Some learners in Taiwan understand far more English than they ever produce, and the wall between the two is the feeling of being judged the moment they open their mouth. For these learners, treating errors as information instead of as a verdict does more for speaking than extra input or a smaller class. The effect has a name in the research, and a measurable size. The learner who knows the answer but will not say it is a familiar figure in any classroom that runs on cold correction.
Where the wall comes from
Horwitz, Horwitz and Cope defined foreign language classroom anxiety in their 1986 Modern Language Journal paper, which introduced the Foreign Language Classroom Anxiety Scale. They described anxiety as made of three strands: communication apprehension, test anxiety, and a fear of negative evaluation. The third strand is the one a teacher can change. A learner who believes every error invites a judgement will ration the number of sentences they attempt, and the sentences they do attempt get shorter and safer.
This pattern is visible in a class. The same three students volunteer, the rest wait for the teacher to call on them, and when called, they produce a five-word answer and look down. The problem is the exposure. The material is rarely at fault.
That rationing carries a cost beyond emotion. MacIntyre and Gardner showed in their 1994 Language Learning study that anxiety disrupts language at each stage of processing: taking in input, organising it, and producing output. An anxious learner underperforms even on material they know, because self-monitoring occupies part of their working memory rather than the language itself.
The size of the effect
Researchers have measured the size of the effect. Teimouri, Goetze and Plonsky pooled 97 reports, 105 independent samples and 19,933 learners from 23 countries in their 2019 meta-analysis in Studies in Second Language Acquisition. Across 216 reported correlations, the mean was about -.36, a moderate negative relationship between anxiety and achievement that held across measures. The authors also report the relationship varying with the type of achievement measure, the learner's level, and the target language, which makes the number a useful average rather than a universal law.
How it shows up
A learner in this state shows the same habits again and again. They rehearse a sentence silently before saying it. They answer in the fewest words that will end the turn. They avoid the passive, the subjunctive, and any structure they might get wrong, so their spoken English stays permanently simpler than their reading. A 34-year-old account manager who reads contracts in English but opens every meeting with a rehearsed two-line greeting is not short of vocabulary. She is spending her working memory on watching herself.
This self-watching is the mechanism the research points to. When attention goes to monitoring the next word for a mistake, less attention remains for the message. The learner sounds more fluent in writing than in speech, and the gap between the two is a useful clue that anxiety, rather than knowledge, is doing the damage.
What a teacher can change
What follows for a teacher is specific. Shrinking a class helps only if the smaller room changes how the teacher handles errors. A four-person class where the teacher corrects every mistake aloud, on the spot, and alone can feel more exposed than a twelve-person class where the teacher treats errors as shared material. The lever is the error climate, and it has concrete parts: correct fewer things, let the learner finish the repair, and separate the correction from the judgement.
A concrete version for a speaking class: let the learner complete the sentence before any correction, correct at most one error per turn, and phrase the correction as a question about the sentence. The goal is a learner who starts a second, longer attempt without being asked twice.
Choc's own correction approach works this way, and there is an article on why the learner should finish the repair. The anxiety angle adds one instruction on top: the teacher should signal, in words, that an error is information about the sentence and not about the person. That signal can be as plain as "good try, one part needs changing," said before any correction.
The boundary
Boundaries matter here. Some learners are not anxious and some pressure genuinely sharpens them, so a blanket rule to remove all evaluation would help nobody. And the meta-analysis reports a correlation, not a proof that anxiety causes low achievement. The arrow can run both ways, with weak learners growing anxious as the material outpaces them. The practical claim is narrower: for the learner whose silence comes from fear of judgement, an error climate that treats mistakes as information removes a barrier that no vocabulary list can.
Tomorrow, pick the one learner in your class who answers in monosyllables. Correct nothing for a full ten minutes. After the first answer, say only "good, now add one detail." Watch whether the second answer grows. The thing to measure is the number of words the learner risks, not accuracy.