooms. This is striking at a time when educational discourse has increasingly emphasised 'evidence-based teaching,' while simultaneously acknowledging the limitations of approaches that ignore the professional judgement teachers exercise daily. Teachers constantly draw upon tacit pedagogical knowledge - what Bernstein calls the 'reservoir' of teacher knowledge. It is shaped by experience, subject expertise, classroom relationships, institutional contexts and responsiveness to individual learners.
AI cannot have any of this. It doesn't care.
When I speak of 'sharpness' in teaching, I mean the precise pedagogical choices expert teachers make continually and often instinctively. A good teacher knows when to explain and when to hold back in order to promote productive struggle. They know when to check understanding before independent work begins, how and when to assess, and what to do with the information assessment provides. They anticipate misconceptions before they emerge. They rephrase explanations in response to confusion. They connect abstract concepts to students’ prior knowledge, personalities, experiences and interests. In one class I taught, a pupil wasn't great at Latin, but he knew an awful lot more about tropical fish than I did (still do). Not much connection in terms of language, but a way in to him as a learner - as a human being. In another class, a different pupil was way ahead of me in recognising subjunctive forms of ancient Greek irregular verbs (and why they were there).
These decisions that teachers make - scores of them in a single lesson, sometimes, are not reducible to algorithms or templates.
AI can imitate the appearance of pedagogy because it can reproduce the language associated with teaching. It can generate the form of a lesson plan, the tone of feedback or the structure of a classroom activity. But imitation is not understanding.
I don't mean that AI has no place in education. It may become a valuable assistant for resource generation, accessibility, adaptation, or administrative efficiency, as I have said before. It may help teachers create wider ranges of materials for students with differing needs and learning preferences. Used critically and carefully, it could support rather than diminish professional practice. But at the moment, it lacks the obviously human face of the real live teacher - the one who is genuinely interested in how the pupil is doing. And it cannot replicate the educational hurly-burly of the live classroom. That is something we mustn't lose.
Comments