Cognitive Psychology · Simplified

Remembering a Word Nobody Said

You will be sure you saw a word that was never on the screen, and you will be able to say who said it.

Study

Two lists, read by two people

Twenty words appear one at a time, about one a second. Each one is labelled with who read it out, Ravi or Nina. Just watch and take them in. There is nothing to press.

What this needs. About four minutes, and no writing anything down. Nothing flashes and nothing moves; the words simply change about once a second.

What happened

The word that was never there

Three kinds of test word: ones that really were in the lists, ones with nothing to do with them, and the two that fit the lists perfectly and were never shown.

Every word you judged


Word	Really in the lists?	You said	How sure	Who you said read it

Key idea: Both lists were built so that every word pointed at one word that was never in them. Taking in the list means the theme gets activated too, and afterwards there is nothing about that activation that marks it as having come from you rather than from the screen. So it feels like a memory, because it is one: it is just a memory of thinking of the word rather than of seeing it. That is why the three judgements can come apart. Saying yes is one thing, being confident is a second, and naming a source is a third, and a false memory can score highly on all three. Confidence is not a measure of accuracy, and the ability to say where something came from is not proof that it came from there.

Two lists and ten test words from one person, so the percentages are rough and a single unlucky guess moves them a long way. The lists here were written for this activity rather than taken from a published set, so they are not calibrated and the effect may be weaker or stronger than the literature reports. And this does not show that memory is generally unreliable, or that anyone is a poor witness. It shows one specific, well-understood way in which remembering can produce something that was never presented, using materials built precisely to produce it.

The longer version adds a third list, a fuller source-monitoring analysis separating the two kinds of error, and a treatment of what warning people in advance does. It is at False Memory and Source Monitoring in the main collection.