Abstract

Statistical learning can be used to gain sensitivity to many important regularities in our environment, including structure that is foundational to language and visual perception. As yet, little is known about how statistical learning takes place in the human brain, especially in children's developing brains and with regard to the broader neurobiology of learning and memory. We therefore explored the relationship between statistical learning and the thickness and volume of structures that are traditionally implicated in declarative and procedural memory, focusing specifically on the left inferior PFC, the hippocampus, and the caudate during early childhood (ages 5–8.5 years). We found that the thickness of the left inferior frontal cortex and volume of the right hippocampus predicted statistical learning ability in young children. Importantly, these regions did not change in thickness or volume with age, but the relationship between learning and the right hippocampus interacted with age such that older children's hippocampal structure more strongly predicted performance. Overall, the data show that children's statistical learning is supported by multiple neural structures that are more broadly implicated in learning and memory, especially declarative memory (hippocampus) and attention/top–down control (the PFC).

INTRODUCTION

Statistical Learning

Decades of research on statistical learning has helped us answer a central question to psychological science: How do humans—especially children—extract meaningful information from a chaotic, information-packed world? Moreover, how does the developing human brain extract regularities from the environment that are not explicitly taught, especially when these (statistical) regularities are foundational to language (Liberman, Cooper, Shankweiler, & Studdert-Kennedy, 1967), visual perception (Turk-Browne, Jungé, & Scholl, 2005), and our ability to perform and plan actions (Nissen & Bullemer, 1987)?

Because statistical learning can be used to gain sensitivity to regularities in the environment (Fiser & Aslin, 2001; Gomez & Gerken, 1999; Saffran, Aslin, & Newport, 1996; Saffran, Newport, & Aslin, 1996) often without a learner's effort or awareness (Saffran, Newport, Aslin, Tunick, & Barrueco, 1997), it offers an important, mechanistic answer to this long-standing question. Initial demonstrations of statistical learning showed that very young infants could learn much more than we ever thought; after only 2 min of exposure to an artificial language, infants treated strings of syllables that regularly occurred in succession (having a high co-occurrence and transitional probability [TP] structure) differently from strings that did not (Saffran, Aslin, et al., 1996). This discovery meant that even infants are able to learn from passive exposure, something that is also true of adults (Saffran, Newport, et al., 1996) and other animals (Toro & Trobalón, 2005; Hauser, Newport, & Aslin, 2001) across a wide range of stimuli, modalities (Conway & Christiansen, 2005; Fiser & Aslin, 2001; Saffran, Johnson, Aslin, & Newport, 1999), and statistical structures (Schapiro, Rogers, Cordova, Turk-Browne, & Botvinick, 2013; Hunt & Aslin, 2001; Aslin, Saffran, & Newport, 1998).

Although statistical learning has been established as a core learning mechanism with rigorous behavioral work, much is unknown about how this learning takes place in the human brain, and especially in children's developing brains. Moreover, little is known about how to situate the neuroscience of statistical learning with regard to other known learning and memory systems. The present work therefore directly relates statistical learning ability to children's brain structure, focusing on neural regions that are implicated in the broader neuroscience of learning and memory. Before describing this work, we briefly review previous neuroimaging work in statistical learning in adults and children with reference to related work in learning and memory.

Statistical Learning in the Adult Brain

In adult neuroimaging studies of statistical learning, the left inferior frontal and superior temporal cortical regions have been implicated along with the hippocampus and the basal ganglia. In the auditory domain, statistical learning was linked with greater recruitment of the left inferior frontal cortex, specifically the pars triangularis and pars opercularis (Karuza et al., 2013). Likewise in the visual domain, the left inferior frontal cortex was shown to be recruited to a greater extent during exposure to structured versus random blocks and learning performance on a posttest positively correlated with greater recruitment in the left inferior frontal cortex (Turk-Browne, Scholl, Chun, & Johnson, 2009). More recent work in the visual modality further shows that the learning of a different kind of statistic (temporal community structure instead of co-occurrence) was associated with repetition enhancement in bilateral inferior frontal cortex (Schapiro et al., 2013). The role of the inferior frontal cortex, especially the left inferior frontal cortex, in statistical learning is often interpreted with regard to the role this region plays in language production and processing (Poeppel & Hickok, 2004). This interpretation is more straightforward for the studies that investigate learning in the auditory domain, especially when the base stimuli are syllables. A more general role of this region in the learning of temporal sequences can, however, be put forth to explain a role of this region in supporting the learning of visual sequences. Also noteworthy is the large body of work that implicates the inferior frontal cortex in active maintenance (Fedorenko, Behr, & Kanwisher, 2011; Fiebach, Rissman, & D'Esposito, 2006) and learning and memory more broadly construed (Wagner et al., 1998).

In addition to the inferior frontal cortex, the lateral and medial-temporal cortices have been linked to statistical learning. The left temporal cortex has been shown to be more active for structured versus nonstructured auditory stimuli (McNealy, Mazziotta, & Dapretto, 2006). Similarly, the temporal gyrus, bilaterally, was active alongside premotor regions for structured auditory stimuli in an fMRI study of statistical learning (Cunillera et al., 2009). The right superior and middle temporal gyri were also differentially recruited during exposure to structured versus random blocks in the visual domain (Turk-Browne et al., 2009). The role of this superior temporal cortex, especially the left, has been discussed with regard to its known role in language (Finn, Hudson Kam, Ettlinger, Vytlacil, & D'Esposito, 2013; Poeppel & Hickok, 2004) and auditory processing. Again, this interpretation is more straightforward for the statistical learning of auditory stimuli but can be applied to the visual paradigms because these regions are likely to play a more general role in the learning of rapid sequences, something that is more commonly done in the auditory domain.

Recent work has shown that the medial-temporal lobe, generally, and the hippocampus, specifically, is important for statistical learning. Indeed, neural responses to stimuli with a high co-occurrence structure become increasingly similar in the hippocampus, perirhinal cortex, and parahippocampal cortex (Schapiro, Kustner, & Turk-Browne, 2012). Likewise, the hippocampus (along with the regions noted above) shows a more similar activation profile for items from the same versus different statistically defined communities (Schapiro, Turk-Browne, Norman, & Botvinick, 2016; Schapiro et al., 2013). In addition, patients with hippocampal lesions show limited statistical learning ability (Covington, Brown-Schmidt, & Duff, 2018; Schapiro, Gregory, Landau, McCloskey, & Turk-Browne, 2014). This work demonstrates that some forms of statistical learning can take place in patients with hippocampal lesions but that this is reduced relative to controls, suggesting that the hippocampus can support statistical learning but that it may not be necessary. Based largely on this work, it has been suggested that we think differently about the role of the hippocampus in learning and memory, that we think of the computational properties of the hippocampus—namely its ability to perform rapid binding (Cohen & Eichenbaum, 1993)—instead of whether the representation can be verbalized or is declarative (Shohamy & Turk-Browne, 2013).

Finally, the basal ganglia have also been shown to be involved in statistical learning. In an ROI analysis, Karuza et al. (2013) identified learning-related activity in the basal ganglia (caudate and putamen) during an auditory statistical learning task. Likewise, differential activity in the caudate was observed during exposure to structured versus random blocks in a visual statistical learning task (Turk-Browne et al., 2009). The role of the basal ganglia in statistical learning fits nicely with previous work demonstrating that the basal ganglia, especially the caudate, are involved in associative motor learning (Gabrieli, 1998; Cohen & Squire, 1980) and the (implicit) learning of probabilistic relationships (Shohamy et al., 2004; Poldrack et al., 2001).

Taken together, studies have demonstrated the involvement of the inferior frontal and superior temporal cortical regions, the hippocampus, and the basal ganglia in adult statistical learning. Importantly, there is a consistent lateralization of these effects. Cortically, the prefrontal and temporal regions are most consistently implicated in the left hemisphere, especially for auditory paradigms. The caudate and hippocampus are most consistently observed on the right (Turk-Browne et al., 2009), but these subcortical findings tend to be more bilateral.

Statistical Learning in the Child Brain

To date, only two studies have explored the neural correlates of this ability in children, one functional and one structural. The functional study found increased recruitment of temporal cortices and greater activity in the left inferior frontal cortex for visually presented statistically regular relative to random items (McNealy, Mazziotta, & Dapretto, 2010). Structurally, the volume of the hippocampus has been shown to predict statistical learning performance across both children and adults. In particular, performance on a visual statistical learning task was associated with smaller hippocampi when controlling for age, sex, and IQ (Schlichting, Guarino, Schapiro, Turk-Browne, & Preston, 2017). This finding was bilateral and specific to the head of the hippocampus and two subregions (the subiculum and CA2/3). Of note, smaller volumes were predictive of better learning in this study, something that could reflect greater pruning of redundant connections. Taken together, this work suggests a similar suite of regions that are implicated in statistical learning across development, especially left inferior frontal regions and the hippocampus. However, as yet, it is largely unknown how structural differences outside the hippocampus relate to success in statistical learning during childhood or if these associations are also observed in auditory/linguistic statistical learning, despite initial work in statistical learning focusing on the segmentation of this kind of stimuli for the purposes of language learning in infancy.

Statistical Learning, Memory, and Development

From a classic neuroscience of memory perspective, this suite of regions—inferior frontal, hippocampus, and basal ganglia (caudate)—does not have just one clear mnemonic home, spanning the traditional declarative (verbalizable or “knowing what”) and nondeclarative/procedural (nonverbalizable, “knowing how”) divide: Prefrontal and hippocampal regions are known to support declarative systems, whereas the basal ganglia (and other cortical regions) are associated with nondeclarative/procedural memory (Gabrieli, 1998; Cohen & Squire, 1980).

Indeed, statistical learning itself appears to span this classic declarative/procedural divide, at least in adults. Recent work has shown that performing well on the classic behavioral index of statistical learning—a forced-choice recognition test—is associated with adults' confidence about their recognition (Batterink, Reber, Neville, & Paller, 2015). Participants are thus aware of when they are able to demonstrate better learning, suggesting that declarative knowledge emerges alongside nondeclarative and procedural representations. The idea that multiple mnemonic processes are likely at play during statistical learning has been echoed elsewhere (Thiessen, 2017; Bays, Turk-Browne, & Seitz, 2016; Shohamy & Turk-Browne, 2013).

From the perspective of development especially, there are many unanswered questions about the neurobiology of statistical learning and how to think of statistical learning in the broader context of the neuroscience of memory. As stated, children and adults both show robust (and fast) statistical learning (Saffran et al., 1997). Yet, we know that many declarative functions (e.g., working memory, long-term memory) are slow to develop and are far from mature in early childhood (ages 6–8 years) and even into adolescence (Finn, Sheridan, Hudson Kam, Hinshaw, & D'Esposito, 2010; Ghetti & Angelini, 2008; Gathercole, Pickering, Ambridge, & Wearing, 2004). Furthermore, development of declarative and nondeclarative systems is asynchronous (Finn et al., 2016). Prefrontal regions (including inferior frontal cortex) and the hippocampus undergo great change, especially through early childhood (Ducharme et al., 2016; Lee, Ekstrom, & Ghetti, 2014; Sheridan, Kharitonova, Martin, Chatterjee, & Gabrieli, 2014; Kharitonova, Martin, Gabrieli, & Sheridan, 2013), whereas procedural or nondeclarative abilities appear to mature earlier (Finn et al., 2016; Amso & Davidow, 2012). Among children, will better learning be more related to the basal ganglia (which are implicated in procedural or implicit learning systems), or will better learning be associated with the same suite of implicit and explicit memory systems that we have observed in adulthood?

The Current Investigation

The current investigation aims to explore the relationship between statistical learning and the thickness and volume of structures across the brain that have been consistently implicated in statistical learning and memory, focusing specifically on the left inferior PFC (pars triangularis and opercularis), the hippocampus, and the caudate. We target a tight age range during early childhood to identify how individual differences in statistical learning ability (and not age per se) relate to neural structure during childhood. This approach allows us to answer the puzzle posed above—how is it that we observe early and robust statistical learning and substantial developmental change in the neural systems thought to underlie statistical learning. Specifically, we focus on early childhood (5–8 years of age) because this is a period during which the primary neural substrates of interest are undergoing great functional and structural change. Age-related change during this period has been observed in the inferior frontal gyrus (Kharitonova et al., 2013; Bunge & Zelazo, 2006), the hippocampus (Lee et al., 2014), and the basal ganglia (Østby et al., 2009). Yet previous work has shown no difference in statistical learning outcomes between this age of children and adults (Saffran et al., 1997).

METHODS

Participants

Fifty-nine children (mean age = 6.9 years, range = 5.3–8.5 years, SD = 0.711, 38 girls; Figure 1A) participated. Written consent was obtained; all participants were English-speaking with normal hearing and no reported history of neurological, psychiatric, or learning disorders. Participants' families received $50 per visit for their participation, which involved acquisition of several behavioral tasks and functional and structural neuroimaging. Data from other parts of this study have been reported elsewhere (Kharitonova, Winter, & Sheridan, 2015).

Figure 1. 

Age and behavioral performance. (A) Density plot of the age distribution of our sample in months. (B) A violin plot (pink) of performance depicts the minimum (bottom of shape) and maximum (top of shape) observed values. Black circles indicate each child's mean percent correct, and the width of the violin plot indicates the probability density of the value on the corresponding y axis across the group. Chance performance is indicated with the dotted line. (C) Scatter plot depicting the relationship between performance and age. The pink line depicts the linear model (predicting percent correct from age), and the opaque width of the line represents the 95% confidence level interval of this model.

Figure 1. 

Age and behavioral performance. (A) Density plot of the age distribution of our sample in months. (B) A violin plot (pink) of performance depicts the minimum (bottom of shape) and maximum (top of shape) observed values. Black circles indicate each child's mean percent correct, and the width of the violin plot indicates the probability density of the value on the corresponding y axis across the group. Chance performance is indicated with the dotted line. (C) Scatter plot depicting the relationship between performance and age. The pink line depicts the linear model (predicting percent correct from age), and the opaque width of the line represents the 95% confidence level interval of this model.

Procedure

Children came to the Center for Brain Imaging at Harvard University for two visits, a mock scanning session and a scanning session, which took place within 1 week of each other and never on the same day. A battery of behavioral measures were collected from children during the first visit after the mock scan, which included subscales to measure IQ and statistical learning measures (described in detail below). Tasks were always presented in the same order; the statistical learning task was presented last. All procedures were approved by the institutional review board at Harvard University and Boston Children's Hospital.

Apparatus and Stimuli

IQ

Each child completed the Matrix Reasoning subscale of the Wechsler Preschool and Primary Scale of Intelligence to estimate IQ. Standard scores were calculated and are used in subsequent analyses.

Statistical Learning Stimuli

Participants were exposed to an artificial language over headphones before being asked questions about the language. After meeting a stuffed alien named Arnie, children were told that they were going to listen to Arnie's home language to help him figure out some words (he bumped his head and forgot a lot of them). The language was similar to those used in previous auditory statistical learning studies (Finn & Hudson Kam, 2008; Saffran et al., 1997; Saffran, Newport, et al., 1996); the stimuli played for 4 min 47.5 sec and consisted of four trisyllabic words made from eight consonants and four vowels, which generated 12 unique syllables. The four words were /tupiɹo/, /golɐbu/, /bidɐku/, and /pɐkoli/. Each word was generated with the text-to-speech program SoftVoice (Katz, 2005). The synthesizer produced syllables with a monotonic F0 (fundamental frequency) of 190 Hz. All vowels were matched for length (250 msec). Words were presented at 45 times each. Presentation order was quasirandom with no pauses and no immediate repetitions. TPs (the probability of XY given X) were therefore deterministic (1.0) for syllable transitions that were word internal and .33 at word boundaries. A sample stream is here: /tupiɹobidɐkupɐkoligolɐbupɐkolitupiɹobidɐku/

Statistical Learning Tests

After exposure, children were asked to complete an auditory two-alternative forced-choice word segmentation test. Children first completed a practice trial in English and were then asked which of two options (a word or a foil; separated by a 700-msec pause) was a better word in Arnie's language, indicating their choice verbally to the experimenter. E-Prime software was used (Schneider, Eschman, & Zuccolotto, 2002) to present 12 test items in one of two counterbalanced orders. There were three types of test foils: nonword, classic-part-word, and fronted-part-word. Nonword foils combined syllables with a TP of 0 at each transition but preserved the relative placement of syllables (e.g., the syllable /go/ in the word /golɐbu/ from exposure would always be in the first position, likewise for middle and final position syllables). An example nonword foil is /godɐli/. A classic-part-word foil started with the middle and last syllables of a word (e.g., /piɹo/ from the word /tupiɹo/) and ended with the first syllable of another word (e.g., /piɹobi/). Unlike nonwords, part words occur during exposure but contain a TP dip (.33) between the second and third syllables. Finally, a fronted-part-word foil also started with the middle and last syllables of a word (e.g., /piɹo/ from the word /tupiɹo/), but these ended with the first syllable of that same word (e.g., /piɹotu/). The TP between the second and third syllables is therefore 0, never having happened before, but these test items contain all of the same elements as a word (just in the wrong order).

MR Data Acquisition

All MRIs were acquired at the Center for Brain Science at Harvard University on a 3-T Siemens Tim Trio MRI system (Berlin, Germany). T1-weighted anatomical scans were acquired with a multiecho MPRAGE sequence that is optimized for pediatric work (repetition time = 2530 msec, echo time = 1640–7040 msec, flip angle = 7°, field of view = 220 mm2, 176 slices, in-plane voxel size = 1 mm3; integrated parallel acquisition techniques = 3 to reduce acquisition time). To reduce motion artifacts, a navigator echo was used before the onset of scan acquisition, slices were compared with this echo online, and up to 20% of slices that did not align with the navigator echo were reacquired (Tisdall et al., 2012).

Behavioral Data Analyses

Analyses were conducted in R (R Core Team, 2012). One-sample t tests were used to compare performance to chance. Repeated-measures ANOVAs were used to analyze omnibus effects pertaining to test type. A generalized mixed-effects model (glmer function in the lme4 package; Bates, Mächler, Bolker, & Walker, 2015) was used to assess whether performance improved with item repetition during the test. This model contained random intercepts and random slopes for item repetition, grouped by subject, and item repetition as a fixed effect (see the Supplemental Material, available at https://finnlandlab.org/wp-content/uploads/2018/11/supp_info_jocn_2018_final.pdf, for exact model specification in R). Finally, a Spearman's rank-order correlation was performed to investigate age-related changes in statistical learning ability.

MR Data Analyses

Analyses were completed using FreeSurfer Version 5.0 (Fischl et al., 2004; Fischl & Dale, 2000) and follow the same protocol as in previous work (Kharitonova et al., 2015). Data were preprocessed to correct for motion; remove nonbrain tissue using a hybrid watershed/surface deformation procedure (Ségonne et al., 2004); transform into Talairach space using an automated technique that has been validated in children (Burgund et al., 2002); and finally segment subcortical white matter, deep gray matter volumetric structures, and cortical gray matter using gyral and sulcal landmarks. FreeSurfer morphometric procedures and those used in the present analyses have been successfully used in multiple developmental studies including children as young as 4 years (Mackey et al., 2015; Kharitonova et al., 2013; Ghosh et al., 2010). Using FreeSurfer, we defined 148 cortical regions (74 for each hemisphere) according to the 2005 Desikan–Killiany Atlas (Desikan et al., 2006) within each participant.

Following segmentation, data were visually inspected for image quality by a trained research assistant who placed “control points” to more accurately define white and gray matter segmentation, removed gray matter voxels from the white matter mask, and removed skull parts, which were inadvertently not removed during brain mask generation to better define pial and white matter boundaries. Following these hand editing steps, data were again run through steps, which segmented and measured cortical gray matter using gyral and sulcal landmarks. This editing and reanalysis were performed up to three times on each participant. These are the recommended procedures from the FreeSurfer development team and have been successfully used in our lab previously when measuring cortical and subcortical structure in children (Kharitonova et al., 2013). Our dependent measure for subcortical regions is volume. For cortical regions—which can be characterized by overall volume, thickness, and surface area (total volume = thickness + surface area)—we chose cortical thickness as our dependent measure because thickness is thought to best reflect the number of cells in a given region (Rakic, 1988) and is maximally comparable with recent developmental work (Ducharme et al., 2016). Overall, these methods appeared to estimate subcortical volume and cortical thickness within typical ranges; however, one participant had estimated the left hippocampal volume to be greater than 3 standard deviations from the group mean; this value was excluded from subsequent analyses. We did not observe any other outliers above or below the group mean for any other ROIs.

Analyses were restricted to three regions previously associated with statistical learning: the ventral PFC, the caudate, and the hippocampus. Specifically, we measured thickness in the lefts pars opercularis and left pars triangularis (both inferior frontal gyrus) and volume in the left and right hippocampus and caudate in each individual's native space using the 2005 Desikan–Killiany Atlas (Desikan et al., 2006). Relationships to performance were investigated in each region using independent regression models (as detailed below).

Structure–Performance Relationships

Multiple linear regression analyses were performed to uncover the relationship between thickness and volume measures and each child's mean performance on the statistical learning task. All models controlled for age, sex, IQ, and total intracranial volume. We chose to control for IQ to isolate relationships that were particular to statistical learning performance and not differences in intelligence and cognitive ability more broadly speaking. Because some children's performance was below chance (50%) on the statistical learning task, additional analyses were run in which below chance performance was recoded as chance (50%; n = 9 children). This alternate approach ensured that associations observed in the primary analysis were not driven by below chance performance.

Structure–Age Relationships

Multiple linear regression analyses were performed to uncover the relationship between thickness and volume measures and age. All models controlled for sex and total intracranial volume.

RESULTS

Behavior

Children learned the statistical regularities presented during training; overall performance was significantly better than chance (50%, one-sample t test: t(59) = 4.67, p < .001, d = .61; Figure 1B). Moreover, word repetition did not impact performance during test (β = 0.106, z = 1.07, p = .285), suggesting that learning did not occur during the test itself. To identify if this learning differed across foil subtypes (i.e., nonword, classic-part-word, and fronted-part-word), a repeated-measures ANOVA with Performance on the three foil subtypes as a within-subject factor was performed. This analysis revealed no difference in learning on foil subtype, F(2, 116) = 1.42, p = .246, ηp2 = .013. Because foil subtypes were not treated differently, we consider them together for all subsequent analyses. Furthermore, age did not predict statistical learning performance (Figure 1C; Spearman's rho: ρ = −.031, p = .815), but statistical learning performance was positively associated with IQ (r = .26, p = .045).

Performance: Thickness, Inferior Frontal Gyrus

As shown in Figure 2A, left pars triangularis thickness was positively associated with performance on the statistical learning test (β = 0.36, p = .008, Bonferroni-corrected α = .008). Likewise, left pars opercularis thickness was also positively associated with performance on the statistical learning test (β = 0.33, p = .016, Bonferroni-corrected α = .008; Figure 2B).

Figure 2. 

Left prefrontal thickness and performance. (A) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left pars triangularis thickness (x axis). The green line depicts the linear model, and the opaque width of the line represents the 95% confidence interval. (B) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left pars opercularis thickness (x axis). The blue line depicts the linear model, and the opaque width of the line represents the 95% confidence interval.

Figure 2. 

Left prefrontal thickness and performance. (A) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left pars triangularis thickness (x axis). The green line depicts the linear model, and the opaque width of the line represents the 95% confidence interval. (B) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left pars opercularis thickness (x axis). The blue line depicts the linear model, and the opaque width of the line represents the 95% confidence interval.

Similar relationships were found when recoding below chance performance as chance in nine children: Thickness was positively associated with performance in the left pars triangualarus (β = 0.27, p = .0142) and marginally so in the left pars opercularis (β = 0.213, p = .112; Supplementary Figure 1A and B).

Performance: Volume, Hippocampus, and Caudate

As shown in Figure 3A, the left hippocampus was not associated with performance (β = 0.062, p = .683); however, the right hippocampus was: Smaller volumes were associated with better performance (β = −0.35, p = .016, Bonferroni-corrected α = .008; Figure 3B). Smaller right hippocampal volumes were likewise observed when recoding below chance performance as chance in nine children (β = −0.322, p = .0213; Supplementary Figure 1C). Post hoc analyses of hippocampal subregions were further carried out. As reported in the supplement, none were significant (all ps > .09).

Figure 3. 

Hippocampus, caudate, and performance. (A) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left hippocampus volume (x axis). This and all other lines represented the depicted linear model. Likewise, this and all other opaque widths represent the 95% confidence interval. (B) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted right hippocampus volume (x axis). (C) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left caudate volume (x axis). (D) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted right caudate volume (x axis).

Figure 3. 

Hippocampus, caudate, and performance. (A) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left hippocampus volume (x axis). This and all other lines represented the depicted linear model. Likewise, this and all other opaque widths represent the 95% confidence interval. (B) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted right hippocampus volume (x axis). (C) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted left caudate volume (x axis). (D) Scatter plot depicting the relationship between adjusted performance (y axis) and adjusted right caudate volume (x axis).

As shown in Figure 3C and D, caudate volume was not associated with performance on statistical learning, either on the left (β = 0.04, p = .771) or the right (β = 0.03, p = .861).

Age

As shown in Figure 4A, age was not associated with left pars triangularis thickness in this relatively tight early childhood age range (β = 0.005, p = .969). This was also true in the left pars opercularis (β = 0.046, p = .732; Figure 4B).

Figure 4. 

Age and structure. (A) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left pars triangularis thickness (x axis). This and all other lines represented the depicted linear model. Likewise, this and all other opaque line widths represent the 95% confidence interval. (B) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left pars opercularis thickness (x axis). (C) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left hippocampus volume (x axis). (D) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted right hippocampus volume (x axis). (E) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left caudate volume (x axis). (F) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted right caudate volume (x axis).

Figure 4. 

Age and structure. (A) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left pars triangularis thickness (x axis). This and all other lines represented the depicted linear model. Likewise, this and all other opaque line widths represent the 95% confidence interval. (B) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left pars opercularis thickness (x axis). (C) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left hippocampus volume (x axis). (D) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted right hippocampus volume (x axis). (E) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted left caudate volume (x axis). (F) Scatter plot depicting the relationship between adjusted age (y axis) and adjusted right caudate volume (x axis).

Age was associated with neither left hippocampal volume (β = 0.236, p =.057; Figure 4C) nor right hippocampal volume (β = 0.143, p = .255; Figure 4D). Finally, age was neither associated with left caudate volume (β = 0.070, p =.592; Figure 4E) nor right caudate volume (β = 0.021, p =.869; Figure 4F).

Interactions: Age, Structure, and Performance

Age did not interact with structure to predict performance in most cases: age and left pars triangularis thickness (β = −1.940, p = .486; Figure 5A), age and left pars opercularis thickness (β = −4.932, p = .204; Figure 5B), age and left hippocampal volume (β = −3.823, p = .070; Figure 5C), age and left caudate volume (β = −3.866, p = .070), and age and right caudate volume (β = −2.933, p = .174); thus, in most ROIs, performance relationships do not differ across the reported age range. Age and right hippocampal volume did interact (β = −4.692, p = .044; Figure 5D), such that the association between hippocampal volume and task performance was stronger for older relative to younger children. Age also interacted with right hippocampal volume to predict performance when recoding below chance performance as chance in nine children (β = −4.62, p = .044), such that the association between hippocampal volume and task performance was stronger for older relative to younger children (Supplementary Figure 1D).

Figure 5. 

Interactions: Age, structure, and performance. Scatter plots depicting the relationship between adjusted performance (y axis) and adjusted volume (hippocampus) or cortical thickness (x axis) with age of participant indicated by the color of the hexagon. (D) The only significant interaction is depicted in the right hippocampus. Colored lines depict the regression between volume and performance separately for younger (purple) and older (orange) children as determined by a median split (at 83 months).

Figure 5. 

Interactions: Age, structure, and performance. Scatter plots depicting the relationship between adjusted performance (y axis) and adjusted volume (hippocampus) or cortical thickness (x axis) with age of participant indicated by the color of the hexagon. (D) The only significant interaction is depicted in the right hippocampus. Colored lines depict the regression between volume and performance separately for younger (purple) and older (orange) children as determined by a median split (at 83 months).

DISCUSSION

This study shows that the thickness of the left inferior frontal cortex and volume of the right hippocampus predict statistical learning ability in young children. Importantly, these regions did not change in thickness or volume with age (nor did the other ROIs), likely because of the tight age range we measured (5–8.5 years). Interestingly, the relationship between learning and the right hippocampus did interact with age, such that older children with relatively smaller hippocampi performed especially well. These data are largely consistent with the two previous investigations of how the child brain relates to statistical learning, with the first showing involvement of the left prefrontal regions in 10-year-old children (McNealy et al., 2010) and the second showing that smaller hippocampal heads are positively related to statistical learning in children (ages 6–18 years; Schlichting et al., 2017). We discuss these findings by neural region and with regard to the broader learning, memory, and developmental literatures.

Inferior Frontal Cortex

The thickness of children's left inferior frontal cortices predicts auditory statistical learning ability. This discovery suggests that frontal cortex is involved in statistical learning during childhood. The direction of this finding is important: Thicker cortex was associated with better learning after controlling for total intracranial volume, IQ, sex, and age. The left inferior frontal regions have shown a similar positive association with performance (on a measure of verbal IQ) in children aged 5–11 years (Sowell et al., 2004). Still, a thicker inferior frontal cortex during childhood could reflect two things. It could be that the inferior frontal cortex is slower to prune relative to the rest of the brain and is therefore thicker because it is taking longer to identify which connections perform prefrontal computations best (see Shaw et al., 2006, for a related argument). It could also mean that the inferior frontal cortex is getting thicker with practice and thus increasing ability, a pattern typical of increasing experience in adults. We cannot distinguish among these possibilities. However, because substantial evidence indicates that age is associated with thinning in the PFC (Lenroot & Giedd, 2006), this observed pattern of thicker associated with better performance is not simply a case of developmentally “older” individuals performing better on this task.

This finding highlights the importance of understanding how and why the inferior frontal cortex relates to statistical learning. One potential explanation would be that the left inferior frontal cortex is observed in studies using linguistic stimuli for statistical learning because this region is specialized for linguistic information, consistent with the known role of the inferior frontal gyrus in language production. However, previous studies have conclusively demonstrated that the involvement of the inferior frontal cortex in statistical learning is not sensory modality specific; involvement of this region is observed in visual paradigms (Schapiro et al., 2013; Turk-Browne & Scholl, 2009) in addition to linguistic paradigms, thus ruling out a purely linguistic explanation. But if the involvement of inferior frontal regions does not have to do with the importance of this region for linguistic processing, then what? An intriguing possibility could have to do with the importance of inferior frontal cortex in declarative memory processes and attentionally taxing aspects of working memory (Bahlmann, Blumenfeld, & D'Esposito, 2015; Blumenfeld, Nomura, Gratton, & D'Esposito, 2013; Fiebach et al., 2006). This region could be involved in statistical learning because it is needed to effortfully hold the fleeting sequential information in working memory, that is, in the mind's ear or eye. Indeed, although statistical learning is typically discussed as an automatic or implicit process (Saffran et al., 1997), attention and declarative processes appear to play an important role, at least in adults (Bays et al., 2016; Batterink et al., 2015; Finn, Lee, Kraus, & Hudson Kam, 2014; Fernandes, Kolinsky, & Ventura, 2010; Toro, Sinnett, & Soto-Faraco, 2005; Turk-Browne et al., 2005). Future functional work should determine how statistical learning relates to these attentional and mnemonic systems more directly and especially during childhood, a time during which these systems undergo great change.

Hippocampus

Similar to previous work (Schlichting et al., 2017), we also found that smaller hippocampal volume was predictive of learning. Unlike this work, our finding was restricted to the right hemisphere. And unlike our findings in PFC, this relationship is negative. Although MRI cannot observe these processes directly and a longitudinal sample is needed to observe changes, better performance associated with smaller volumes could be due to increased pruning of superfluous connections (Cowan, Fawcett, O'Leary, & Stanfield, 1984), a process that could improve performance, in general, and processing speed, in particular (Chechik, Meilijson, & Ruppin, 1998).

Interestingly, the relationship between statistical learning and the right hippocampus interacted with age, with older children who have smaller hippocampi performing especially well. The direction of this relationship suggests that individual differences in hippocampal volume matter more for children as they get older. This may be consistent with interpreting the main effect as reflecting hippocampal maturation. However, several limitations increase the caution with which we interpret these effects. First, the observed interaction represents a moderate effect (p = .044), which does not survive Bonferroni correction (alpha of .008). Second, although other studies have observed associations between hippocampal volume and statistical learning, this is the first report of an age by volume interaction. Thus, follow-up studies are needed to observe whether this pattern replicates.

Much recent attention has been paid to the involvement of the hippocampus in statistical learning and reframing the role of the hippocampus as important for more than just declarative memory process (Schapiro, Turk-Browne, Botvinick, & Norman, 2017; Schlichting et al., 2017; Shohamy & Turk-Browne, 2013). As noted above, it is also possible that statistical learning is more declarative than we originally thought (Batterink et al., 2015), and the relationship between hippocampal volume and performance may reflect the involvement of declarative memory systems in statistical learning. Finally, the present hippocampal finding extends previous literature by showing that the involvement of the hippocampus extends to statistical learning in the auditory, not just the visual, modality during early childhood.

Caudate

Finally, we did not find that differences in the volume of the caudate, either on the left or the right, predicted learning in young children. This is counter to functional imaging studies in adulthood (Karuza et al., 2013). In addition, given that the caudate is most often implicated in implicit or procedural memory (Shohamy, 2011; Gabrieli, 1998) and that procedural memory develops early relative to other forms of memory (Finn et al., 2016), we might have expected differences in these structures to map onto differences in learning outcomes. Although this null result must be interpreted with caution, given our sample size, the fact that we observed associations between performance and structure in both the inferior frontal gyrus and the hippocampus, but not the caudate, should motivate further research into the role of the striatum in statistical learning in early childhood.

Statistical Learning and Development

A more comprehensive picture of how statistical learning is achieved in childhood and infancy in terms of neural and cognitive mechanisms is greatly needed. Here, we observed individual differences in statistical learning to only be associated with neural structures traditionally linked with declarative mnemonic processes, the inferior frontal cortex and hippocampus. Given that declarative processes are slower to develop than procedural (Finn et al., 2016), this finding is quite remarkable and is consistent with a recent emphasis on the link between declarative mnemonic processes and statistical learning (Bays et al., 2016; Batterink et al., 2015; Schapiro et al., 2014). Indeed, developmental work further exploring this line of research not only will serve to tell us the practical question about how children learn but also will likely yield more information about how to situate statistical learning in a broader neural and mnemonic framework.

Conclusions

The present data show that statistical learning is related to hippocampal and left inferior prefrontal structure during childhood. Thus, in childhood, statistical learning is supported by multiple neural structures that are more broadly implicated in learning and memory systems, especially declarative memory (hippocampus) and attention and top–down control (the PFC). This work supports the idea that statistical learning is achieved with domain general learning systems that are implicated in other, declarative and attentionally mediated top–down—forms of learning and memory.

Acknowledgments

This work was funded by the National Institutes of Mental Health (K01MH09255 to M. A. S.) and in part by the National Institutes of Health (National Research Service Award to A. S. F.). We thank John Gabrieli, Warren Winter, and Jenna Snyder for their support, thoughtful discussion, and assistance in data collection.

Reprint requests should be sent to Amy S. Finn, Department of Psychology, University of Toronto, 100 St. George Street, 4th floor, Sidney Smith Hall, Toronto, ON M5S 3G3, Canada, or via e-mail: finn@psych.utoronto.ca.

REFERENCES

REFERENCES
Amso
,
D.
, &
Davidow
,
J. Y.
(
2012
).
The development of implicit learning from infancy to adulthood: Item frequencies, relations, and cognitive flexibility
.
Developmental Psychobiology
,
54
,
664
673
.
Aslin
,
R. N.
,
Saffran
,
J. R.
, &
Newport
,
E. L.
(
1998
).
Computation of conditional probability statistics by 8-month-old infants
.
Psychological Science
,
9
,
321
324
.
Bahlmann
,
J.
,
Blumenfeld
,
R. S.
, &
D'Esposito
,
M.
(
2015
).
The rostro-caudal axis of frontal cortex is sensitive to the domain of stimulus information
.
Cerebral Cortex
,
25
,
1815
1826
.
Bates
,
D.
,
Mächler
,
M.
,
Bolker
,
B. M.
, &
Walker
,
S. C.
(
2015
).
Fitting linear mixed-effects models using lme4
.
Journal of Statistical Software
,
67
,
1
48
.
Batterink
,
L. J.
,
Reber
,
P. J.
,
Neville
,
H. J.
, &
Paller
,
K. A.
(
2015
).
Implicit and explicit contributions to statistical learning
.
Journal of Memory and Language
,
83
,
62
78
.
Bays
,
B. C.
,
Turk-Browne
,
N. B.
, &
Seitz
,
A. R.
(
2016
).
Dissociable behavioural outcomes of visual statistical learning
.
Visual Cognition
,
23
,
1072
1097
.
Blumenfeld
,
R. S.
,
Nomura
,
E. M.
,
Gratton
,
C.
, &
D'Esposito
,
M.
(
2013
).
Lateral prefrontal cortex is organized into parallel dorsal and ventral streams along the rostro-caudal axis
.
Cereberal Cortex
,
23
,
2457
2466
.
Bunge
,
S. A.
, &
Zelazo
,
P. D.
(
2006
).
A brain-based account of the development of rule use in childhood
.
Current Directions in Psychological Science
,
15
,
118
121
.
Burgund
,
E. D.
,
Kang
,
H. C.
,
Kelly
,
J. E.
,
Buckner
,
R. L.
,
Snyder
,
A. Z.
,
Petersen
,
S. E.
, et al
(
2002
).
The feasibility of a common stereotactic space for children and adults in fMRI studies of development
.
Neuroimage
,
17
,
184
200
.
Chechik
,
G.
,
Meilijson
,
I.
, &
Ruppin
,
E.
(
1998
).
Synaptic pruning in development: A computational account
.
Neural Computation
,
10
,
1759
1777
.
Cohen
,
N. J.
, &
Eichenbaum
,
H.
(
1993
).
Memory, amnesia, and the hippocampal system
.
Cambridge, MA
:
MIT Press
.
Cohen
,
N. J.
, &
Squire
,
L. R.
(
1980
).
Preserved learning and retention of pattern-analyzing skill in amnesia: Dissociation of knowing how and knowing that
.
Science
,
210
,
207
210
.
Conway
,
C. M.
, &
Christiansen
,
M. H.
(
2005
).
Modality-constrained statistical learning of tactile, visual, and auditory sequences
.
Journal of Experimental Psychology: Learning, Memory, and Cognition
,
31
,
24
39
.
Covington
,
N. V.
,
Brown-Schmidt
,
S.
, &
Duff
,
M. C.
(
2018
).
The necessity of the hippocampus for statistical learning
.
Journal of Cognitive Neuroscience
,
30
,
680
697
.
Cowan
,
W. M.
,
Fawcett
,
J. W.
,
O'Leary
,
D. D.
, &
Stanfield
,
B. B.
(
1984
).
Regressive events in neurogenesis
.
Science
,
225
,
1258
1265
.
Cunillera
,
T.
,
Càmara
,
E.
,
Toro
,
J. M.
,
Marco-Pallares
,
J.
,
Sebastián-Galles
,
N.
,
Ortiz
,
H.
, et al
(
2009
).
Time course and functional neuroanatomy of speech segmentation in adults
.
Neuroimage
,
48
,
541
553
.
Desikan
,
R. S.
,
Ségonne
,
F.
,
Fischl
,
B.
,
Quinn
,
B. T.
,
Dickerson
,
B. C.
,
Blacker
,
D.
, et al
(
2006
).
An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest
.
Neuroimage
,
31
,
968
980
.
Ducharme
,
S.
,
Albaugh
,
M. D.
,
Nguyen
,
T.-V.
,
Hudziak
,
J. J.
,
Mateos-Pérez
,
J. M.
,
Labbe
,
A.
, et al
(
2016
).
Trajectories of cortical thickness maturation in normal brain development—The importance of quality control procedures
.
Neuroimage
,
125
,
267
279
.
Fedorenko
,
E.
,
Behr
,
M. K.
, &
Kanwisher
,
N.
(
2011
).
Functional specificity for high-level linguistic processing in the human brain
.
Proceedings of the National Academy of Sciences, U.S.A.
,
108
,
16428
16433
.
Fernandes
,
T.
,
Kolinsky
,
R.
, &
Ventura
,
P.
(
2010
).
The impact of attention load on the use of statistical information and coarticulation as speech segmentation cues
.
Attention, Perception, & Psychophysics
,
72
,
1522
1532
.
Fiebach
,
C. J.
,
Rissman
,
J.
, &
D'Esposito
,
M.
(
2006
).
Modulation of inferotemporal cortex activation during verbal working memory maintenance
.
Neuron
,
51
,
251
261
.
Finn
,
A. S.
, &
Hudson Kam
,
C. L.
(
2008
).
The curse of knowledge: First language knowledge impairs adult learners' use of novel statistics for word segmentation
.
Cognition
,
108
,
477
499
.
Finn
,
A. S.
,
Hudson Kam
,
C. L.
,
Ettlinger
,
M.
,
Vytlacil
,
J.
, &
D'Esposito
,
M.
(
2013
).
Learning language with the wrong neural scaffolding: The cost of neural commitment to sounds
.
Frontiers in Systems Neuroscience
,
7
,
85
.
Finn
,
A. S.
,
Kalra
,
P. B.
,
Goetz
,
C.
,
Leonard
,
J. A.
,
Sheridan
,
M. A.
, &
Gabrieli
,
J. D. E.
(
2016
).
Developmental dissociation between the maturation of procedural memory and declarative memory
.
Journal of Experimental Child Psychology
,
142
,
212
220
.
Finn
,
A. S.
,
Lee
,
T.
,
Kraus
,
A.
, &
Hudson Kam
,
C. L.
(
2014
).
When it hurts (and helps) to try: The role of effort in language learning
.
PLoS One
,
9
,
e101806
.
Finn
,
A. S.
,
Sheridan
,
M. A.
,
Hudson Kam
,
C. L.
,
Hinshaw
,
S.
, &
D'Esposito
,
M.
(
2010
).
Longitudinal evidence for functional specialization of the neural circuit supporting working memory in the human brain
.
Journal of Neuroscience
,
30
,
11062
11067
.
Fischl
,
B.
, &
Dale
,
A. M.
(
2000
).
Measuring the thickness of the human cerebral cortex from magnetic resonance images
.
Proceedings of the National Academy of Sciences, U.S.A.
,
97
,
11050
11055
.
Fischl
,
B.
,
van der Kouwe
,
A.
,
Destrieux
,
C.
,
Halgren
,
E.
,
Ségonne
,
F.
,
Salat
,
D. H.
, et al
(
2004
).
Automatically parcellating the human cerebral cortex
.
Cerebral Cortex
,
14
,
11
22
.
Fiser
,
J.
, &
Aslin
,
R. N.
(
2001
).
Unsupervised statistical learning of higher-order spatial structures from visual scenes
.
Psychological Science
,
12
,
499
504
.
Gabrieli
,
J. D. E.
(
1998
).
Cognitive neuroscience of human memory
.
Annual Review of Psychology
,
49
,
87
115
.
Gathercole
,
S. E.
,
Pickering
,
S. J.
,
Ambridge
,
B.
, &
Wearing
,
H.
(
2004
).
The structure of working memory from 4 to 15 years of age
.
Developmental Psychology
,
40
,
177
190
.
Ghetti
,
S.
, &
Angelini
,
L.
(
2008
).
The development of recollection and familiarity in childhood and adolescence: Evidence from the dual-process signal detection model
.
Child Development
,
79
,
339
358
.
Ghosh
,
S. S.
,
Kakunoori
,
S.
,
Augustinack
,
J.
,
Nieto-Castanon
,
A.
,
Kovelman
,
I.
,
Gaab
,
N.
, et al
(
2010
).
Evaluating the validity of volume-based and surface-based brain image registration for developmental cognitive neuroscience studies in children 4 to 11 years of age
.
Neuroimage
,
53
,
85
93
.
Gomez
,
R. L.
, &
Gerken
,
L.
(
1999
).
Artificial grammar learning by 1-year-olds leads to specific and abstract knowledge
.
Cognition
,
70
,
109
135
.
Hauser
,
M. D.
,
Newport
,
E. L.
, &
Aslin
,
R. N.
(
2001
).
Segmentation of the speech stream in a nonhuman primate: Statistical learning in cotton-top tamarins
.
Cognition
,
78
,
B53
B64
.
Hunt
,
R. H.
, &
Aslin
,
R. N.
(
2001
).
Statistical learning in a serial reaction time task: Access to separable statistical cues by individual learners
.
Journal of Experimental Psychology: General
,
130
,
658
680
.
Karuza
,
E. A.
,
Newport
,
E. L.
,
Aslin
,
R. N.
,
Starling
,
S. J.
,
Tivarus
,
M. E.
, &
Bavelier
,
D.
(
2013
).
The neural correlates of statistical learning in a word segmentation task: An fMRI study
.
Brain and Language
,
127
,
46
54
.
Katz
,
J.
(
2005
).
SoftVoice (Demo Program)
.
Los Angeles
:
SoftVoice, Inc
.
Kharitonova
,
M.
,
Martin
,
R. E.
,
Gabrieli
,
J. D. E.
, &
Sheridan
,
M. A.
(
2013
).
Cortical gray-matter thinning is associated with age-related improvements on executive function tasks
.
Developmental Cognitive Neuroscience
,
6
,
61
71
.
Kharitonova
,
M.
,
Winter
,
W.
, &
Sheridan
,
M. A.
(
2015
).
As working memory grows: A developmental account of neural bases of working memory capacity in 5- to 8-year old children and adults
.
Journal of Cognitive Neuroscience
,
27
,
1775
1788
.
Lee
,
J. K.
,
Ekstrom
,
A. D.
, &
Ghetti
,
S.
(
2014
).
Volume of hippocampal subfields and episodic memory in childhood and adolescence
.
Neuroimage
,
94
,
162
171
.
Lenroot
,
R. K.
, &
Giedd
,
J. N.
(
2006
).
Brain development in children and adolescents: Insights from anatomical magnetic resonance imaging
.
Neuroscience & Biobehavioral Reviews
,
30
,
718
729
.
Liberman
,
A. M.
,
Cooper
,
F. S.
,
Shankweiler
,
D. P.
, &
Studdert-Kennedy
,
M.
(
1967
).
Perception of the speech code
.
Psychological Review
,
74
,
431
461
.
Mackey
,
A. P.
,
Finn
,
A. S.
,
Leonard
,
J. A.
,
Jacoby-Senghor
,
D. S.
,
West
,
M. R.
,
Gabrieli
,
C. F. O.
, et al
(
2015
).
Neuroanatomical correlates of the income-achievement gap
.
Psychological Science
,
26
,
925
933
.
McNealy
,
K.
,
Mazziotta
,
J. C.
, &
Dapretto
,
M.
(
2006
).
Cracking the language code: Neural mechanisms underlying speech parsing
.
Journal of Neuroscience
,
26
,
7629
7639
.
McNealy
,
K.
,
Mazziotta
,
J. C.
, &
Dapretto
,
M.
(
2010
).
The neural basis of speech parsing in children and adults
.
Developmental Science
,
13
,
385
406
.
Nissen
,
M. J.
, &
Bullemer
,
P.
(
1987
).
Attentional requirements of learning: Evidence from performance measures
.
Cognitive Psychology
,
19
,
1
32
.
Østby
,
Y.
,
Tamnes
,
C. K.
,
Fjell
,
A. M.
,
Westlye
,
L. T.
,
Due-Tønnessen
,
P.
, &
Walhovd
,
K. B.
(
2009
).
Heterogeneity in subcortical brain development: A structural magnetic resonance imaging study of brain maturation from 8 to 30 years
.
Journal of Neuroscience
,
29
,
11772
11782
.
Poeppel
,
D.
, &
Hickok
,
G.
(
2004
).
Towards a new functional anatomy of language
.
Cognition
,
92
,
1
12
.
Poldrack
,
R. A.
,
Clark
,
J.
,
Paré-Blagoev
,
E. J.
,
Shohamy
,
D.
,
Creso Moyano
,
J.
,
Myers
,
C.
, et al
(
2001
).
Interactive memory systems in the human brain
.
Nature
,
414
,
546
550
.
Rakic
,
P.
(
1988
).
Specification of cerebral cortical areas
.
Science
,
241
,
170
176
.
R Core Team
. (
2012
).
R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing
.
Vienna, Austria
. http://www.r-project.org/
Saffran
,
J. R.
,
Aslin
,
R. N.
, &
Newport
,
E. L.
(
1996
).
Statistical learning by 8-month-old infants
.
Science
,
274
,
1926
1928
.
Saffran
,
J. R.
,
Johnson
,
E. K.
,
Aslin
,
R. N.
, &
Newport
,
E. L.
(
1999
).
Statistical learning of tone sequences by human infants and adults
.
Cognition
,
70
,
27
52
.
Saffran
,
J. R.
,
Newport
,
E. L.
, &
Aslin
,
R. N.
(
1996
).
Word segmentation: The role of distributional cues
.
Journal of Memory and Language
,
35
,
606
621
.
Saffran
,
J. R.
,
Newport
,
E. L.
,
Aslin
,
R. N.
,
Tunick
,
R. A.
, &
Barrueco
,
S.
(
1997
).
Incidental language learning: Listening (and learning) out of the corner of your ear
.
Psychological Science
,
8
,
101
105
.
Schapiro
,
A. C.
,
Gregory
,
E.
,
Landau
,
B.
,
McCloskey
,
M.
, &
Turk-Browne
,
N. B.
(
2014
).
The necessity of the medial temporal lobe for statistical learning
.
Journal of Cognitive Neuroscience
,
26
,
1736
1747
.
Schapiro
,
A. C.
,
Kustner
,
L. V.
, &
Turk-Browne
,
N. B.
(
2012
).
Shaping of object representations in the human medial temporal lobe based on temporal regularities
.
Current Biology
,
22
,
1622
1627
.
Schapiro
,
A. C.
,
Rogers
,
T. T.
,
Cordova
,
N. I.
,
Turk-Browne
,
N. B.
, &
Botvinick
,
M. M.
(
2013
).
Neural representations of events arise from temporal community structure
.
Nature Neuroscience
,
16
,
486
492
.
Schapiro
,
A. C.
,
Turk-Browne
,
N. B.
,
Botvinick
,
M. M.
, &
Norman
,
K. A.
(
2017
).
Complementary learning systems within the hippocampus: A neural network modelling approach to reconciling episodic memory with statistical learning
.
Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences
,
372
,
20160049
.
Schapiro
,
A. C.
,
Turk-Browne
,
N. B.
,
Norman
,
K. A.
, &
Botvinick
,
M. M.
(
2016
).
Statistical learning of temporal community structure in the hippocampus
.
Hippocampus
,
26
,
3
8
.
Schlichting
,
M. L.
,
Guarino
,
K. F.
,
Schapiro
,
A. C.
,
Turk-Browne
,
N. B.
, &
Preston
,
A. R.
(
2017
).
Hippocampal structure predicts statistical learning and associative inference abilities during development
.
Journal of Cognitive Neuroscience
,
29
,
37
51
.
Schneider
,
W.
,
Eschman
,
A.
, &
Zuccolotto
,
A.
(
2002
).
E-Prime user's guide
.
Pittsburgh, PA
:
Psychology Software Tools
.
Ségonne
,
F.
,
Dale
,
A. M.
,
Busa
,
E.
,
Glessner
,
M.
,
Salat
,
D.
,
Hahn
,
H. K.
, et al
(
2004
).
A hybrid approach to the skull stripping problem in MRI
.
Neuroimage
,
22
,
1060
1075
.
Shaw
,
P.
,
Greenstein
,
D.
,
Lerch
,
J.
,
Clasen
,
L.
,
Lenroot
,
R.
,
Gogtay
,
N.
, et al
(
2006
).
Intellectual ability and cortical development in children and adolescents
.
Nature
,
440
,
676
679
.
Sheridan
,
M.
,
Kharitonova
,
M.
,
Martin
,
R. E.
,
Chatterjee
,
A.
, &
Gabrieli
,
J. D. E.
(
2014
).
Neural substrates of the development of cognitive control in children ages 5–10 years
.
Journal of Cognitive Neuroscience
,
26
,
1840
1850
.
Shohamy
,
D.
(
2011
).
Learning and motivation in the human striatum
.
Current Opinion in Neurobiology
,
21
,
408
414
.
Shohamy
,
D.
,
Myers
,
C. E.
,
Grossman
,
S.
,
Sage
,
J.
,
Gluck
,
M. A.
, &
Poldrack
,
R. A.
(
2004
).
Cortico-striatal contributions to feedback-based learning: Converging data from neuroimaging and neuropsychology
.
Brain
,
127
,
851
859
.
Shohamy
,
D.
, &
Turk-Browne
,
N. B.
(
2013
).
Mechanisms for widespread hippocampal involvement in cognition
.
Journal of Experimental Psychology: General
,
142
,
1159
1170
.
Sowell
,
E. R.
,
Thompson
,
P. M.
,
Leonard
,
C. M.
,
Welcome
,
S. E.
,
Kan
,
E.
, &
Toga
,
A. W.
(
2004
).
Longitudinal mapping of cortical thickness and brain growth in normal children
.
Journal of Neuroscience
,
24
,
8223
8231
.
Thiessen
,
E. D.
(
2017
).
What's statistical about learning? Insights from modelling statistical learning as a set of memory processes
.
Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences
,
372
,
20160056
.
Tisdall
,
M. D.
,
Hess
,
A. T.
,
Reuter
,
M.
,
Meintjes
,
E. M.
,
Fischl
,
B.
, &
van der Kouwe
,
A. J. W.
(
2012
).
Volumetric navigators (vNavs) for prospective motion correction and selective reacquisition in neuroanatomical MRI
.
Magnetic Resonance in Medicine
,
68
,
389
399
.
Toro
,
J. M.
,
Sinnett
,
S.
, &
Soto-Faraco
,
S.
(
2005
).
Speech segmentation by statistical learning depends on attention
.
Cognition
,
97
,
B25
B34
.
Toro
,
J. M.
, &
Trobalón
,
J. B.
(
2005
).
Statistical computations over a speech stream in a rodent
.
Perception & Psychophysics
,
67
,
867
875
.
Turk-Browne
,
N. B.
,
Jungé
,
J. A.
, &
Scholl
,
B. J.
(
2005
).
The automaticity of visual statistical learning
.
Journal of Experimental Psychology: General
,
134
,
552
564
.
Turk-Browne
,
N. B.
, &
Scholl
,
B. J.
(
2009
).
Flexible visual statistical learning: Transfer across space and time
.
Journal of Experimental Psychology: Human Perception and Performance
,
35
,
195
202
.
Turk-Browne
,
N. B.
,
Scholl
,
B. J.
,
Chun
,
M. M.
, &
Johnson
,
M. K.
(
2009
).
Neural evidence of statistical learning: Efficient detection of visual regularities without awareness
.
Journal of Cognitive Neuroscience
,
21
,
1934
1945
.
Wagner
,
A. D.
,
Schacter
,
D. L.
,
Rotte
,
M.
,
Koutstaal
,
W.
,
Maril
,
A.
,
Dale
,
A. M.
, et al
(
1998
).
Building memories: Remembering and forgetting of verbal experiences as predicted by brain activity
.
Science
,
281
,
1188
1191
.