Edited by: Ilaria Grazzani, University of Milano-Bicocca, Italy
Reviewed by: Eliza L. Nelson, Florida International University, United States; Chiara Cantiani, Eugenio Medea (IRCCS), Italy
This article was submitted to Developmental Psychology, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
While a number of studies have found that an improvement in object shape recognition is associated with language growth in infants and toddlers, no published studies have investigated the longitudinal relation between early shape recognition, and language abilities in later childhood. An electrophysiological measure of semantic processing (the N400) was used to assess shape recognition and general object recognition in a naming context in 20-month-olds. The measures of shape recognition strongly predicted language and cognitive abilities at 6–7 years even after controlling for toddler vocabulary size. The electrophysiological measures of general object recognition were not related to future language or cognitive abilities. These results suggest that early shape recognition abilities may play a role in language acquisition and influence even long-term language outcomes.
The growth of children’s early vocabularies has been systematically linked to the development of the so-called “shape bias”: the tendency to extend newly learned words to objects similar in shape (
Important work has demonstrated that supporting the development of a shape bias can have a direct beneficial effect on vocabulary growth. By familiarizing toddlers with novel object categories organized by shape,
While the predictive validity of early shape recognition for language outcomes at school entry has not been explored, many studies show that individual differences in early word learning ability predict language skills in later childhood. Vocabulary size in toddlers is often measured using parent report (e.g.,
Electrophysiological methods have also proved useful in investigating how infants and toddlers process and understand the meaning of words in real-time. By presenting words in contexts that are semantically matching or mismatching, children’s understanding of word meanings can be directly measured without requiring a response. Results from such paradigms indicate that individual differences in the N400 effect, an event-related potential (ERP) component which indexes lexical-semantic processing, are associated with language skills. The N400 component is an ERP component elicited by words and other meaningful stimuli, and its’ amplitude is attenuated by any aspect of the context that is associated with, or helps predict, the word. In contrast, cues that are incongruent to the word meaning increase the N400 amplitude (for reviews on the N400, see
Using an ERP picture-word paradigm, we recently reported online brain measures of toddlers’ ability to recognize object shape and object part information during word comprehension (
In the present study, we assessed the same children at age 6–7 years with a large battery of standardized tests measuring language and other cognitive abilities. This provides the first study of the relevance of shape recognition for language skills in school-aged children, as well as the first study to investigate the longer-term predictive validity of brain measures of online semantic processing in toddlers. In a retrospective study,
The sample size included in the various analyses differed due to many children contributing only to parts of the data set. The ERP data at 20 months suffered the highest attrition rates, and thus the main longitudinal analyses between ERP data at 20 months and language and cognitive tests at 6–7 years were performed on a sample of 23 children.
The full toddler sample consisted of 77 children (36 boys) at 20 months of age (±3 weeks). Inclusion criteria were: typical development according to parent report, monolingual Swedish speaking, and full term birth (>36 gestational weeks). Reliable electrophysiological data was obtained from 38 children (17 boys). The other 39 children were excluded from EEG analysis due to fussiness, technical problems, or too few artifact-free trials in one or more of the analyzed conditions. Developmental data from questionnaires was obtained for 74 children. The children returned at 24 months (±3 weeks) for the same experimental procedure with novel stimuli. The present study only reports ERP data from the 20 months session, since this time point best captured the period of emergence of the shape recognition/shape bias skills (50–100 words vocabulary, and in between 18 and 24 months). In addition to the ERP data, the current analyses included measures of vocabulary size from both 20 and 24 months. Participants were recruited mainly through child healthcare centers in and around Lund, Sweden, and an information campaign sent by mail to all children in certain areas close to Lund that would fall within the appropriate age range during the study period. The project was granted ethical approval by the Regional Ethical Review Board, according to the decision DNR-2009-383, and the parents of all participants provided written informed consent.
Families that participated at 20 months were invited to participate in the follow-up study, and 36 (16 girls) of those contacted agreed. At the time of testing, the children were between 6:03 and 7:09 years old. Twenty-three of these participants had contributed reliable ERP data at 20 months. The parents of all subjects gave written informed consent in accordance with the Declaration of Helsinki. The project was approved by the Institutional Review Board at the Section of Logopedics, Phoniatrics, and Audiology at Lund University.
The event-related potential experiment at 20 months contained 30 common count nouns and 30 pseudowords which were phonotactically legal in Swedish. Words were recorded in an anechoic chamber by a female voice, speaking in an infant-directed manner, and presented through speakers placed in front of the participants. The visual stimuli consisted of cartoon images
Tests were chosen from several different standardized batteries, to measure the following linguistic and cognitive abilities: receptive and expressive language abilities, verbal short-term and working memory, communicative skills (by parent questionnaire), and fluid intelligence. See
Test battery at 6–7 years.
Sentence comprehension | Test for reception of grammar (TROG-2) ( |
Expressive vocabulary | Clinical evaluation of language fundamentals (CELF-4) ( |
Similarities (expressive) | CELF-4 |
Similarities (receptive) | CELF-4 |
Digit span forward | CELF-4 |
Digit span backward | CELF-4 |
Speeded naming | A developmental neuropsychological assessment (NEPSY-II) ( |
Word fluency | NEPSY-II |
Non-word repetition | Sound information processing system (SIPS) ( |
Segment subtraction | Illinois test of psycholinguistic abilities (ITPA-3) ( |
Auditory analysis | ITPA-3 |
Overall communication skills (parent report) | Children’s communication checklist (CCC-2) ( |
Fluid intelligence | Raven’s colored progressive matrices (CPM) ( |
Children sat on their parent’s lap, with a screen placed around them in order to block out distractions. Pictures were presented on a 17-inch computer screen (34 cm × 27 cm) approximately 35 cm from the child, and words were presented from a speaker next to the screen. Breaks were taken between blocks if necessary, with the possibility of showing a short video clip to recapture the child’s attention. A camera placed in front of the child recorded the child’s behavior throughout the experiment, allowing for exclusion of trials where the child was inattentive.
The stimuli were organized into ten independent blocks, with each block containing three real words and objects and three pseudowords and novel objects. Please see
Overview of experimental design containing real words and objects. Figure reproduced from
The test battery was administered in the children’s homes. It took between 2 and 3 h to complete, including breaks. Sometimes a caregiver was present during testing, but instructed to remain silent. The tests were administered in the following order: TROG-2, word fluency, expressive vocabulary, segment subtraction, Raven’s colored progressive matrices (CPM), non-word repetition, digit span forward and backward, similarities receptive and expressive, speeded naming, and auditory analogies.
EEG data was pre-processed according to description in
Also following recommendations by
Results were scored according to each test’s manual, and normed when appropriate test-norms were available. The results section reports scores expressed in percentiles. Speeded Naming and non-word Repetition did not have available norms, so results from these tests are reported as raw scores. The Speeded Naming test produces two measures: accuracy and speed.
The age-normed results from the Raven’s CPM test showed a strong ceiling effect, with a mean score of percentile 83, and 16 participants obtaining the top score of percentile 95. The norms for this test are old (from 1982) and known to be too lenient. In order to obtain an age-adjusted measure of the Raven’s score that better differentiated between participants, we transformed the raw scores by performing a linear regression with age as an independent variable. The residuals were then saved as a new variable, with an added constant of 5 to remove negative numbers. By this procedure, the effect of age could be removed from the raw scores, creating a new variable that was normally distributed.
Statistical analyses were performed using IBM SPSS Statistics, version 22. All measures were checked for normality of distribution using the Shapiro-Wilks test. Only two school-age measures were normally distributed (Raven’s CPM raw scores, and non-word repetition), while all toddler measures except Productive Vocabulary at 20 months were normally distributed. To adjust the positively skewed distribution of this variable, a log10 transformation was applied. To explore the longitudinal relations between all toddler and school age measures, we first calculated bivariate correlations using Spearman’s rank order correlation for non-normally distributed variables, and Pearson’s correlations for normally distributed variables. A correlation matrix for all individual variables can be seen in
In order to reduce the number of variables at 6–7 years, and also facilitate interpretation of the longitudinal relations, further analyses focused on the eight variables that correlated with the toddler ERP measures (see
Descriptive statistics for all behavioral measures.
20 m rec. vocabulary | Raw | 33 | 183 | 190.88 (69.15) | 86–319 |
20 m prod. vocabulary | Raw | 33 | 60 | 108 (102.25) | 7–391 |
24 m prod. vocabulary | Raw | 31 | 313 | 290 (161.22) | 15–565 |
TROG-2 | Perc | 36 | 70 | 64.11 (25.92) | 1–96 |
Exp. vocabulary | Perc | 36 | 84 | 66.44 (29.62) | 1–98 |
Similarities (exp) | Perc | 36 | 37 | 46.89 (26.38) | 5–91 |
Similarities (rec) | Perc | 36 | 37 | 41.28 (25.39) | 5–84 |
Digit span forward | Perc | 36 | 25 | 38.31 (26.10) | 2–91 |
Digit span backward | Perc | 36 | 63 | 53.94 (23.76) | 16–98 |
Segment subtraction | Perc | 36 | 87.50 | 67.39 (33.42) | 9–99 |
Auditory analysis | Perc | 36 | 95 | 78.89 (26.66) | 16–99 |
Word fluency | Perc | 36 | 75 | 65.94 (27.99) | 2–99 |
CCC2 | Perc | 36 | 53.50 | 54.31 (28.21) | 2–97 |
Raven’s CPM | Perc | 36 | 90 | 83.19 (19.09) | 10–95 |
Raven’s CPM | Raw | 36 | 26 | 25.81 (5.47) | 13–34 |
Raven’s CPM raw age-adj. | Residuals | 36 | 5.07 | 5.00 (0.98) | 2.88–6.90 |
Speeded naming (acc.) | Raw | 36 | 84 | 102.06 (28.06) | 53–135 |
Speeded naming (time) | Raw | 36 | 150 | 167 (55.26) | 104–305 |
Non-word repetition | Raw | 35 | 13 | 13.20 (4.02) | 4–22 |
The resulting two factors were subsequently used as outcome variables instead of the seven measures that entered the analysis. Further longitudinal analyses focused on how the toddler measures were able to predict these two factors as well as Raven’s CPM as a measure of fluid intelligence. The three outcome variables were tested for normality of distribution using Shapiro-Wilks normality test. While two of the variables were clearly normally distributed, factor 1 fell just below the threshold of statistical significance (
The temporal PCA produced a total of 16 factors, of which only the first two factors accounted for a variance greater than 0.1 (all others had a variance <0.065). Therefore, only the first two components were further analyzed. Factor one (variance 0.37) was a late component, with a peak latency of 1092 ms. This component had its largest negative peak at channel 14 (frontopolar), and largest positive peak at channel 90 (right occipital). Factor two (variance 0.21) was a mid-latency component with a peak latency of 520 ms, a peak negativity at channel 74 (mid occipital), and a peak positivity at channel 122 (right frontal, F8). These two components were further examined in terms of the experimental effect of congruity (amplitude difference between condition where the word is presented with the correct vs. incorrect object referent), in both the regular object conditions and the shape object conditions. An incongruity effect in the regular object conditions indicates that the child recognizes an object from a typical illustration and knows the correct label for it, while an incongruity effect in the shape conditions is a measure of the child’s ability to also recognize the object from a sparse silhouette representation, i.e., object shape recognition. The mid-latency component (factor 2) showed a congruity effect over parietal and occipital areas, in both the regular and shape object conditions (see
Illustration of the two temporal factors (TF01 and TF02) from the PCA analysis. The figure includes topographical plots of the components in each of the four conditions (regular object vs. shape, and congruous vs. incongruous presentations), where the color blue represents a negative amplitude, and red represents a positive amplitude. The graphs show the temporal properties of the components in each of the four conditions (negative amplitude plotted downward). Electrodes showing the largest effect of incongruity were chosen to illustrate the component waveforms, and for TF01 the topography of this effect differed in the regular object and shape conditions, which is why we display both a central electrode (Cz) for the shape incongruity effect, and parietal/occipital electrode (E75) for the regular incongruity effect.
Repeated measures ANOVAs confirmed that all these congruity effects were statistically significant (see section “Materials and Methods” for details): shape N400,
Descriptive statistics for the cognitive measures at age 6–7 years, as well as 20 to 24 months, are presented in
Of the four ERP measures at 20 months, analyses showed that only the two components from the shape conditions correlated with future cognitive measures (see
The principal components analysis analysis of the selected outcome measures (see section “Materials and Methods”) resulted in two factors. Factor 1 loaded most heavily on the variables Expressive vocabulary and TROG, while the Factor 2 loaded most heavily on the backward digit span test, but also on the forward digit span and speeded naming accuracy measure. From the pattern of loadings, we found it reasonable to conceptualize Factor 1 as “language ability” (receptive and expressive), and Factor 2 as “verbal executive function” (see
Pattern matrix for the PCA analysis.
TROG | 0.975 | –0.154 |
Expressive vocabulary | 0.955 | –0.148 |
Non-word repetition | 0.744 | 0.195 |
Digit span forward | 0.367 | 0.609 |
Digit span backward | –0.260 | 0.921 |
Similarities expressive | 0.540 | 0.339 |
Speeded naming (accuracy) | 0.072 | 0.452 |
Instead of the original 15 outcome variables, our main analyses focused on three cognitive variables at age 6–7 years: the latent variables “language ability,” “and “verbal executive function” as well as the observable variable “fluid intelligence” (see section “Materials and Methods”). These three cognitive variables correlated with each other: language ability and fluid intelligence were most strongly related
Longitudinal correlations between toddler measures and latent factors in school age.
Shape N400 20 m | |||
Shape late neg. 20 m | 0.307 | –0.086 | |
> 0.001 | |||
Productive vocab. 24 m | 0.148 | ||
Receptive vocab. 20 m | 0.200 | ||
0.019 | 0.273 | 0.008 | |
31 | 32 | 33 |
Scatter plots of relation between the shape N400 difference score (negative difference score represents a larger N400 incongruity effect) and the three outcome measures at 6–7 years.
With a regular alpha-level of 0.05, the shape N400 component strongly predicted language ability, but also verbal executive function and fluid intelligence. The relation between the shape N400 and language ability remained significant after controlling for verbal executive function (
A linear regression model demonstrated that the two predictors Shape N400 effect at 20 months and productive vocabulary size at 24 months together explained 57% (adjusted
This study investigated individual differences in electrophysiological measures of semantic processing and object recognition in toddlers and how they relate to the children’s cognitive development 4–5 years later. The results show that the regular N400-effect, which in our paradigm depends on both semantic processing of words and recognition of common objects, did not differentiate between children on any future measures of language or cognition. The incongruity responses related to object shape recognition, however, correlated with several measures at 6–7 years. These incongruity responses provide an electrophysiological measure of the toddlers’ ability to recognize common objects from only overall shape. Thus, relations between these toddler and school-age measures show that individual differences toddlers’ shape recognition can predict future cognitive skills. To understand the pattern of correlations, and to improve the psychometric properties of our measures by creating normally distributed variables, we used a principal components analysis, where the cognitive tests clustered into two factors. We will argue that it is reasonable to understand one factor as
The tests that most clearly differentiated between the two factors were TROG-2 (receptive grammar comprehension) and expressive vocabulary on the one hand (factor 1), and digit span backward on the other hand (factor 2). The other tests that were included in factor 1 were similarities expressive (abstract conceptual reasoning) and non-word repetition. Thus, factor 1 clearly involves both receptive and expressive language skills, and both semantic, syntactic and phonological processing. Therefore, it seemed reasonable to interpret the factor as general
Finally, the third outcome measure used in the analyses was Raven’s Colored Progressive Matrices. We chose to include this test in the study because the Raven’s matrices series of tests is the most widely used measure of fluid intelligence. The concept of fluid intelligence refers to the ability to solve novel problems using logical reasoning, and the matrix task is a non-verbal measure of this ability (
A main conclusion from our results is that brain measures of shape recognition in toddlers predict language ability 4–5 years later. Toddlers with larger N400 incongruity effects in the shape conditions developed better language skills. However, the regular N400 incongruity effect was unrelated to language development. This pattern of results indicates that the process of quickly recognizing object silhouettes and mobilizing the associated verbal label is more predictive of future language development than general efficiency of semantic processing. Importantly, the incongruity response during the shape condition depends on more than simple vocabulary knowledge (captured by the regular object condition), rather it requires that the child has paid attention to the global shape properties of the object. This finding extends previous research on the children’s shape recognition, which has established that toddler’ language development is associated with an improvement in shape recognition and the subsequent development of a shape bias (
Our analysis of the ERP data indicates that the task of recognizing objects and quickly processing whether an object is semantically related to a word involves at least two distinct stages of processing that are affected by the semantic relatedness. The first stage, captured by a classic N400 effect, may be reasonably interpreted as a stage of lexical access, where the picture of an object primes the associated word and thus facilitates lexical access of words in congruous presentations (for reviews on the N400, see
There also seems to be a second stage of processing the semantic information, indexed by a late ERP component also demonstrating an incongruity effect. Although it is not clear whether this effect can be classified as a particular established ERP component, later stage ERP components are often interpreted as involving contextual integration or memory updating (
To our knowledge there are no previous reports that N400 responses have predictive value over such a long period of time. Other ERP responses related to word processing have been shown to capture individual differences in language ability, although few studies show longer-term predictive value. The N200-500 component, which relates to word form familiarity or word recognition (for a review, see
We know from previous research that the stability of rank order in language ability in young children is fairly low, which makes it difficult to use toddler vocabulary measures to predict language delays even a couple of years later (
The cognitive test battery used in the present study merits some consideration. The PCA analysis of the selected outcome measures resulted in two latent factors labeled language ability and verbal executive function. The factor language ability loaded most heavily on the three following variables: language comprehension (comprehension of grammar), expressive vocabulary and non-word repetition. This did not come as a surprise since these tests are highly important predictors of language development. Non-word repetition not only predicts vocabulary development, but is even considered a clinical marker of developmental language disorder in children this age (
As is often the case with longitudinal studies, the present study has weaknesses related to data attrition and the risk of obtaining false positive results due to many data variables. The most important results reported rest on a data-set of 23 participants. A sample of this size renders the correlational analyses low-powered and increases the risk of false positive results. On the other hand, the type of data we offer, electrophysiological developmental data as well as longitudinal data from toddlerhood to school entry, is rare and therefore may illuminate relationships which have not been observed previously. The choice of using principal components analyses for both the ERP data and the cognitive data was an attempt to improve properties of the data sets. For the ERP data, the results of the PCA are potentially less noisy and more precise than alternative measures such as mean difference amplitude across a certain time window. For the test measures at 6–7 years, the PCA resulted both in fewer variables (which decreases the risk of false positive correlations) and variables with better statistical properties (normally distributed) so that better methods for data analysis could be used. The many bivariate correlations calculated in
The current study provides the first evidence that electrophysiological responses in tasks measuring semantic processing have predictive value from toddlerhood to early school age. A measure of object shape recognition at age 20 months was found to predict language abilities and verbal executive functions 4–5 years later, even when language abilities in toddlerhood were taken into consideration. Electrophysiological measures of general object recognition were not associated with later language and cognitive abilities, suggesting that the predictive value was specific to object shape.
The datasets generated for this study are available on request to the corresponding author.
All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Institutional Review Board at the Section of Logopedics, Phoniatrics, and Audiology, Lund University.
KB had overall responsibility for the study design, collecting all the data at 20 months, as well as all data processing, statistical analysis, and writing of the first draft of the manuscript. BS led the designing of the follow-up study and supervising the students who performed the standardized tests at 6–7 years. JT and ML were highly involved in the designing of the ERP study, and analyzing and interpreting of the ERP data. All authors contributed to interpreting all the results, and reviewing and editing of the manuscript.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Many thanks to Andrea Fölster and Jenny Hansson for their contribution in the data collection and to the families who generously participated in this study.
The Supplementary Material for this article can be found online at:
The toddler data presented in this paper were part of a large project investigating several different aspects of semantic processing and word learning at 20 and 24 months, and the results have been reported previously (