Sleep-based memory processing facilitates grammatical generalization: Evidence from targeted memory reactivation

Abstract

Generalization-the ability to abstract regularities from specific examples and apply them to novel instances-is an essential component of language acquisition. Generalization not only depends on exposure to input during wake, but may also improve offline during sleep. Here we examined whether targeted memory reactivation during sleep can influence grammatical generalization. Participants gradually acquired the grammatical rules of an artificial language through an interactive learning procedure. Then, phrases from the language (experimental group) or stimuli from an unrelated task (control group) were covertly presented during an afternoon nap. Compared to control participants, participants re-exposed to the language during sleep showed larger gains in grammatical generalization. Sleep cues produced a bias, not necessarily a pure gain, suggesting that the capacity for memory replay during sleep is limited. We conclude that grammatical generalization was biased by auditory cueing during sleep, and by extension, that sleep likely influences grammatical generalization in general.

Keywords: Abstraction; Generalization; Language acquisition; Learning; Memory consolidation; Sleep; Syntax; Targeted memory reactivation.

Copyright © 2015 Elsevier Inc. All rights reserved.

Figures

Figure 1

Schematic diagram of artificial language learning task. Participants learned to correctly order phrases according to abstract hidden grammatical rules through an interactive, trial-and-error based learning procedure. Participants selected individual words one at a time and were given auditory and visual feedback on whether their choice was correct or not. For a correct choice, the word moved to the bottom of the screen and the spoken word was presented out loud. Auditory recordings of the phrases were embedded in the learning paradigm and presented later during sleep. Examples of both a correct and incorrect response are shown.

Figure 2

Data contributing to the Accuracy Residual. The proportion of phrases predicted to be correct based on Pre-Nap learning performance is shown in dark blue for each group. Observed Post-Nap performance is shown in light red. A significant interaction was found, indicating that participants in the grammar-cued condition showed significantly higher Accuracy Residual values than did participants in the tone-cued condition.

Figure 3

The Accuracy Residual in the grammar-cued group (n=18) and the tone-cued group (n=17), as well as participants who were not successfully cued due to low sleep quality, the no-cues group (n=9). The Accuracy Residual was computed as the difference between each participant's predicted Post-Nap performance, as estimated based upon performance during the Pre-Nap learning task, and his or her observed Post-Nap performance. A larger Accuracy Residual indicates that the participant performed better at the Post-Nap test than would be expected based upon Pre-Nap learning.