Two-year-olds consolidate verb meanings during a nap

Introduction

Young children learn new words quickly and then expand and refine their representations of meaning with increasing exposure (see He & Arunachalam, 2017, for a review on word learning). Children’s initial representations of a novel word’s meaning must therefore be sufficiently robust to be retained in memory until the next encounter. The evidence, thus far derived almost exclusively from the acquisition of novel nouns, suggests that initial representations are indeed retained. At issue, however, is whether young word learners encounter the same success in acquiring the meaning of novel verbs. Verbs pose an extra challenge because the initial representations are often gleaned from “fragmented” input. Caregivers rarely label events while they are ongoing, and so the learner cannot observe the referent event while hearing the verb (e.g., Gleitman & Gleitman, 1992; Tomasello & Kruger, 1992). In situations like this, when linguistic and observational information become available at different time points, are young word learners’ initial representations sufficiently robust to sustain a delay?

Years of research have established that both linguistic and observational context provide important information about verb meaning (e.g., Gillette, Gleitman, Gleitman, & Lederer, 1999; Gleitman, 1990; Landau & Gleitman, 1985). The linguistic context provides structural information; for example, blick in “the boy blicked the girl” denotes a causative event whereas in “the boy and the girl blicked,” it denotes a non-causative event. The observational context further instantiates the semantic content of the event (e.g., how a blicking event unfolds). These two sources of information, however, are often temporally “fragmented.” Over half of the verbs in child-directed speech refer to events that occur after the verb has been uttered (e.g., “Let’s pour some milk into the glass”) (Tomasello & Kruger, 1992). Nevertheless, 2-year-old children successfully harvest this fragmented input, integrating linguistic information about a novel verb (e.g., a conversation between two actors who use the verb in informative sentences) with observational evidence about its meaning (e.g., Arunachalam, 2013; Arunachalam, Escovar, Hansen, & Waxman, 2013; Arunachalam & Waxman, 2010; Dautriche et al., 2014; Messenger, Yuan, & Fisher, 2015; Scott & Fisher, 2009; Yuan & Fisher, 2009). For example, Arunachalam and Waxman (2010) and Arunachalam et al. (2013) introduced children to novel verbs, presented in either transitive or intransitive syntactic frames in a dialogue between two actors, without providing any visual cues to the verb’s meanings. Children were subsequently shown two candidate referent scenes, one in which one actor performed a causative action on the other (e.g., spinning) and another in which the two actors engaged in independent non-causative actions (e.g., waving). Children mapped novel verbs presented in transitive frames, but not verbs in intransitive frames, to the causative scene. Because these representations were established from temporally fragmented input, they may well be fragile. Our goal in the current investigation is to test children’s ability to retain such fragmented representations over a delay.

Although the evidence for retaining initial representations of novel nouns over delays is promising (e.g., Carey & Bartlett, 1978; Dollaghan, 1985; Golinkoff, Hirsh-Pasek, Bailey, & Wenger, 1992; Goodman, McDonough, & Brown, 1998; Heibeck & Markman, 1987; Jaswal & Markman, 2001, 2003; Markson & Bloom, 1997; Mervis & Bertrand, 1994; Waxman & Booth, 2000; Wilkinson & Mazzitelli, 2003; Wilkinson, Ross, & Diamond, 2003; Woodward, Markman, & Fitzsimmons, 1994)¹, the evidence concerning verb learning is considerably sparser. Yuan and Fisher (2009) reported that 28-month-olds retained an initial verb representation after a delay of one or two days. This is impressive, but because delays of this duration necessarily include sleep, it raises a compelling question: Is sleep an essential ingredient in maintaining verb representations over delays? In the current investigation, we address this directly, introducing a shorter delay period during which we manipulate whether or not the child napped.

There is substantial evidence that new memories are consolidated during sleep—stabilized, strengthened, and integrated into long-term memory (Diekelmann & Born, 2010; Rasch, Büchel, Gais, & Born, 2007), and that short naps also have a consolidation effect (e.g., Lahl, Wispel, Willigens, & Pietrowsky, 2008). Young children tend to sleep longer at night than adults and typically take daytime naps (Iglowstein, Jenni, Molinari, & Largo, 2003; Ohayon, Carskadon, Guilleminault, & Vitiello, 2004; Weissbluth, 1995). But what remains unclear is whether young word learners’ retention and retrieval of initial representations of novel words are affected by sleep. Some studies report a sleep advantage (e.g., Friedrich, Wilhelm, Born, & Friederici, 2015; Horváth, Myers, Foster, & Plunkett, 2015; Williams & Horst, 2014), but others do not (Werchan & Gómez, 2014). And again, most existing evidence is on nouns.

To our knowledge, there is only a single report of the effect of sleep in learning verbs. Sandoval, Leclerc, and Gómez (2017) found that only children who napped shortly after learning, but not those who stayed awake, retained and generalized the meaning of novel verbs. However, because the novel verbs in their design were presented concurrently with a referent scene, this work cannot address the acquisition of meaning when they must lay down an initial representation based on the linguistic context alone, and then integrate it with observational information when it later becomes available.

To assess how fragmented representations of verb meaning fare over a delay, we invited 2-year-old children and their parents to participate in a study with the two distinct visits (Visit 1 and Visit 2) separated by a 4-hour delay during which the child either slept (Nap Condition) or remained awake (Wake Condition). We compared performance across the two visits to ascertain whether children’s initial representations of novel verb meanings were sufficiently robust to withstand a delay and whether their representations were enhanced with sleep. Specifically, we predicted that children would retain initial representations if the delay included sleep, but that without sleep, the representations would decay.

We take as a starting point the robust evidence that children successfully establish an initial representation of a novel verb even from fragmented input (e.g., Arunachalam, 2013; Arunachalam & Waxman, 2010; Yuan & Fisher, 2009). Adopting the stimuli and design of Arunachalam and Waxman (2010), we focus specifically on learning novel transitive verbs, asking whether and how children’s verb learning is affected by a delay with or without sleep. We target 27-month-olds, children in an active phase of acquiring new verbs.

Methods

Participants.

Forty-two typically-developing, monolingual English-learning children (21 females, 21 males; ages 25.1-29.9 months, mean = 26.8 months) were included in the final sample. Parents reported that all children typically took a daytime nap. We randomly assigned children to the Nap or Wake Condition, adjusting the time of their lab visits so that for half, Visit 1 was before their regular nap time and Visit 2 after, and for the remaining half, Visit 1 and 2 did not span their typical nap time. Families received $25 compensation for participation. Fourteen additional children participated but were excluded from analysis because of failure to conform to condition assignment (5) (see Supplementary Materials), fussiness (5), failure to return for Visit 2(1), or insufficient eye-tracking data (3)².

Stimuli.

The visual and auditory stimuli were identical to those reported in Arunachalam

Procedure.

Visits 1 and 2 were separated by a delay of 4 hours. At Visit 1, the parent provided informed consent while the child played with an experimenter. Next, they escorted to the testing room, where the child sat 18 inches from the eye-tracker monitor, either in a car seat or on the parent’s lap. Each visit included four different trials, each comprising a Dialogue Phase and a Test Phase, featuring a different novel verb. See Figure 1 for an example trial and Supplementary Materials for all trials.

Figure 1.

Design of one example trial (from a total of 4). The visual stimuli were recordings of live actors in conversation and live actors acting on other actors or on objects. The auditory stimuli were recordings of a female native English speaker speaking in child-directed speech. The materials were presented on a 24-inch widescreen eye-tracker monitor (Tobii T60XL; Studio 2.0), which recorded the coordinates of children’s gaze at a rate of 60 frames/sec.

In the Dialogue Phase, the actors mentioned the novel verb 8 times, always in transitive frames (e.g., “The boy is going to biff the lady”). Only linguistic cues to the novel verb’s meaning were available. In the Test Phase, observational materials were presented, with sentences stripped of informative syntax. This phase consisted of three windows: (1) Salience: Children heard “Look!” and saw the two test scenes simultaneously on opposite sides of the screen—one depicting a causative event and the other a non-causative event; (2) Central Fixation: Children saw a centered yellow star to draw attention to the midpoint of the screen and heard a prompt (e.g., “Where’s mooping?”); (3) Response: The two test scenes reappeared in their original locations, and the child heard another prompt (e.g., “Do you see mooping?”) and another 2 seconds later (e.g., “Find mooping”).

After Visit 1, the child and parent left the lab, returning for Visit 2 approximately 4 hours later. Visit 2 included only the Test Phase for each trial, no Dialogue Phase. Therefore, to succeed, children had to retrieve their initial representations for the novel verbs—without the benefit of additional linguistic support—and use them to identify the causative test scenes.

Sleep Information.

Parents logged their child’s sleep/wake activities at 30-minute intervals throughout the 4-hour delay. We also collected parental reports on children’s typical sleep hours outside of the context of the study. See the Supplementary Materials for more information.

Coding and Analysis.

Data points where no eye gaze was captured (e.g., blinks) were excluded (15% of all data points). The included data were then coded at each frame, as 1 (gaze directed to the causative scene) or 0 (directed to either the synchronous scene or to neither scene). We focused on the first 2.5 seconds of the Test Response window as in Arunachalam et al. (2013), calculating the proportion of frames on which gaze was directed to the causative scene for each participant on each trial out of all frames on which a valid gaze coordinate was recorded (i.e., excluding blinks but including looks to neither scene). If sleep strengthens children’s initial verb representations, this should be expressed in a Condition x Visit interaction.

Results and Discussion.

At Visit 1, children’s mean proportion of looking to the causative scene in the Nap Condition was 0.41 (SD = 0.17), and in the Wake Condition, 0.43 (SD = 0.13). At Visit 2, mean target look was 0.48 (SD = 0.19) and 0.40 (SD = 0.13) in the Nap and Wake Condition respectively. To assess whether and how looking time in each condition and each visit varied over the course of 2.5 seconds, we submitted the data to a Growth Curve Analysis (Mirman, 2014), with Condition and Visit as fixed effects, Time as a continuous predictor, and Participant as a random factor. We fit the data with a first-order linear model and a second-order orthogonal polynomial model. Because the latter provided a better fit (χ²(9) = 1196,p < 0.001, ΔAIC=−1178, ΔBIC = −1121), we report the second-order model. Statistical significance for individual parameter estimates was evaluated using an alpha level of 0.05, using normal approximation (i.e., t-values treated as z-values). See Figure 2.

Figure 2.

Fixation time course and growth curve model fit

The model yielded a main effect of Visit (t = 3.09, p = 0.002) and a Condition x Visit interaction (t = 6.32, p < 0.001). To better capture the interaction, we submitted data from each condition separately to a second-order orthogonal polynomial model, with Visit as a fixed effect, Time as a continuous predictor, and Participant as a random variable. In the Nap condition, the proportion of attention devoted to the causative scene increased from Visit 1 to Visit 2 (p < 0.001, β = 0.25); in the Wake Condition, this proportion decreased (p = 0.014, β = −0.085). This reveals that children’s initial verb representations, built upon fragmented information, were enhanced with sleep, an outcome consistent with evidence that initial encodings are strengthened and consolidated with sleep (see Diekelmann & Born, 2010, for review). See Supplementary Materials for more details about the models.

General Discussion

Young children acquire vocabulary at an astonishing rate. Doing so requires that they not only identify the referent of a new word, but also retain their representation of that word’s meaning over a delay until the word is encountered again. In the current work, we have focused on verb learning in particular, because typically the linguistic and observational information are decoupled—children often hear a novel verb in an utterance without the benefit of concurrent observational information (e.g., Tomasello & Kruger, 1992). At issue then was whether children’s initial representations for verbs, formed from this “fragmented” input, were too fragile to withstand delays. Given the role of sleep in learning and memory, we asked whether a daytime nap would better support retention of a novel verb representation over a delay than a period of wakefulness. We found that 27-month-olds’ initial verb representations withstood a delay of 4 hours if sleep was included; without sleep, the representations decayed over the delay.

This result provides several new insights. First, it provides evidence that children’s representations, built upon fragmented input are indeed fragile and subject to decay. Second, these representations, although sparse, are nonetheless sufficient to be retained so that sleep-related memory consolidation can bolster them. This outcome converges well with other evidence on the behavioral, cognitive, and neurodevelopmental benefits of sleep (e.g., Dahl, 1999; Fondell et al., 2011; Hill, Hogan, & Karmiloff-Smith, 2007; Jansen et al., 2011; Touchette et al., 2007).

Third, the current results move beyond seminal work of Yuan and Fisher (2009) in two crucial ways. First, we systematically manipulated sleep; half of the children napped in the delay and half remained wakeful. This manipulation permitted us to test the effect of sleep directly. Second, Yuan and Fisher (2009) imposed a delay between the linguistic exposure, waiting a day or two before providing the observational input. Thus, they tested retention of a representation formed from linguistic information alone. In contrast, children in the current experiment retained a representation built from both linguistic and observational input, presented at different times. The current finding suggests that although this temporal gap between linguistic and observational information may indeed result in fragile representations, memory consolidation during sleep offers one key to integrating them.

Two additional key questions remain. First, what is the nature of the representations that children formed? At Visit 1, children encountered four different novel verbs, all transitive, each depicting a different causative event. To succeed at Visit 2, they could have remembered a) only the linguistic representation established from the Dialogue Phase—that the verb was transitive, b) only the mapping between each verb and its event referent (e.g., moop names the event where one person spins another), or c) both—that the verb was transitive as well as its specific event referent. Ideally, learners would remember both, because this would allow them to distinguish a “mooping” event from other causative events. But the current design cannot tease these apart.

Second, what is the mechanism underlying sleep-related memory consolidation? Does sleep consolidate the initial representation, or simply remove interference from the external stimuli that one would naturally encounter while awake? If the latter, we would expect a wakeful delay that is restful, with minimal external stimuli, would serve the same purpose as a nap. A control condition with children staying awake but engaging in minimal activity would potentially disentangle these possibilities, but we imagine this would be very difficult to carry out with young children. Nevertheless, we think it is unlikely that the nap provided only protection from interference. Importantly, we found an increase in looking to the target scene in the Nap condition, rather than simply no change. Moreover, research on consolidation of other kinds of memories (specifically, visuospatial declarative memories) indicates that naps provide more than just protection against interference for preschoolers (Kurdziel, Duclos, & Spencer, 2013).

In sum, although questions concerning the contribution of sleep remain, the current results reveal that sleep is a key ingredient to the recipe of learning.