Memory for nonadjacent dependencies in the first year of life and its relation to sleep

The study addresses whether infants in their first year of life can form and retain memory for nonadjacent dependencies (NADs) in speech—relationships between elements separated in time that are central to grammar learning—and how sleep affects such memories. NAD learning supports hierarchical processing in language and cognition and is not uniquely human, having been observed in apes, monkeys, and songbirds. Prior infant research shows that even 4-month-olds can encode NADs in an unfamiliar natural language and 7-month-olds can track NADs in artificial grammars, but retention of these dependencies in the first year had not been examined. Sleep is known to support memory consolidation in infancy, with sleep spindles (10–15 Hz) during NonREM sleep implicated in lexical-semantic generalization and detailed episodic-like memory. The present work tests whether 6- to 8-month-olds retain morphosyntactic NADs after a short retention interval with or without a nap, whether they generalize to new verb stems, and whether specific sleep features (e.g., spindle activity) relate to memory for general regularities versus specific phrases.

- Infants can encode NADs as early as 4 months in natural language stimuli and at 7 months show behavioral tracking of NADs in artificial grammars. - Sleep supports both the nature and magnitude of retention in infants’ early lexical-semantic learning, with spindles associated with generalization (central/parietal regions) and detailed memory (frontal regions) in older infants. - In adults, lexical-semantic and detailed episodic memory are declarative (hippocampus-dependent), whereas native-language syntax is often linked to non-declarative systems; such distinctions are unclear in preverbal infants. Prior work suggests prefrontal cortex activity can alter the type of memory formed for NADs in adults, reflected in ERP changes (e.g., N400 emergence under certain learning conditions).

Design: Two-session ERP study with an encoding (familiarization) session followed by a retention interval (0.5–1.5 h) and a memory test. Between-subjects factor: Group (wake vs nap during retention). Within-subject factors during test: Regularity (regular A–B/C–D vs irregular A–D/C–B), Old/New (old verb stems presented during encoding vs new stems), Laterality, and Region. Participants: N = 85 monolingual 6–8-month-old infants (41 female; mean age ~7 months). An additional 40 infants were tested but excluded for standard reasons (artefacts, fussiness, technical issues, failure to sleep, atypical sleep EEG). Ethical approval obtained; parental consent provided. Stimuli: Sentences from an Italian miniature language used previously for NAD learning studies. Structure: noun phrase (e.g., "la sorella" or "il fratello") + verb phrase consisting of an auxiliary or modal (A = sta; C = può), an intervening verb stem (X or Y), and a suffix (B = -ando; D = -are) dependent nonadjacently on A or C. Prosody controlled via splicing; two counterbalanced versions ensured no acoustic-phonological cues predicted suffix. Procedure: Infants passively listened while watching a silent cartoon to minimize movement. Each session ~10 min with 128 sentences total across sessions. Encoding: familiarization with one language version; 16 verb stems presented in both NAD frames (each specific phrase, e.g., cant-ando, occurred four times). Retention: one group napped (average nap ~39.6 min; total retention ~73.2 ± 12.9 min), another stayed awake with a shorter retention to maintain alertness (wake retention ~37.8 ± 8.8 min). Memory test: presented regular sentences consistent with learned NADs and irregular sentences violating them; half of each set used old stems (seen in encoding) and half new stems (novel) to test retention vs generalization. EEG/ERP acquisition: 64-channel EEG (Ag/AgCl; 10–20 system), 1000 Hz sampling; impedances <10 kΩ. Band-pass filtering and artefact rejection applied; epochs time-locked to suffix onset. Minimum ~10 artefact-free trials per condition aimed. Time windows: early (100–300 ms), middle (400–600 ms), late (encoding: 700–850 ms; memory test: 750–900 ms); an additional 600–750 ms window for spindle subgroup analyses. Regions of interest (frontal, central, parietal-occipital; lateralities defined) were analyzed. Statistical analyses predominantly repeated-measures ANOVAs with Greenhouse–Geisser corrections; t-tests and correlations as appropriate; multiple-comparisons control applied. Sleep recording and spindle analysis (nap group): Portable polysomnography with frontal, central, parietal leads; EOG and EMG recorded. Sleep staged visually. Fast sleep spindles were detected around each infant’s individual peak frequency (~13.6 ± 0.45 Hz) using band-pass filtering and RMS-based thresholding; mean peak-to-peak amplitudes computed per channel and averaged within regions (left/mid-frontal, central, parietal). Subgroups formed by median split into low- vs high-frontal spindle amplitude (each N = 24).

- Infants formed immediate familiarity during encoding, reflected by ERP familiarity effects at early, mid, and late latencies (including an infant Nc-like late frontal negativity increase with familiarity), consistent across wake and nap groups. - Memory test showed retention of NADs: regular structures elicited ERP responses indicating greater familiarity than irregular ones (early-latency memory effect; e.g., Regularity F1,53 = 14.172, η²p ≈ 0.146). - Generalization: Infants generalized NAD knowledge to new verb stems, showing memory effects for both old and new stems. - Sleep not necessary for retention: Both wake and nap groups exhibited memory for NAD regularities after the retention interval. - Sleep altered processing stage: A mid-latency ERP effect diverged by group. The wake group showed a mid-latency pattern resembling encoding familiarity (e.g., right-central 400–500 ms positivity difference; t(36) = 2.913, P = 0.004, d ≈ 0.48). The nap group showed a polarity-inverted N400-like mid-latency effect (irregular more negative than regular) peaking ~500–600 ms over central and parietal-occipital sites (e.g., central: t(27) = 3.303, P = 0.004; parietal-occipital: t(27) = 3.289, P = 0.002), suggesting sleep-dependent evolution of memory representations. - Spindle-related specificity: In nap infants, higher frontal sleep spindle amplitudes were associated with ERP differences indexing specific memory for old phrases (left frontal-central region), including a polarity-inversed effect for old stems in high-spindle infants at 600–750 ms over left frontal areas (t(23) = 2.260, P = 0.034). Correlation between frontal spindle amplitude and left frontal-central ERP difference for old stems: r = 0.600, P ≈ 7.1×10⁻³; no relation for new stems (r = 0.045, P = 0.763), and the correlations differed significantly (z = -2.544, P = 0.004). - Subgroup pattern: Low-spindle subgroup showed a generalized N400-like effect for new stems (500–750 ms), while the high-spindle subgroup displayed both generalized memory for new stems and an additional left frontal effect indexing specific memory for old stems, indicating co-existing generalized and highly specific memory traces post-nap. - Overall, 6–8-month-olds retain and generalize morphosyntactic NADs; sleep contributes to qualitative changes in processing and supports consolidation of highly specific phrase memories linked to frontal spindle activity.

The research question—whether infants retain NADs and how post-encoding sleep affects these memories—is addressed by ERP evidence showing robust retention and generalization across both wake and nap groups. Thus, basic memory mechanisms for grammar-relevant nonadjacent regularities are present by 6–8 months and do not require sleep to persist over a short interval. However, sleep qualitatively modifies the processing stage engaged during retrieval: nap infants exhibited an N400-like mid-latency effect not observed during encoding nor in the wake group, consistent with sleep-dependent evolution of memory representations and potentially reflecting emerging involvement of prefrontal mechanisms, analogous to adult findings where PFC engagement alters the type of memory representation for NADs. Furthermore, frontal spindle activity specifically related to detailed memory for individual, previously encountered phrases (particularly with old verb stems) and predominantly over left frontal cortex, paralleling findings in older infants’ lexical-semantic learning. These results suggest that during infancy, sleep can promote both consolidation of generalized rule knowledge and the emergence or preferential access to specific episodic-like or associative memories, possibly involving immature PFC–hippocampal networks.

This study demonstrates that by 6–8 months, infants retain and generalize morphosyntactic nonadjacent dependencies after brief delays. Sleep is not necessary for short-term retention of these regularities but changes the nature of retrieval, introducing an N400-like processing stage and supporting consolidation of highly specific phrase memories tied to frontal spindle activity. The findings indicate that foundational mechanisms for grammar learning are present early in life and that sleep can concurrently promote generalized and specific memory consolidation. Future research should clarify the causal role of sleep spindles, delineate contributions of prefrontal and hippocampal systems to infant memory for linguistic regularities, and explore how sleep-dependent changes unfold over longer intervals and across developmental stages.

- ERP components of interest could not be specified a priori because no prior ERP studies had examined NAD memory in the first year; analyses therefore referenced familiarity effects observed during encoding to assess memory in the test phase. - Potential individual trait or developmental differences may contribute to the observed relation between left frontal spindle activity and specific memory; causality cannot be inferred. - It remains unclear whether frontal spindles directly mediate consolidation or index a general propensity for verbal memory consolidation; baseline measures of spindle activity were not included. - Group sizes and some procedural constraints (e.g., naturalistic nap scheduling, differing retention durations to maintain wakefulness) may introduce variability not fully controlled experimentally.