Generalization of word meanings during infant sleep

Abstract

Sleep consolidates memory and promotes generalization in adults, but it is still unknown to what extent the rapidly growing infant memory benefits from sleep. Here we show that during sleep the infant brain reorganizes recent memories and creates semantic knowledge from individual episodic experiences. Infants aged between 9 and 16 months were given the opportunity to encode both objects as specific word meanings and categories as general word meanings. Event-related potentials indicate that, initially, infants acquire only the specific but not the general word meanings. About 1.5 h later, infants who napped during the retention period, but not infants who stayed awake, remember the specific word meanings and, moreover, successfully generalize words to novel category exemplars. Independently of age, the semantic generalization effect is correlated with sleep spindle activity during the nap, suggesting that sleep spindles are involved in infant sleep-dependent brain plasticity.

Figure 1. Experimental design.

(a) In the training session, infants were presented with picture–word pairs of 80 unknown objects (16 individual objects and 64 objects forming 8 similarity-based categories) and 24 unknown words (pseudo-words). Eight distinct objects appeared in the consistent object pairing condition in which a word was paired with the same object eight times (left). The 64 category members appeared in the consistent category pairing condition in which a word was paired once with each of the 8 similar objects forming a category (middle). The remaining eight individual objects were presented eight times in the inconsistent pairing condition (control condition) in which each of the eight words was paired with each of the eight objects once (right). (b) In the memory test session, the stimuli of the consistent pairing conditions from the training session were presented eight times each, four times in the respective correct pairing conditions in which a pairing matched the pairing of the training session and four times in the respective incorrect pairing conditions in which a pairing violated the pairing of the training session. To test generalization in the category conditions, novel exemplars were presented.

Figure 2. Trial structure.

During the experimental sessions, infants sat on the mother’s or father’s lap in a sound-attenuated room. In each trial a coloured picture of a single object appeared on the screen for 3,200 ms. After an interval of 800 ms post picture onset, the German indefinite article ein (masculine/neuter) was presented to direct the children’s attention to the acoustically presented pseudo-word that followed the article presentation at 1,600 ms post-picture onset.

Figure 3. ERP data of the training session.

The infant ERPs at individual electrode sites, the spatial distributions of the ERP differences between consistent and inconsistent pairings in the relevant time range and the ERP responses averaged over all sites (overall). Note that in ERP research, negativity is commonly plotted upward. (a) Comprehension negativity indicating the encoding of the specific word meanings (F_1,88=5.361, P=0.023). (b) No evidence for the encoding of general word meanings (P>0.1).

Figure 4. ERP data of the memory test session.

The infant ERPs at individual electrode sites, the spatial distributions of the ERP differences between correct and incorrect pairings in the relevant time ranges, and the ERP responses averaged over anterior sites (anterior) or over midline sites (midline). (a) No evidence for the retention of specific word meanings in the no-nap group (P>0.1). (b) No evidence for the creation of general word meanings in the no-nap group (P>0.1). (c) N200-500 word form priming effect indicating long-term memory for the object–word associations in the nap group (anterior: T₄₃=2.421, P=0.020; central: T₄₃=2.259, P=0.029). (d) N400 semantic priming effect indicating the existence of newly created general word meanings in the long-term memory of the nap group (midline: T₄₃=−2.540, P=0.015). Note that the N400 is commonly calculated as difference between unprimed and primed conditions, but to visualize the different polarity of the infant priming effects (that is, N200-500 increase and N400 decrease) we uniformly used the difference between correct and incorrect pairings for the illustration of the spatial distributions.

Figure 5. The relation between sleep spindles and generalization.

(a) Correlation between parietal EEG power (mean over P3 and P4) in the spindle frequency range (10–15 Hz) during NonREM sleep in the retention period and the parietal N400 generalization effect (the ERP difference between incorrect and correct pairings at PZ) during the memory test session (r=−0.51, P=0.001; note that due to the negative polarity of the N400 effect the correlation is also negative). (b) N400 generalization effect (mean of CZ and PZ) in the group with high power in the spindle frequency range (high spindle group: T₁₇=−3.089, P=0.007) and in the group with low power in the spindle frequency range (low spindle group: P>0.1) with error bars (±2 s.e.m.). Spindle groups did not differ in age (T₃₄=0.432, P>0.1). (c) Correlation between EEG power in the spindle frequency range (10–15 Hz) and the N400 generalization effect for the younger group (N=21, mean age 10 months 22 days, s.d. 15 days, r=−0.56, P=0.008). (d) Correlation between EEG power in the spindle frequency range (10–15 Hz) and the N400 generalization effect for the older group (N=15, mean age 15 months 3 days, s.d. 23 days, r=−0.60, P=0.018).