Abstract
Objective
Quantity of talk and interaction in the home during early childhood correlates with socioeconomic status (SES) and predicts early language and cognitive outcomes. This study tested the effectiveness of automated early language environment estimates for children 2–36 months to predict cognitive and language skills 10 years later and examined effects for specific age periods of early development.
Methods
Daylong audio recordings for 146 infants and toddlers were completed monthly for 6 months, and the total number of adult words and adult-child conversational turns for recording days were automatically estimated with LENA software. Follow-up evaluations at 9–14 years of age included language and cognitive testing using the Wechsler Intelligence Scale for Children-Fifth Edition (WISC-V – IQ and Verbal Comprehension Index, VCI), Peabody Picture Vocabulary Test (PPVT), and Expressive Vocabulary Test (EVT). Language exposure for three age groups was assessed: 2–17, 18–24 and 25+ months. Pearson correlations and multiple linear regression analyses were conducted.
Results
Conversational turn counts at 18–24 months of age accounted for 14–27% of the variance in IQ, verbal comprehension, and receptive/expressive vocabulary scores ten years later, after controlling for SES. Adult word counts between 18–24 months correlated with language outcomes but were weakened after controlling for SES.
Conclusion
These data support the hypothesis that early talk and interaction, particularly during a relatively narrow developmental window of 18–24 months, predicts school-age language and cognitive outcomes. These findings underscore the need for early intervention programs that support parents to create an optimal early language learning environment in the home.
Table of Contents Summary:
This 10-year study explores the relationship between very early child language experience and language and cognitive skills in late childhood.
Language experience in the second year of life predicts language outcomes in late childhood
In their landmark study, Hart & Risley quantified the language environments of typically developing infants and toddlers, finding that adult word exposure between 10–36 months of age predicted child IQ at age three.1,2 Their work and subsequent research provide strong evidence that early language exposure predicts developmental outcomes.3–6 In response, pediatric interventions have been developed to help parents and caregivers boost talk and interaction with young children; several of these incorporate LENA (Language ENvironment Analysis) software to characterize early language experience via automated analysis of daylong audio recordings.7–10 Although these interventions reportedly have increased early talk and improved child language skills, research is needed on the long-term relationship between automated measures of early language experience and later developmental outcomes. The current study tested whether cognitive and language skills in children 9–14 years of age correlated with automated estimates of their early language experience and examined whether long-term outcomes were predicted differentially during three periods of early development. Note that we use predictiveness throughout in a statistical rather than explanatory or causative sense.
Decades of research have provided empirical evidence linking early language exposure and developmental outcomes.1–3,5,6 In one study, 14- to 26-month-olds exposed to more adult words had higher rates of vocabulary development than those exposed to fewer words.4 In another, high-socioeconomic status (SES) mothers of 18- to 29-month-olds spoke more often and with more varied vocabulary than mid-SES mothers, and their children demonstrated more advanced lexical development.11 Research connecting adult word exposure with higher rates of vocabulary development generally has focused on children with spoken lexicons in the second and third years of life, when many children undergo linguistic changes that may influence interlocutors and patterns of interaction in their environment.12,13 In fact, it has long been argued that parents adjust their speech to infants and children based on recognition of their level of development and awareness that child speech complexity changes with age.12,13 For example, children under 18 months of age rarely engage others using word combinations.14 Then around 18 months children commonly produce their first combinatorial speech (two-word utterances), and their spoken vocabulary increases rapidly, a phenomenon sometimes called the ‘word spurt.’15–17 The existence and nature of a literal word spurt is debatable – some researchers explain accelerated vocabulary learning and the appearance of word combinations by invoking a “naming insight,” and others suggest accelerated acquisition may be a by-product of parallel word learning and variation in the time to learn new words.18–20 Ganger & Brent argue against a word spurt but concede that “it is uncontroversial that a child’s rate of word learning increases during the 2nd year of life” (p. 621).21 Regardless of its origins, a landscape change in language use is observed around this age. Subsequently, children beyond 24 months start to produce longer utterances including grammatical morphemes and multi-clausal sentences.14
Although follow-up with Hart & Risley’s original sample showed that child word complexity and length of utterances between 10–36 months of age predicted academic outcomes in third grade (e.g., expressive/receptive language, spelling, reading), analyses were presented in aggregate and did not examine effects for specific age groups.22 Little is known regarding whether language experience during different developmental periods may uniquely impact long-term outcomes. This study addresses the question longitudinally using LENA language experience metrics extracted from automatically-analyzed, full-day audio recordings collected from infants and toddlers to predict their language and cognitive skills in middle school. These relationships are examined across the full span of recording ages as well as within subgroups of children 2–17, 18–24, and 25+ months of age. Analyses within these age groups address the possibility that the long-term impact of a child’s parent-generated language environment may depend in part on developments in child utterance complexity.
Methods
Phase I
Study Design.
The initial 2006 phase of this research has been reported in detail.23 Briefly, 329 children predominantly between 2–36 months of age (nine between 38–47 months) were recruited from the Denver-metro area, matching the US census on an SES proxy (mother’s attained education). Families completed daylong (12-hour) audio recordings monthly for 6 months.
Apparatus.
LENA software automatically processed audio recordings to quantify adult word exposure, child vocalization, and turn-taking interactions throughout the day based on algorithmic analysis.7 The adult word count (AWC) algorithm does not recognize words directly but analyzes acoustic information, e.g., related to syllable counts and consonant distributions, to estimate counts. The recording device registers all speech near the child, thus AWC includes both overheard and child-directed speech. Child vocalizations (CVs) quantify speech-related vocalizing by the child. The conversational turn count (CTC) quantifies adult-child alternations (vocal initiations with responses that occur within 5 seconds). Both intentional vocal responses and accidental vocal contiguity can be included in the CTC measure.24
Reliability of LENA’s automated speaker segmentation has been extensively reported, with accuracy of identification of adult and child segments between 68–82 percent.25–28 Reliability for LENA measures was based on 5,000 minutes of transcribed recording data from 94 children ages 2–48 months, 30–70 minutes each (mean = 53.2, SD = 12.7 minutes). AWC, CV, and CTC correlated highly with human transcription counts, r = .95, .82, and .83 respectively, all P < .001. Differences between transcribed counts and LENA estimates were uncorrelated with age for AWC, r = −.12, P = .27, CV, r = .16, P = .11, or CTC, r = .06, P = .57. Concurrent validity of LENA measures in the Phase I sample was demonstrated by significant correlations with language assessments administered by a certified Speech-Language Pathologist (SLP).23
Measures.
Participant children were evaluated by a certified SLP on a battery of assessments. A composite child language skills score (mean = 100, SD = 15) was generated by averaging total (expressive and receptive) language standard scores from the Preschool Language Scale-Fourth Edition and the Receptive Expressive Emergent Language Test-Third Edition.29,30 Parents completed an age-appropriate version of the MacArthur-Bates Communicative Development Inventory, from which the child’s vocabulary size was included.31 Figure 1 illustrates growth in vocabulary across age for 90 participants for whom it was available concurrently with recordings and indicates that sample children demonstrated accelerating vocabularies around 18 months.
Figure 1.
Phase I Vocabulary Size by Child Age Month
Note. Each marker represents one child.
Phase II
Study Design.
In Phase II, letters sent to Phase I families invited the now early school age children (9–13 years old) to complete follow-up language and cognitive assessments with a clinical psychologist, during which time the parent completed a demographic questionnaire. Upon completion, children were given a $50 gift card. Participants were not provided with assessment results, and the evaluator was blind to Phase I data and results.
Participants.
Participant addresses were updated via phone calls and email correspondences between phases. Figure 2 shows the derivation of the study sample in a simplified flow diagram. Ultimately, 146 Denver-area families provided informed consent approved by Heartland Institutional Review Board and participated in Phase II (see Table 1). Over 95% of Phase II children were 36 months of age or younger at Phase I onset. No differences were found between Phase II participant and non-participant families (n = 183) on child gender or age at recording, but more Phase II mothers had attended college (64.4% versus 44.8%), χ2 = 12.51, P < .001.
Figure 2.
Phase II Participant Recruitment Flow
Table 1.
Number of Phase II Participants by Phase I Child Age Group
| Age Sample: | 2–17 Months n = 51 |
18–24 Months n = 44 |
25+ Months n = 51 |
Full Sample N = 146 |
|---|---|---|---|---|
| Child Gender | ||||
| Female | 26 | 21 | 25 | 72 |
| Male | 25 | 23 | 26 | 74 |
| Mother’s Educationa | ||||
| Some High School | 3 | 6 | 1 | 10 |
| High School Diploma | 9 | 17 | 16 | 42 |
| Some College | 21 | 10 | 15 | 46 |
| Bachelor’s Degree + | 18 | 11 | 19 | 48 |
Mother’s highest attained education level includes: Some high school without a diploma; high school diploma or equivalent (e.g., GED); some college but less than bachelor’s degree; bachelor’s or more advanced degree.
Measures.
Participant children were administered the Wechsler Intelligence Scale for Children-Fifth Edition (WISC-V), the Peabody Picture Vocabulary Test (PPVT), and the Expressive Vocabulary Test (EVT).32–34 The WISC-V, for children 6–16 years of age, comprises five Primary Index Scales (Verbal Comprehension, Visual-Spatial, Fluid Reasoning, Working Memory, Processing Speed) which produce the Full-Scale Intelligence Quotient (IQ); of these, Full-Scale IQ and Verbal Comprehension Index (VCI) were included in this study. The PPVT is a widely used measure of receptive vocabulary for ages 2–90+ years in which respondents indicate which of four pictures matches a given word. The EVT, an expressive counterpart to the PPVT for the same ages, provides pictures that participants are asked to name. WISC-V administration generally takes 60 minutes, and the PPVT and EVT each take ~15 minutes. All assessment scores were standardized (mean = 100, SD = 15) against the general population.
Analysis
All valid recordings contributed by families over the six-month Phase I study were included for the full sample. LENA metrics were age-standardized (mean = 100, SD = 15) against a LENA Foundation corpus of 3,384 recordings from 378 families of typically developing children (including current participants) collected during Phase I and subsequent studies. These values were then averaged within family across recordings to produce one representative value and to minimize random variation in monthly scores. For this study, the early child language experience is characterized by AWC and CTC; CV is included only as a measure of child volubility.
Pearson correlations were calculated for AWC and CTC with outcome measures and then recalculated adjusted for SES and repeated by age subgroups. For analyses within age groups, only age-appropriate recordings were used, and each family was represented in only one age group, the one for which they contributed the maximum number of recordings (with preference given to the 18–24 months group to improve sample balance). That is, one set of LENA values that covered the full six-month period and one set restricted to an age subgroup (e.g., 18–24 months) were analyzed. Mean full sample and age-restricted values were highly correlated for AWC, CV, and CTC (r = .98, .96, and .97, respectively, all P < .001). Additional analyses examined the impact of a single recording per family and corrected for inequality of variance in age subgroups. Finally, to examine the possibility that CTC might not only reflect meaningful caregiver/child interaction but could act as a proxy for other child language characteristics (e.g., volubility), we conducted a multiple linear regression analysis that controlled for children’s Phase I CV, language skills, and vocabulary size. See Table 2 for more detail.
Table 2.
Phase I Language Measures by Child Age Group
| Age Sample: | 2–17 months n = 51 |
18–24 months n = 44 |
25+ months n = 51 |
Full Sample N=146 |
|---|---|---|---|---|
| Mean (SD) Range |
Mean (SD) Range |
Mean (SD) Range |
Mean (SD) Range |
|
| Age (mos) at Recording | 8.97 (4.08) 2.66–16.24 |
21.43(1.72) 19.14–24.43 |
34.22(5.42) 26.66–48.21 |
21.48(11.41) 2.66–48.21 |
| Recordings per Family | 4.82(1.73) 1–7 |
3.77(1.55) 1–6 |
5.14(1.13) 2–7 |
5.17(1.35) 1–7 |
| Conversational Turns SS | 99.93(10.13) 82.11–127.35 |
97.76(15.5) 61.30–131.02 |
100.82(12.59) 77.59–131.06 |
99.99(12.64) 61.30–135.42 |
| Adult Word Count SS | 105.52(20.69) 62.55–152.74 |
96.72(16.12) 57.41 – 129.03 |
101.01(15.74) 68.75–141.42 |
101.56(18.05) 57.41–152.74 |
| Child Vocal-ization SS | 101.26(9.85) 81.34–130.71 |
97.32(14.41) 60.27–122.77 |
99.87(10.88) 71.65–124.69 |
99.94(11.30) 60.27–130.71 |
| Conversational Turns Count | 285 (120) 118–750 |
440(221) 59 – 967 |
548 (232) 212–1166 |
428 (224) 59–1,166 |
| Adult Word Count | 14,484(6,022) 4,256–30,794 |
11,768(4,159) 3,372–21,069 |
12,928(4,342) 5,305–25,810 |
13,218(5,079) 3,372–30,794 |
| Child Vocal-ization Count | 1,110(396) 449 – 2,227 |
1,736 (716) 321–3,439 |
2,316(728) 711–4,176 |
1,730 (786) 321–4,176 |
| aLanguage Skills | 100.70(8.81) 80.5–122.5 |
104.06(13.2) 76.00–130.00 |
112.21(15.74) 56.00–139.00 |
105.69(13.70) 56.00–139.00 |
| bVocabulary | 13.95(25.38) 0–125 n = 44 |
177.48(143.86) 1–470 n = 27 |
455.79(168.96) 187–670 n=19 |
156.29(203.55) 0–670 N = 90 |
Note: LENA SS (standard scores) age-normalized to mean=100, SD=15. LENA Counts reflect totals across a 12-hour recording day.
Language Skills score was unavailable for one child in the 25+ months age group.
MacArthur-Bates Vocabulary scores were unavailable for some participants outside the assessment age range.
Results
Full sample
Table 3 summarizes Phase II assessment scores for the full sample and age subgroups. Pearson correlations with LENA measures are shown in Table 4. CTC was associated with VCI, PPVT, and EVT but did not significantly predict IQ. AWC predicted only VCI. CTC and AWC demonstrated a strong concurrent relationship with each other, r = .74 [95% confidence interval (CI): .66 – .81], P < .001, R2 = .55.
Table 3.
Phase II Language Standard Scores by Phase I Child Age Group
| Age Sample: | 2–17 months n = 51 |
18–24 months n = 44 |
25+ months n = 51 |
Full Sample N = 146 |
|---|---|---|---|---|
| Mean (SD) Range |
Mean (SD) Range |
Mean (SD) Range |
Mean (SD) Range |
|
| Age (yrs) at Assessment | 10.69 (0.57) 9.74 – 12.24 |
11.79 (0.53) 10.99 – 12.99 |
12.74 (0.64) 11.70 – 14.59 |
11.74 (1.04) 9.74 – 14.59 |
| IQ | 103.78 (17.04) 73 – 137 |
104.82 (13.36) 71 – 132 |
102.69 (12.65) 73 – 131 |
103.71 (14.46) 71 – 137 |
| VCI | 105.10 (15.06) 68 – 133 |
105.36 (14.04) 68 – 127 |
101.84 (10.63) 76 – 124 |
104.04 (13.35) 68 – 133 |
| PPVT | 112.12 (16.16) 76 – 146 |
113.57 (14.85) 76 – 137 |
111.02 (11.41) 76 – 134 |
112.17 (14.19) 76 – 146 |
| EVT | 106.61 (14.06) 75 – 143 |
109.93 (14.12) 78 – 143 |
106.75 (9.87) 77 – 124 |
107.66 (12.78) 75 – 143 |
Note: All scores age-normalized to mean=100, SD=15. IQ = Wechsler Adult Intelligence Scale Full Scale Intelligence Quotient; VCI = Verbal Comprehension Index; PPVT = Peabody Picture Vocabulary Test; EVT = Expressive Vocabulary Test.
Table 4.
Correlations between Phase I LENA Measures and Phase II Outcomes
| Conversational Turns | Adult Words | ||||||
|---|---|---|---|---|---|---|---|
| N | r (95% CI) | P | R2 | r (95% CI) | P | R2 | |
| All Ages | |||||||
| IQ | 146 | .14 [−.02 – .30] | .08 | .02 | .08 [−.09 – .24] | .35 | .01 |
| VCI | 146 | .27 [.11 – .42] | .001 | .07 | .20 [.04 – .35] | .02 | .04 |
| PPVT | 146 | .21 [.04 – .36] | .01 | .04 | .14 [−.03 – .29] | .10 | .02 |
| EVT | 146 | .21 [.04 – .36] | .01 | .04 | .10 [−.07 – .25] | .25 | .01 |
| 2–17 mos | |||||||
| IQ | 51 | −.05 [−.32 – .23] | .73 | .00 | −.11 [−.37 – .17] | .45 | .01 |
| VCI | 51 | .05 [−.23 – .32] | .74 | .00 | .02 [−.26 – .29] | .90 | .00 |
| PPVT | 51 | .02 [−.26 – .29] | .92 | .00 | −.01 [−.28 – .27] | .95 | .00 |
| EVT | 51 | .10 [−.18 – .37] | .48 | .01 | .09 [−.19 – .35] | .54 | .01 |
| 18–24 mos | |||||||
| IQ | 44 | .44 [.16 – .65] | .003 | .19 | .37 [.08 – .60] | .01 | .14 |
| VCI | 44 | .57 [.32 – .74] | <.001 | .32 | .42 [.14 – .64] | .005 | .18 |
| PPVT | 44 | .43 [.16 – .65] | .003 | .19 | .33 [.04 – .57] | .03 | .11 |
| EVT | 44 | .45 [.17 – .66] | .002 | .20 | .27 [−.03 – .52] | .08 | .07 |
| 25+ mos | |||||||
| IQ | 51 | .08 [−.20 – .35] | .58 | .01 | .08 [−.20 – .34] | .60 | .01 |
| VCI | 51 | .25 [−.03 – .49] | .07 | .06 | .22 [−.06 – .47] | .12 | .05 |
| PPVT | 51 | .19 [−.09 – .44] | .07 | .04 | .18 [−.10 – .43] | .21 | .03 |
| EVT | 51 | .02 [−.25 – .30] | .87 | .00 | −.10 [−.37 – .18] | .49 | .01 |
Note: IQ = Wechsler Adult Intelligence Scale Full Scale Intelligence Quotient; VCI = Verbal Comprehension Index; PPVT = Peabody Picture Vocabulary Test; EVT = Expressive Vocabulary Test.
Age groups and SES
Subsequent analyses examined possible systematic variation related to child age. To generate a smoothed representation of the relationship between early language experience and later outcomes, a moving average age window was defined such that recording values were averaged within-family for each target age month +/− 3 months. For example, language values for age 18 months were computed as the average of available values from 15–21 months. Figure 3 displays the relationships between CTC/AWC and primary outcome measures for the resulting seven-month age window. The strongest relationships (solid lines indicate statistical significance) occurred in a middle period starting at ~18 months.
Figure 3.
Correlations between Phase I LENA Values and Phase II Outcome Measures
The sample was split into three exploratory age groupings (2–17, 18–24, and 25+ months). Pearson correlations between language experience predictors and outcomes within each age group are provided in Table 4. Essentially no significant relationships were observed for the 2–17 and 25+ age groups. However, both CTC and AWC strongly predicted outcomes in the 18–24 months group. Repeating these analyses controlling for maternal attained education as a marker for SES (Table 5) demonstrated that correlations in this group remained significant for CTC with IQ, VCI, PPVT, and EVT, but the predictive power of AWC was weakened considerably.
Table 5.
Correlations between LENA Metrics at 18–24 Months and Phase II Outcomes Adjusted for SES and for Equal Age Group Variance
| Conversational Turns | Adult Words | ||||||
|---|---|---|---|---|---|---|---|
| n | r (95% CI) | P | R2 | r (95% CI) | P | R2 | |
| SES Adjusted | |||||||
| IQ | 44 | .37 [.08 – .60] | .01 | .14 | .24 [−.06 – .50] | .12 | .06 |
| VCI | 44 | .52 [.26 – .71] | <.001 | .27 | .30 [.00 – .55] | .05 | .09 |
| PPVT | 44 | .37 [.08 – .60] | .01 | .14 | .21 [−.09 – .48] | .18 | .04 |
| EVT | 44 | .38 [.10 – .61] | .01 | .15 | .12 [−.19 – .40] | .46 | .01 |
| Equal Variance | |||||||
| IQ | 35 | .48 [.17 – .70] | .004 | .23 | .34 [.00 – .60] | .05 | .11 |
| VCI | 35 | .61 [.34 – .78] | <.001 | .37 | .36 [.03 – .62] | .04 | .13 |
| PPVT | 35 | .58 [.31 – .77] | <.001 | .34 | .38 [.05 – .63] | .03 | .14 |
| EVT | 35 | .54 [.25 – .74] | .001 | .29 | .18 [−.16 – .49] | .30 | .03 |
Note: IQ = Wechsler Adult Intelligence Scale Full Scale Intelligence Quotient; VCI = Verbal Comprehension Index; PPVT = Peabody Picture Vocabulary Test; EVT = Expressive Vocabulary Test.
Assessment of Sampling Issues
LENA metrics generally derived from multiple recordings per family (Table 2). To test whether similar results held using a single recording, we randomly selected one recording per family within each age group. Overall correlation patterns were similar; magnitudes were reduced somewhat but remained significant. For example, the multiple recordings 18–24 months CTC/VCI correlation was r = .57, P < .001; correlations computed from two random draws of one recording for this group were r = .43 and .47, both P < .01.
CTC distributions were further examined to investigate whether increased predictiveness for the 18–24 months age group could have resulted from greater variance in the language environment compared to younger or older children. The Levene test statistic W was computed to compare the homogeneity of variance among age groups. CTC variance in the 18–24 months group (SD = 15.5) was significantly larger than that in the younger group (SD = 10.1), W(1, 93) = 11.09, P = .001, and marginally so compared to the older group (SD = 12.6), W(1, 93) = 3.68, P = .06. Nine cases with the highest squared errors (compared to the group mean) were excluded from this group to achieve homogeneity of variance with the other age groups, W(2,134) = 1.16, P = .32. Recomputed correlations (controlled for SES) between CTC and language outcomes (Table 5) were higher than with the cases included, supporting the interpretation that the predictive strength for this group did not derive solely from increased CTC variance.
The CTC/VCI relationship in the 18–24 months group was the strongest observed. To evaluate whether this correlation could be accounted for by other child characteristics, we added three contemporaneously collected, potentially related measures of child language development.23 CTC correlated significantly with CV, language skills, and vocabulary size, r = .84, .57, and .63 respectively (all P < .001). However, results from multiple linear regression analyses predicting VCI from CTC controlling for these factors demonstrated that their addition to the model did not significantly alter the predictive power of CTC. Table 6 displays regression metrics for three models: Model 1 – turns only; Model 2 – turns plus vocalization and language skills; Model 3 – turns plus vocalization, language skills, and vocabulary size.
Table 6.
Multiple Linear Regression Predicting Verbal Comprehension (VCI) in Phase II from Phase I Measures at 18–24 Months
| Regression Model | n | Standardized Coefficient |
Unstandardized Coefficient |
t-test Value |
P | Model R2 |
Model Adj. R2 |
|---|---|---|---|---|---|---|---|
| Model 1 | |||||||
| Conversational Turns | 44 | .51 | .57 | 4.43 | <.001 | .32 | .30 |
| Model 2 | |||||||
| Conversational Turns | 44 | .67 | .74 | 3.06 | .004 | .33 | .28 |
| Child Vocalizations | 44 | −.19 | −.20 | −0.78 | .44 | ||
| Language Skills | 44 | −.01 | −.01 | −0.05 | .96 | ||
| Model 3 | |||||||
| Conversational Turns | 44 | .75 | .88 | 2.52 | .02 | .28 | .28 |
| Child Vocalizations | 44 | −.28 | −.29 | −0.86 | .40 | ||
| Language Skills | 44 | .04 | .04 | 0.13 | .90 | ||
| Vocabulary Size | 44 | −.01 | −.09 | −0.33 | .74 | ||
Note: The Child Vocalizations and Language Skills scores in Models 2 and 3 were available for all participants in the 18–24 months subgroup; MacArthur-Bates Vocabulary Size in Model 3 was available for 27/44 participants in this group. The Language Skills score was computed by averaging total language standard scores from the Preschool Language Scale – Fourth Edition and the Receptive Expressive Emergent Language Test – Third Edition.
Discussion
These results support the hypothesized relationship between early language experience and school age developmental and language outcomes using an objective, automated method to estimate conversational turns and adult words in the language environment. But, the strength of the association based on automated counts was age-dependent; in subgroup analyses the mean number of CTCs for young children specifically between 18–24 months of age predicted IQ, verbal comprehension, and expressive and receptive language skills at 9–13 years. Importantly, these correlations remained significant after adjustments for SES or child language development, suggesting that the impact of increased early interaction on long-term developmental outcomes extends beyond the influence of socioeconomic factors and child skills.
It is possible that the automated procedure is not yet sensitive enough to capture significant relationships outside the 18–24 months range, and that improvement in automatic detection will make it possible to observe significant correlations across all early ages. It is also possible that, although not indicated in our transcriptional analysis, the automated procedure is particularly well suited to accurate counts of conversational turns during the 18–24 months range. In short, though these findings show strong predictiveness of outcome measures by conversational turns in the 18–24 months range, they do not rule out possible relationships at earlier or later ages.
Assuming, however, that a most sensitive period for prediction of language outcomes by early language experience really does begin around 18 months (with individual variation expected), these findings might be explained by a newly emergent cognitive process such as a naming insight or with a more mechanistic proposal not necessitating cognitive insights. But whatever the cause for the increased vocabulary growth and onset of combinatorial speech observed during this period, the data presented here support the possibility that developmental changes associated with complexity in child vocalizing are concurrent with a particularly sensitive period for adult-child interaction.14 If specialized cognitive processes contribute to the onset of more frequent word usage based on symbolic reference to external entities, then perhaps during this period children increasingly engage in especially impactful, referentially meaningful exchanges. These turn-taking exchanges may prepare the child’s cognitive and linguistic capacities for enhanced growth as indicated by the correlational patterns depicted in Figure 3.19 Empirical evaluation of this possibility requires further investigation. Although little is known about the neuronal mechanisms underlying accelerated vocabulary acquisition after 18 months, sensitive periods for language acquisition suggest that brain architecture may be differentially receptive to environmental input at different periods in early childhood.35–37 Numerous studies have explored such sensitive periods and the long-term effects that can result when normal patterns of experience are disturbed during development.38 The present work supports research that suggests exposure to tailored experience during specific periods of early development may have important effects on later development.35–37
The current study expands on previous research by addressing certain limitations in Hart & Risley’s work. First, they reported a correlation between early word exposure and IQ at age three but did not report results controlled for SES. Second, although their adult word counts correlated with 36-month IQ scores, turn-taking quantity did not, which is surprising given what is now known about the relationship between parent-child interactions and cognitive development.1,39 However, Hart & Risley’s “parent behavioral turns” differed from LENA’s conversational turns metric and included both verbal and nonverbal responses. Third, although additional research on a subset of the original Hart & Risley sample found measures of child language production (e.g., MLU and child vocabulary use) correlated with developmental outcomes in elementary school, analyses on early adult word and turn counts were not reported.22 This study fills these gaps, demonstrating significant correlations between turn-taking interactions early in life and cognitive and speech/language skills at 9–13 years of age.
There are several possible reasons CTC is better correlated with long term developmental outcomes than AWC. AWC does not measure alternation and so is more likely to contain overheard speech, whereas CTC requires alternation between vocalizations of the child and adult speakers and so is more likely to include child-directed speech. Further, unlike AWC, CTC incorporates child speech which may be predictive of later developmental outcomes.40 This finding is supported by other reports showing that adult-child turn-taking is more important to early development than is simple word exposure.26,41 Romeo et al. specifically identified a possible neural mechanism, reporting that LENA conversational turns predicted fMRI activation in language areas of the brain for 4- to 6-year-olds, whereas AWC did not.42 Their study represents the first empirical research linking a direct measure of neural functioning to early language environment and supports the current finding that turns are more strongly related to long term outcomes than is simple exposure to adults words. Consequently, we suggest that the long term predictiveness of turn-taking reported here coupled with empirical evidence for its relationship to neural functioning provides strong support for the pivotal roles of the early language environment in healthy cognitive development.
Hart & Risley showed that the early language environment is important in predicting developmental outcomes. But their laborious transcription methods severely limited most clinical applications. Automated analyses from daylong recordings are unquestionably easier to obtain than labor-intensive transcriptions, and the ability to predict long-term IQ and language skills even from a single recording has implications for developmental intervention and prevention programs. Potential issues in a child’s language experience may be identified early; if one or two recordings are completed before the 18-month well-baby visit and impoverished language environments are identified, families could be supported through appropriate intervention.
One limitation to these results is their correlational nature – although we refer to statistical predictiveness, we cannot infer causality. For example, other developmental changes occurring during the 18–24 months period may primarily account for cognitive and language skills later. Another limitation is that although the sample spanned a relatively large range of mother’s education levels, only 10 children were from the lowest SES group. In addition, the sample is not ethnically diverse and includes only monolingual English-speakers, so the generalizability of results to other languages and cultures is unknown.
Conclusion
Our findings support the concept that a child’s early language experience may predict developmental outcomes years later. This study expands on previous research by using an automated system to estimate language experience. Conversational turn-taking between the ages of 18–24 months was highly correlated with later language and cognitive skills. The use of automated recordings in the home language environment provides an objective and relatively non-invasive method for assessing the strengths and weaknesses of a child’s language environment and an opportunity to design individualized family feedback and offer education and support to enhance child development, potentially altering developmental trajectories especially of children living in impoverished language environments.
What’s Known on This Subject:
Previous studies using transcriptions of short recordings have shown that quantity of talk and interaction experienced by infants and toddlers is correlated with early language and cognitive abilities.
What This Study Adds:
Automated estimates of turn-taking interaction with children 18–24 months of age predict their IQ and language skills ten years later, suggesting that a child’s language experience during this relatively narrow early age window may predict later language and cognitive development
Acknowledgments
We gratefully acknowledge the Paul family for their wisdom and philanthropy, without which none of this work would have been possible; the LENA parents and children; the members of the LENA Scientific Advisory board; and the LENA employees past and present who contributed to this study.
Funding Source: All phases of this study were funded by the LENA Research Foundation.
Abbreviations
- SES
Socioeconomic status
- SLP
Speech-Language Pathologist
- WISC-V
Wechsler Intelligence Scale for Children-Fifth Edition
- IQ
Intelligence Quotient
- VCI
Verbal Comprehension Index
- PPVT
Peabody Picture Vocabulary Test
- EVT
Expressive Vocabulary Test
- LENA
Language ENvironment Analysis
- AWC
Adult Word Count
- CTC
Conversational Turn Count
- CV
Child Vocalizations
- CI
Confidence Interval
Footnotes
Financial Disclosure: Three of the authors (Gilkerson, Richards and Russo) are full time employees of the LENA Research Foundation (LRF). LRF is a non-profit 501(c)3 Public Charity committed to conducting sound scientific research over the past decade. Three of the authors are unpaid members of the LRF Scientific Advisor Board (Warren, Oller, Vohr). Advisory board members were paid consultants for Infoture, Inc. 2006–2009. In February 2009, Infoture, Inc. was dissolved and gifted to LRF. Since then, all advisory board members have been unpaid volunteers.
Potential Conflicts of Interest: Three of the authors (Gilkerson, Richards and Russo) are full time employees of the LENA Research Foundation, a 501(c)(3) public charity that developed and distributes the automated approach used to analyze the data described here The salaries of LRF scientists are in no way associated with data analyses or research results. The other three authors (Warren, Oller, Vohr) have no conflicts of interest to disclose.
References
- 1.Hart B, Risley TR. American parenting of language-learning children: Persisting differences in family-child interactions observed in natural home environments. Developmental Psychology. 1992;28(6):1096–1105. [Google Scholar]
- 2.Hart B, Risley TR.Meaningful differences in the everyday experience of young American children. Baltimore, MD: Paul H. Brookes Publishing Co; 1995. [Google Scholar]
- 3.Hoff E The specificity of environmental influence: Socioeconomic status affects early vocabulary development via maternal speech.Child Development. 2003;74(5):1368–1378. [DOI] [PubMed] [Google Scholar]
- 4.Huttenlocher J, Haight W, Bryk A, Seltzer M, Lyons T. Early vocabulary growth: Relation to language input and gender. Developmental Psychology. 1991;27(2):236–248. [Google Scholar]
- 5.Rowe ML. Child-directed speech: Relation to socioeconomic status, knowledge of child development and child vocabulary skill. Journal of Child Language. 2008;35(01):185–205. [DOI] [PubMed] [Google Scholar]
- 6.Landry SH, Smith KE, Swank PR. Responsive parenting: Establishing early foundations for social, communication, and independent problem-solving skills.Developmental Psychology. 2006;42(4):627–642. [DOI] [PubMed] [Google Scholar]
- 7.Xu D, Yapanel U, Gray S, Gilkerson J, Richards JA, Hansen J. Signal processing for young child speech language development Proceedings of the 1st Workshop of Child, Computer and Interaction. 2008. [Google Scholar]
- 8.Suskind D, Leffel KR, Graf E, Hernandez MW, Gunderson EA, Sapolich SG, et al. A parent-directed language intervention for children of low socioeconomic status: A randomized controlled pilot study. Journal of Child Language. 2016;43(2):366–406. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Providence Talks. Providence Talks: Pilot findings & next steps. 2015.
- 10.Zhang Y, Xu X, Jiang F, Gilkerson J, Xu D, Richards JA, et al. Effects of quantitative linguistic feedback to caregivers of young children: A pilot study in China. Communication Disorders Quarterly. 2015;37(1):16–24. [Google Scholar]
- 11.Hoff-Ginsberg E The relation of birth order and socioeconomic status to children’s language experience and language development. Applied Psycholinguistics. 1998;19(04):603–629. [Google Scholar]
- 12.Bruner J Vygotsky: A historical and conceptual perspective In: Wertsch JV, editor. Culture, communication, and cognition: Vygotskian perspectives. Cambridge: Cambridge University Press; 1985. p. 21–34. [Google Scholar]
- 13.Vygotsky LS. Mind in society: The development of higher psychological processes. Cambridge, MA: Harvard University Press; 1978. [Google Scholar]
- 14.Fromkin V, Rodman R, Hyams N.An introduction to language (10th ed). Boston, MA: Cengage Learning; 2013. [Google Scholar]
- 15.Dapretto M, Bjork E. The development of word retrieval abilities in the second year and its relation to early vocabulary growth. Child Development. 2000;71(3):635–648. [DOI] [PubMed] [Google Scholar]
- 16.Berk LE.Child Development. 6th ed Boston, MA: Allyn & Bacon; 2003. [Google Scholar]
- 17.Slobin DI.Universals of grammatical development in children. (Working Paper No 22). Berkeley, CA: University of California 1969. [Google Scholar]
- 18.Dore J Conditions for the acquisition of speech acts In: Markova I, editor.The social context of language. Hoboken, NJ: John Wiley & Sons; 1978. p. 87–111. [Google Scholar]
- 19.Nazzi T, Bertoncini J. Before and after the vocabulary spurt: Two modes of word acquisition? Developmental Science. 2003;6(2):136–142. [Google Scholar]
- 20.McMurray B Defusing the childhood vocabulary explosion. science. 2007;317(5838):631. [DOI] [PubMed] [Google Scholar]
- 21.Ganger J, Brent MR. Reexamining the vocabulary spurt. Developmental Psychology. 2004;40(4):621–632. [DOI] [PubMed] [Google Scholar]
- 22.Walker D, Greenwood C, Hart B, Carta J. Prediction of school outcomes based on early language production and socioeconomic factors. Child Development. 1994;65(2):606–621. [PubMed] [Google Scholar]
- 23.Gilkerson J, Richards JA, Warren SF, Montgomery JK, Greenwood CR, Oller DK, et al. Mapping the early language environment using all-day recordings and automated analysis. American Journal of Speech-Language Pathology. 2017;26(2):248–265. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Romeo RR, Leonard JA, Robinson ST, West MR, Mackey AP, Rowe ML, et al. Beyond the “30 Million Word Gap:” Children’s conversational exposure is associated with language-related brain function. psychological science. 2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Xu D, Richards JA, Gilkerson J. Automated analysis of child phonetic production using naturalistic recordings. Journal of Speech, Language, and Hearing Research 2014;57(5):1638–1650. [DOI] [PubMed] [Google Scholar]
- 26.Zimmerman FJ, Gilkerson J, Richards JA, Christakis DA, Xu D, Gray S, et al. Teaching by listening: The importance of adult-child conversations to language development. Pediatrics. 2009;124(1):342–349. [DOI] [PubMed] [Google Scholar]
- 27.Oller DK, Niyogi P, Gray S, Richards JA, Gilkerson J, Xu D, et al. Automated vocal analysis of naturalistic recordings from children with autism, language delay, and typical development. Proceedings of the National Academy of Sciences. 2010;107(30):13354–13359. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Xu D, Yapanel U, Gray S, Baer CT. The LENA language environment analysis system: The Interpreted Time Segments (ITS) file Technical Report No LTR-04–2; 2008; Boulder, CO: LENA Research Foundation. [Google Scholar]
- 29.Bzoch KR, League R, Brown VL. Receptive-Expressive Emergent Language Test. 3rd ed Austin, TX: Pro-ed; 2003. [Google Scholar]
- 30.Zimmerman IL, Steiner VG, Pond RE. Preschool Language Scale. 4th ed San Antonio, TX: The Psychological Corporation; 2002. [Google Scholar]
- 31.Fenson L, Marchman VA, Thal DJ, Dale PS, Reznick JS, Bates E. MacArthur-Bates Communicative Development Inventories: User’s guide and technical manual 2nd ed Baltimore, MD: Paul H. Brookes Publishing Co.; 2007. [Google Scholar]
- 32.Wechsler D Wechsler intelligence scale for children. 5th ed Bloomington, MN: NCS Pearson, Inc; 2014. [Google Scholar]
- 33.Dunn LM, Dunn DM.Peabody picture vocabulary test. 4th ed Bloomington, MN: NCS Pearson, Inc; 2007. [Google Scholar]
- 34.Williams KT.EVT-2: Expressive vocabularly test. 2nd ed Bloomington, MN: PsychCorp; 2007. [Google Scholar]
- 35.Knudsen EI, Heckman JJ, Cameron JL, Shonkoff JP. Economic, neurobiological, and behavioral perspectives on building America’s future workforce. Proceedings of the National Academy of 33 Sciences. 2006;103(27):10155–10162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Kuhl PK. Early language acquisition: Cracking the speech code. Nature Reviews Neuroscience. 2004;5(11):831–843. [DOI] [PubMed] [Google Scholar]
- 37.National Scientific Council on the Developing Child. The timing and quality of early experiences combine to shape brain architecture. (Working Paper No 5). 2007. URL: http://developingchild.harvard.edu/index.php/resources/reports_and_working_papers/working_papers/wp5/. Accessed March 27, 2018.
- 38.Bornstein M Sensitive periods in development: Interdisciplinary perspectives. Mahwah, NJ: Erlbaum; 1987. [Google Scholar]
- 39.Warren SF, Brady NC. The role of maternal responsivity in the development of children with intellectual disabilities. Mental Retardation and Developmental Disabilities Research Reviews. 2007;13(4):330–338. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Ribot KM, Hoff E, Burridge A. Language use contributes to expressive language growth: Evidence from bilingual children. Child Development. 2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Ambrose SE, VanDam M, Moeller MP. Linguistic input, electronic media, and communication outcomes of toddlers with hearing loss. Ear and Hearing. 2014;35(2):139–147. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Romeo RR, Leonard JA, Robinson ST, West MR, Mackey AP, Rowe ML, et al. Beyond the “30 Million Word Gap:” Children’s conversational exposure is associated with language-related brain function Psychological Science. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]