EEG-based multivariate and univariate analyses reveal the mechanisms underlying the recognition-based production effect: evidence from mixed-list design

Abstract

The production effect (PE) is a phenomenon where reading words aloud, rather than silently, during study leads to improved recognition memory. Human recognition memory can be divided into recollection (recognition based on complex contextual information) and familiarity (recognition based on a sense of familiarity). This study explored how reading aloud affects recollection and familiarity using electroencephalography (EEG) in a mixed-list design. Participants encoded each list item, either aloud or silently during the study phase and made remember/know/new judgments in the test phase, while EEG data were recorded. The behavioral results replicated the classic PE pattern and indicated that the PE was present in both recollection and familiarity. At the Event-Related Potential (ERP) level, the recollection-based LPC (late positive complex) old/new effect at test was largest in the aloud condition; however, the familiarity-based FN400 old/new effect was equivalent when comparing the aloud condition and the silent condition. Moreover, this study was the first to employ multivariate pattern analysis (MVPA) to decode the time course between two distinct memory strategies (aloud vs. silent). The results revealed significant decoding between 760 and 840 ms, which is consistent with the LPC old/new effect. The paper discusses both traditional theories and the Feature Space Theory based on our results, highlighting inconsistencies with assumptions regarding unconscious retrieval in the Feature Space Theory. In summary, the current results support the role of distinctiveness (enhanced memory for auditory or action information, consistent with recollection) in the PE, rather than the role of strength (enhanced memory trace, consistent with familiarity). This study suggests that enhanced distinctiveness/recollection may be a shared mechanism underlying certain advantageous memory strategies.

Keywords: FN400; LPC; MVPA; reading aloud; silent reading.

Figure 1

Flowchart of this study. During the study phase, two colors of fonts appeared randomly mixed. Afterward, participants entered the test phase, and ERP data were recorded during the stimulus stage indicated by the red box.

Figure 2

Behavioral data: overall recognition, recollection, and familiarity by condition. Error bars indicate the standard error of each mean. ***p < 0.001.

Figure 3

Amplitude of LPC and FN400 (Test Phase) ERP components by condition. (A) Waveforms and topography of LPC (test phase). (B) Evoked LPC amplitudes (test phase). (C) Waveforms and topography of FN400 (test phase). (D) Evoked FN400 amplitudes (test phase). (A, C) Shaded regions represent the defined time windows. (B, D) Error bars indicate the standard error of each mean. *p < 0.05 and ***p < 0.001.

Figure 4

Mean classification performance and spatio-temporal decoding in test phase. (A) Spatio-temporal decoding of old/new contrast result in recognition. Classifier accuracy was threshold (cluster-based correction, p < 0.05). Gray lines indicate classifier accuracy. Solid black lines indicate significant clusters. Gray shaded contours in classification performance plots represent standard error of the mean. (B) Temporal generalization matrices. Saturated colors indicate significant samples (p < 0.05).