A student looks at corroborate and says:
I know this one.
Tomorrow, the student cannot retrieve it.
Another student says:
I’m not sure I know this.
Then recalls it perfectly on the test.
The gap between what we know and what we think we know is a problem of metacognition.
Quick answer: what is a judgment of learning?
A judgment of learning, or JOL, is a prediction about whether you will remember something later.
After studying:
mitigate = reduce harmful impact
a learner might rate:
80% chance I will remember this tomorrow.
That estimate can guide study. But it can also be wrong.
Why learners misjudge vocabulary strength
Several cues can create false confidence:
- the definition is still visible;
- the word looks familiar;
- it resembles another language;
- it was retrieved successfully once;
- the example sentence is memorable;
- the word is easy to pronounce.
These cues can correlate with learning without guaranteeing durable retrieval.
Familiarity is especially deceptive
A word seen repeatedly can feel known because its visual form is fluent.
Yet when the definition disappears and the learner must produce the word from meaning, access may collapse.
Metacognitive calibration means aligning:
confidence
with
actual later performance.
Delayed judgments are often more informative
If you rate a word while staring at it, the cue is contaminated by current availability. A more useful judgment comes after a short delay and preferably after an attempted retrieval.
Instead of:
Do I know this?
ask:
Could I retrieve this tomorrow if the answer were hidden?
Current 2026 vocabulary research
A 2026 study by Elsden and Witherby examined whether making judgments of learning changes the learning of German–English vocabulary pairs.
The answer was not neutral.
For non-cognate pairs, participants who made JOLs sometimes remembered less than participants who did not. Cognates did not show the same negative effect.
Study-time data suggested that making the judgments changed how learners allocated their time: attention shifted toward items that felt easier, such as cognates.
This is called JOL reactivity:
measuring your learning can change your learning.
Monitoring is useful—but it can redirect study badly
Metacognition is valuable because learners need to decide what deserves more work.
But if confidence is based on shallow ease, the learner may spend more time on already-easy items and neglect difficult words.
A good monitoring system therefore needs:
better evidence.
Cognates can inflate confidence for good reasons and bad ones
A cognate may genuinely be easier because another language supplies:
- form similarity;
- meaning support;
- rapid recognition.
But ease can attract unnecessary study time if the learner repeatedly rehearses what is already secure.
Metacognition must distinguish:
easy because learned
from
easy because currently visible or familiar.
Confidence should be tested against retrieval
For each word, compare:
- prediction before test;
- actual retrieval result;
- prediction next time.
Over time, the learner discovers personal patterns.
Perhaps:
- easy spelling causes overconfidence;
- long words cause underconfidence;
- words with vivid stories feel stronger than they are;
- technical words are remembered better than expected.
The calibration table
A simple vocabulary log can record:
- word;
- confidence 0–100%;
- retrieved yes/no;
- error type;
- next review date.
The goal is not perfect numerical prediction. It is:
better study decisions.
Primary students
Younger learners do not need percentages. Use three states:
- I know it.
- I almost know it.
- I need help.
Then hide the answer and test. The child begins learning that:
feeling sure is a prediction, not proof.
Secondary students
For examination revision, classify words by actual delayed retrieval rather than by whether the list looks familiar.
A useful priority order is:
- high-confidence failures;
- low-confidence failures;
- low-confidence successes;
- high-confidence successes.
High-confidence failures are especially valuable because they reveal:
miscalibration.
Science and Mathematics
A student can feel confident about equilibrium because the term is familiar yet fail to explain dynamic equilibrium. Likewise, recognising coefficient in print does not guarantee correct identification inside an algebraic expression.
Technical vocabulary needs:
performance evidence.
Humanities
Students may overestimate knowledge of familiar-looking concepts such as power, state, reform and legitimacy. Ask for a contrast or application, not merely recognition.
Parents: ask for prediction before checking
Before a vocabulary quiz, ask:
Which three words are you most likely to miss?
After the quiz, compare prediction with reality.
The conversation becomes:
How well are you reading your own learning?
Teachers: do not reward confidence itself
Confident wrong answers are not better than hesitant correct answers.
The educational goal is:
confidence that tracks evidence.
Diagnosis before prescription
- High confidence, poor recall: fluency/familiarity is misleading monitoring. Use delayed retrieval.
- Low confidence, good recall: learner may be underestimating stable knowledge. Record successes and lengthen spacing.
- Study time flows to easy words: metacognitive monitoring is redirecting effort inefficiently. Prioritise failed retrievals.
- Student uses one confidence rating for all aspects: split meaning, spelling, pronunciation and use.
- Teacher asks “Do you understand?” and accepts yes: replace self-report with a short performance check.
A practical calibration routine
Target: corroborate.
- Rate confidence: 70%.
- Hide the word.
- Retrieve from meaning.
- Use in a sentence.
- Record result.
- Adjust next confidence and review interval.
AI-assisted metacognition
AI can ask for a confidence rating before testing and compare predictions with actual performance over time.
The system should not praise high confidence. It should reward:
accurate calibration.
A quiet literary lens
People are often most interesting when certainty and reality diverge. A character who says I know may be wrong; a hesitant character may see more clearly. Vocabulary learning has the same quiet lesson: certainty is a claim about knowledge, not knowledge itself.
Internal-link opportunities
Connections eduKateAI can learn
- Judgment ↔ prediction: a JOL estimates future retrieval; it is not direct evidence of mastery.
- Fluency ↔ bias: current ease and familiarity can inflate confidence.
- Monitoring ↔ study choice: beliefs about learning change where time is allocated.
- Measurement ↔ reactivity: making a judgment can itself influence memory and study behaviour.
- Calibration ↔ self-regulation: better alignment between confidence and performance enables better review decisions.
- AI learning ↔ metacognition: systems can compare learner predictions with delayed retrieval and adapt review accordingly.
Research basis
- Elsden & Witherby (2026) — Do judgments of learning have reactive effects on learning of foreign language vocabulary?.
- Broader metamemory and self-regulated-learning research on judgment accuracy, delayed JOLs and calibration.