The freeze is a capacity problem wearing an emotional coat. Watching how you sound draws on the same limited attention that assembling the sentence needs, so being nervous does not merely feel bad: it removes, before you open your mouth, a share of the resource the sentence was going to be built with.
That is why the collapse is so disproportionate. Somebody who writes competently and reads comfortably produces three broken words in a room, and the gap is not a gap in their English. It is that in a room, a large part of what they have is already committed to watching themselves have it.
Eight things worth knowing before you open your mouth
- mistakes do not stop understanding. Almost every learner error leaves the meaning intact, and the ones that do not are a short list.
- most people are not listening for grammar. They are listening for what you mean, and they are considerably less interested in you than the fear suggests.
- errors fossilise from avoidance, not from speaking. A structure you never attempt never gets corrected, which is the actual risk.
- the listener is not marking you unless they are an examiner, and in that case they are marking communication first.
- prepare three topics in advance before any event where you will have to speak, and the preparation removes most of the anxiety.
- a script for the first minute. The opening is where nerves are worst, so having it memorised carries you into the part that flows.
- start with a question. It moves the speaking to the other person while you settle.
- short sentences are not simple sentences. Clear short English reads as confident, and long constructions are where errors cluster.
Everything is drawing on one budget
Producing a spoken sentence in a second language runs several jobs at once and they are not independent. The sounds have to be made. The words have to be found. The grammar has to be applied. And on top of all of it sits the layer that is watching the other three and asking whether this is coming out well.
The top layer is the only optional one, and it is the one that inflates under pressure. Take a competent speaker, put them in front of people they want to impress, and the monitoring expands until the layers underneath have nothing left. The words stop arriving, which is experienced as forgetting the language, and which is not forgetting anything.
This also explains the observation everybody makes and nobody trusts: your English is better after a drink, better when tired, better when you do not care about the outcome. Those are all conditions that shrink the monitoring layer, and none of them adds a word to your vocabulary.
Where it stops: waiting until you are ready
The stall is postponement, and it is the most expensive decision available in this subject. A learner decides to speak once their English is good enough, works on everything else in the meantime, and keeps deferring the start because the standard moves with them.
What makes it a genuine trap rather than mere procrastination is that the thing being waited for is produced by the thing being avoided. Speaking is what builds the retrieval that would have made speaking easier, so the delay does not shorten the eventual difficulty, it lengthens it, and it adds years of accumulated dread to a task that was never going to be comfortable on the first attempt anyway.
The related discovery is worth stating plainly because it surprises people. Most learners are more frightened of speaking in front of other learners than in front of native speakers, and it is the wrong way round. A native speaker is not evaluating your grammar; they are trying to work out what you want. Another learner knows exactly what mistake you just made.
Somewhere the stakes are genuinely low
Discord is where a great deal of free spoken practice now happens, in voice channels attached to language communities that are open at any hour with nobody keeping a register. Its value is not instruction. It is that the cost of a bad sentence there is approximately zero: you are one voice among several, nobody is assessing you, and you can leave a channel without explaining yourself.
It replaced waiting for a scheduled class to be the only place you were allowed to speak, which made every sentence expensive by making the opportunities rare. The limitation is that you get hours and nothing else. There is no structure, no correction and no guarantee anyone will talk to you, so it is a place to lower the stakes rather than a place to be taught.
What it looks like when it goes wrong
The recognisable failure is the sentence that is planned in full and never released, because a complete sentence takes long enough to build that the moment for it passes. What looks like shyness is usually a design decision: the speaker has decided the sentence must be correct before it can start.
The repair is to start before it is ready. Say the first three words, or repeat their last three words, or say right, so, and build the rest while your mouth is already moving. It works partly because it buys time and mostly because it removes the thing being protected. A sentence you have already started badly cannot be spoiled by continuing it.
The second failure is preparing an answer while the other person is still talking, which spends the attention that was decoding the rest of their sentence and produces an answer to a question they did not ask.
What the freed capacity gets spent on, and why large stored phrases cost so much less than built ones, is speed without accuracy, accuracy without speed, and how they meet, and the low-stakes room where you can be as bad as you like is drills that work with nobody in the room. Both sit under getting from correct sentences to an actual conversation.
What none of this touches is the other half of a conversation. Lowering the monitoring frees attention for producing speech, and the same attention is what you were using to follow the person opposite, so a speaker who has just relaxed often discovers that they are now missing more of what is said to them: real speed, real accents, no subtitles.