zum Inhalt springen

Philosophische FakultätLingCologne

Can we explain false memories for predictable but not encountered words by means of simple semantic relation? - Probably not

Katja Haeuser & Jutta Kray

Saarland University, Germany

Recent studies show that disconfirmed predictions can induce false memories: In word recognition tasks, participants are more likely to falsely recognize predictable but not actually encountered words, compared to genuinely new words. An open question is whether these false memory effects are due to predictive processing – they might as well be a consequence of participants using the semantic relatedness between predictable words (1c) and sentence contexts retrospectively to determine if words were old.

(1) At the hospital, the nurse patches up the pants.

  1. Old: pants
  2. New: paper
  3. Predictable lure: wound
  4. Unpredictable lure: band aid

We present two experiments indicating that predictability-based false memories are not merely a consequence of relatedness. Participants read sentences which disconfirmed initial predictions (Example 1). Subsequently, their recognition memory and their confidence for “old” and “new” words was probed. We also presented initially predictable lures and unpredictable associates, which were semantically related to the sentence contexts, but not likely to have been predicted (cloze probability < .05). Results showed elevated levels of false recognition both for predictable and unpredictable lures (vs new nouns), but false recognition was reduced for unpredictable lures (Figure 1). In addition, false recognition increased with higher confidence for predictable lures, but decreased for unpredictable lures, suggesting that while semantic associates elicited familiarity, only predictable lures elicited vivid remembering. Correct rejections of predictable lures took significantly longer than correct rejections of semantic associates (Figure 2). Thus, predictability-based false memories seem qualitatively distinct from false memories emerging through semantic relatedness.

Figure 1. Partial effects plot showing the effects of condition and confidence on trial-by-trial “old” judgments in the recognition tasks of Experiments 1 and 2. Error bars are SE. For previously seen “old” words, the lines show hits; for “new” and “lure” words, the lines show false alarms. Lure-Predictable: Initially predictable but not presented lure. Lure-Unpredictable: Unpredictable semantic associates to sentence context.
Figure 2. Reaction times (in ms; ± SE) of accurate memory judgments in the recognition task. For lures and new items, the bars reflect correct rejections; for old items, the bars reflect correct endorsments. Lure-Predictable: Initially predictable but not presented lure. Lure- Unpredictable: Unpredictable semantic associates to sentence context.