Studying medicine in English as a second language

There is a specific frustration that non-native English speakers in medical school describe, and almost no study advice addresses it directly.

You read the textbook chapter and understand it. You could explain the topic to a friend in your own language without hesitating. Then you sit the exam, or you get asked a question in an oral, and it comes out worse than what you know. Your mark says you understood the material less well than you did.

That gap is not an English problem in the way people assume. Fixing it with a better dictionary does not work, which is why students who try that approach stay stuck. It is worth understanding what is actually happening, because the right tool is different from the obvious one.

What the language barrier actually costs you

It is not vocabulary

By second year, most students in an English-medium programme have the vocabulary. Medical English is in some ways easier for a non-native speaker than everyday English — it is Greek and Latin, standardised and regular, and every student in the room is learning it for the first time. Nobody is a native speaker of anastomosis.

If you were losing marks on vocabulary, a dictionary would have solved it already. Something else is going on.

It is working memory

Here is the actual mechanism. Following a lecture requires holding several things at once: the sentence being spoken now, the argument it belongs to, the diagram on the slide, and how it connects to last week. That is close to the limit of what anyone can hold.

Processing a second language is not free. It is fast and it feels automatic, but it consumes some of the same capacity. So a native speaker spends nearly all of their working memory on the medicine, and you spend most of yours on the medicine and the rest on the language.

Under normal conditions the difference is invisible. Under load — a fast lecturer, an unfamiliar topic, an oral exam, the last hour of a long paper — the reserve runs out, and it runs out on the medicine, because the language processing is not optional.

This explains the thing that confuses students most: why you understand the same slide perfectly at home and lose it in the lecture hall. Nothing about your knowledge changed. The time pressure did.

It costs you twice at the end

The second cost lands in exams, and it is worse in some formats than others.

In a written exam you are producing medicine and producing English, under time pressure, and the examiner cannot separate the two when marking. A short-answer question that a native speaker finishes in three minutes takes you four. Multiply across a paper and you are answering a materially shorter exam.

In an oral exam it is more direct. You have to retrieve the fact, assemble a sentence, and speak it aloud with an examiner watching — and hesitation reads as uncertainty about the medicine, even when it is a search for a word. This is where the gap between what students know and what they demonstrate is widest.

Why translation tools do not fix it

The obvious response is to translate. Translate the slides, read the lecture in your own language, translate back for the exam.

This fails for a reason worth spelling out.

Translating adds a step to every retrieval. If you learned a concept in your first language and are examined in English, then every time you need it you must retrieve it and convert it. That conversion happens under exam pressure, which is precisely when you have no spare capacity. You have moved the cost from studying, where you had time, to the exam, where you do not.

It also weakens the memory. Recall is strongest when the conditions match those in which you learned. If your exam is in English, English is the language you need to have practised retrieving in — not the one you comfortably studied in.

There is a narrow, legitimate use for translation: unblocking a single concept you genuinely cannot parse. Get unstuck in your own language, then re-encode the concept in English immediately. Use it as a bridge, not as a workflow.

The risk nobody warns you about

This one is specific to AI tools and it deserves its own section, because it is the way a non-native speaker is most likely to be quietly harmed.

Language models produce fluent, confident, well-organised English. That is what they are optimised for. Fluency and accuracy are separate properties, and a model can be completely wrong in flawless prose.

Native speakers have a partial defence they are not aware of using. They pick up small signals — an unnatural phrasing, an over-hedged sentence, a claim stated with more certainty than the topic warrants — and those signals prompt a second look. That intuition is built from decades of exposure to how confident English usually sounds.

If English is your second language, you have less of that signal, and you have it in the same domain where you are least able to check the content independently. You are more exposed to a confidently-worded error, and less likely to catch it. This is not a comment about anyone’s ability. It is a straightforward consequence of where the safety net comes from.

The defence is structural, not linguistic: use tools that show you where an answer came from.

If a summary tells you something and links it to the slide it came from, you do not need to judge whether it sounds trustworthy. You open the slide. That converts an unfair test of linguistic intuition into a check anyone can perform in ten seconds. We have written about how compounding a small error rate across a topic plays out, and the arithmetic is unforgiving — but the fix is simple and it is the same one every time. Traceability beats trust.

If a tool will not tell you where a claim came from, you are being asked to evaluate it on how confident it sounds. That is a bad deal for anyone, and a worse one for you.

What actually helps

Four things, in order of how much they matter.

1. Move the language work out of the lecture

The lecture is where you have the least spare capacity, so stop trying to do comprehension there. Read the slides beforehand — even ten minutes — so that the unfamiliar terms are already familiar when they are spoken. You are not pre-learning the medicine; you are removing the language cost from the moment where you cannot afford it. This is the single highest-return change available, and it is free. There is a longer playbook for this.

2. Work from your own course material, in English

Whatever you use should be built from your lectures, in the language of your exam. Not a translated version, and not a resource written for a different syllabus in another country. Your examiner’s terminology and emphasis is the thing you are actually being tested on, and it is the one thing no external resource contains.

3. Practise retrieving in English, not recognising in English

Recognition and retrieval are different skills, and the language gap is much larger for the second. If your practice is entirely multiple-choice, you are training the skill where your disadvantage is smallest and neglecting the one where it is largest. Short-answer practice — where you produce the answer rather than pick it — is uncomfortable for exactly the reason it is valuable. We measured how much less information a multiple-choice question carries, and for a non-native speaker the gap is wider still, because recognition lets you lean on familiarity in a way production does not.

4. Practise speaking, badly, in private, repeatedly

Oral exams are where the gap is widest and where students prepare least, for an obvious reason: it takes another person willing to question you repeatedly, and asking a classmate to do that four times a week is not a real option.

This is worth stating plainly, because students blame their own discipline for it. It is not a discipline problem. It is a logistics problem. The fluency needed to answer a viva question smoothly comes from having assembled that kind of sentence many times before, and the reason you have not is that there was nobody to practise with at eleven at night.

Being able to run a spoken examination alone, on your own material, as many times as you want, is the part of this that changed recently — and it is the single largest thing AI tools have added for non-native speakers specifically. Not the explaining. The rehearsal.

Where KoiSwarm fits, and where it does not

We build a study workspace for medical students, and the design assumptions line up with the four points above: it works from your uploaded lectures rather than a national syllabus, its notes cite the slide behind each fact, it produces short-answer and spoken practice rather than only multiple choice, and it runs an oral examination that will question you at any hour without needing a partner.

We are not claiming this removes the language barrier, and you should be suspicious of anything that does. Working memory is working memory. What tooling can do is stop wasting your reserve on things that were never the point — hunting for a term, deciding whether a summary is trustworthy, waiting for a study partner who is busy.

What it cannot do is make the exam easier. You will still be doing two things at once while the person next to you does one.

The short version

Your vocabulary is not the problem. Your capacity under time pressure is, and it is finite, so protect it: front-load the language work into moments where you have time, keep everything in the language of your exam, and practise producing answers rather than recognising them.

And insist on tools that show their sources, because the ability to judge confident English by ear is the one advantage you do not have, and it is the one that matters most when a tool is wrong.