Leslie Rescorla
Bryn Mawr College,
Bryn Mawr, PA Purpose: This study examined whether late talkers identified at 2431 months
continued to have weaker language and reading skills at 17 years of age than
typically developing peers.
Method: Language and reading outcomes at 17 years of age were examined in
26 children identified as late talkers with normal nonverbal ability and normal
receptive language at intake and in 23 typically developing children matched at
intake on age, socioeconomic status (SES), and nonverbal ability.
Results: Although late talkers performed in the average range on all language and
reading tasks at 17 years of age, they obtained significantly lower Vocabulary/
Grammar and Verbal Memory factor scores than SES-matched peers. The age 17
Vocabulary/Grammar factor had large correlations with the age 17 Verbal Memory
and Reading/Writing factors. The age 17 Vocabulary/Grammar and Reading/
Writing factors were strongly predicted by comparable factors at 13 years of age.
Age 2 Language Development Survey ( L. Rescorla, 1989) vocabulary score
explained 17% of the variance in the age 17 Vocabulary/Grammar and Verbal
Memory factors.
Conclusions: Results suggest that slow language development at 2431 months is
associated with a weakness in language-related skills into adolescence relative to
skills manifested by typically developing peersfindings that are consistent with a
dimensional perspective on language delay.
KEY WORDS: late talkers, adolescent outcomes, early language delay
C
hildren with normal nonverbal cognitive ability, adequate hear-
ing, and typical personality development who have delayed
receptive and /or expressive language skills are generally
referred to as having specific language impairment (SLI; Bishop, 2006;
Leonard, 1998.). When such children are identified at 23 years of age,
they have often been referred to as late talkers, a term that is more de-
scriptive and less diagnostic in nature. One reason this appellation de-
veloped was because of a belief held by many that SLI cannot be reliably
diagnosed at 2 years of age. However, the same criteria are used for
diagnosing a late talker and a preschooler with SLInamely a sig-
nificant language delay in the absence of a more primary condition. A
major justification used for making a diagnostic distinction between
these two groups has been that late talkers appear to manifest a better
outcome than preschoolers with SLI (Paul, 1996; Rescorla & Lee, 2000;
Whitehurst & Fischel, 1994). Evaluation of the merits of this diagnostic
16 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009 D American Speech-Language-Hearing Association
1092-4388/09/5201-0016
distinction may depend on whether one espouses a cat- Children with more severe and prolonged language de-
egorical or a dimensional theoretical perspective on lays have weaker skills in a wider variety of language-
language delay. related abilities and hence fall further from the mean on
the impaired side of the distribution than children
whose delays are milder or more short-lived. Toddlers
Contrasting Theoretical Perspectives who have receptive as well as expressive delays and tod-
on Language Delay dlers whose expressive delays last long enough for them
to be diagnosed with SLI at 4 years of age are both pre-
Viewing late talkers and preschoolers with SLI as sumed to fall further away from the mean on this ability
manifesting two distinct disorders suggests a categorical spectrum than late talkers with no receptive problems
perspective toward language delay whereby the two con- whose delays resolve by 4 years of age.
ditions represent qualitatively different disorders hav-
Family environment factors and speech-language
ing different etiologies and outcomes. The alternative
therapy clearly affect language skills (Girolametto, Pearce,
dimensional theoretical formulation postulates a spec-
& Weitzman, 1996; Hart & Risley, 1995). Furthermore,
trum of language ability whereby late talkers and pre-
changes in language input in the home or provision of
schoolers with SLI differ along a quantitative dimension,
speech-language therapy (in addition to the effects of
with both groups having weaker language ability than
time) may result in many preschoolers with SLI as well
typically developing children.
as many late-talking toddlers moving into the average
Researchers who take the categorical approach to range of language skills as they get older. Nonetheless,
language impairment (Gopnik & Crago, 1991; Rice, 2003; the dimensional approach predicts that preschoolers with
Rice & Wexler, 1996) emphasize as central to defining SLI SLI or late talkers are very likely to continue to have
such discrete and focal grammatical deficits as the ex- weaker language skills than comparison peers who never
tended optional infinitive. Research based in this categor- manifested language delay because of what is presumed
ical approach tends to look for clinical markers and specific to be their more compromised language endowment. Out-
genetic causes for what is conceptualized as a discrete come studies of early language delay therefore provide a
disorder (Bishop, 2006). As Dollaghan (2004) noted, the- crucial test for determining whether this prediction from
orizing that SLI constitutes a discrete category promotes the dimensional approach is empirically supported by
research aimed at specifying a unique phenotype, longitudinal evidence.
etiology, base rate, and treatment regimen (p. 464).
In contrast, the dimensional theoretical formulation
postulates a spectrum of language ability. Early ac-
Outcomes for Children With SLI
counts of this dimensional perspective were offered by Although many children diagnosed with SLI as pre-
Bishop and Edmundson (1987), who argued for a spec- schoolers have persisting language impairment and later
trum of impairment for language problems, and by learning problems, many do not (Aram, Ekelman, &
Leonard (1987, 1991), who argued that SLI represents Nation, 1984; Aram & Hall, 1989; Bishop & Adams, 1990;
the tail of a normal distribution of language abilities. Hall & Tomblin, 1978). Despite the fact that a significant
Dollaghan (2004, p. 464) noted that the dimensional the- percentage of children with SLI move into the average
oretical approach implies the potential for more het- range over time, many outcome studies of children with
erogeneity in symptoms, origins, and causal influences SLI indicate considerable stability in individual differ-
than does the categorical approach. Bishop (2006, p. 220) ences in language skills. Some representative studies
used findingssuch as the finding that one monozygotic are summarized below.
twin may be diagnosable with SLI, whereas her co-twin The study of preschool SLI with the most compre-
may have only subclinical language problemsto argue hensive follow-up data is that of Bishop and colleagues
that investigations of SLI should be looking for dimen- (Bishop & Edmundson, 1987). When the children were
sions of impairment rather than discrete subtypes. seen at age 52 years, 44% of the sample of 68 four-year-olds
The dimensional perspective has also been outlined with SLI were judged to have good outcome (i.e., above
in Rescorla (2002, 2005), in which it was proposed that the 10th percentile in expressive language skills; Bishop
late talkers have below average endowment in a set of & Edmundson, 1987). At age 82, years, this age 5 good
intercorrelated yet diverse language-related abilities, outcome group performed at an average level on all lan-
analogous to differential endowments for intelligence. guage and reading measures (Bishop & Adams, 1990). In
The term endowment conveys the assumption that in- contrast, the 56% of the sample with a poor outcome at
dividual differences in language ability are to some ex- age 5 years (i.e., scores below the 10% percentile) per-
tent constitutionally basedan assumption supported formed significantly worse than controls on 9 out of
by findings from genetic studies, such as those by Dale 11 measures at age 82 years. Not surprisingly, at age
et al. (1998) and Bishop, Price, Dale, and Plomin (2003). 15 years (Stothard, Snowling, Bishop, Chipchase, &
18 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009
scored 85 on the Reynell Expressive Language Scale of Productive Syntax ( IPSyn; Scarborough, 1990), and
at 54 months; and (c) typically developing children, grammar skills at age 5 years as measured by the Pat-
558 children who scored 10th percentile on the CDI at terned Elicitation Syntax Test (Young & Perachio, 1983)
24 months and who scored 85 on the Reynell Expres- collectively explained 35% of the variance in age 8 scores
sive Language Scale at 54 months. Differences between on the Clinical Evaluation of Language Fundamentals
the three groups persisted through fifth grade on the Revised (CELF-R; Semel, Wiig, Secord, & Sabers, 1987).
WoodcockJohnsonRevised ( WJ-R; Woodcock & Age 2 LDS score significantly predicted age 9 reading out-
Johnson, 1989) Picture Vocabulary, LetterWord Iden- come on the WJR (Woodcock & Johnson, 1989), but its
tification, and Memory for Sentences subtests, although beta weight declined once age 5 reading and phonology
the late-talker group scored in the average range on the scores were entered as predictors. These three predictors
first two subtests at all time points. On Memory for Sen- explained 54% of the variance in age 9 reading.
tences, the late talkers scored below the average range Rescorla (2005) reported language and reading out-
at 54 months and at first grade. For all measures, the comes at age 13 years for 28 late talkers and 25 typically
late-talker group performed worst, the typically devel- developing children from the initial intake sample, with
oping group performed best, and the late-bloomer group the two groups matched at intake on age, SES, and non-
performed in between the other two groups. The gaps verbal ability. As a group, late talkers performed in the
between the three groups evident at 54 months did not average range on all standardized language and reading
change significantly through fifth grade. These data tasks at age 13 years. However, they scored significantly
provide strong support for the dimensional perspective lower than SES-matched peers on aggregate measures
of early language delay, with the late bloomers having a of Vocabulary, Grammar, and Verbal Memory, as well as
mild delay and the late talkers having more severe lan- on Reading Comprehension. They were similar to compar-
guage weaknesses, both of which persisted over time. ison peers in Reading Mechanics and Writing. Correlations
Rescorla (2002) reported school-age outcome data among outcome measures were moderately high, suggest-
for 34 late talkers and 25 comparison children matched ing some shared variance. Regression analyses indicated
at intake on age, SES, and norverbal cognitive ability. By that age 2 LDS vocabulary score was a significant predictor
age 6 years (Rescorla, 2002), only 6% of the late talkers of age 13 Vocabulary, Grammar, Verbal Memory, and Read-
had scores on at least two TOLD (Newcomer & Hammill, ing Comprehension. Thus, findings suggested that slow
1988a) subtests below the 10th percentile benchmark. language development at age 2 to 22 years was asso-
Nevertheless, the group means on language measures ciated with a weakness in language-related skills into
were significantly lower than those for typically develop- early adolescence relative to typically developing peers.
ing comparison children. Significant group differences In summary, Rescorlas (2005) age 13 findings are
were found on aggregate Vocabulary score at ages 6, 7, consistent with Stothard et al. (1998), Tomblin et al.
and 8, with Cohens ds of 0.85, 1.53, and 1.05, respec- (2003), Girolametto et al. (2001), Moyle et al. (2007),
tively. Significant group differences on the Grammar Armstrong et al. (2007), and Paul et al. (1997). Findings
aggregate were found at 6 and 8 years of age, with from all of these studies support the dimensional ac-
d values of 0.64 and 0.94. The group difference on the count of early language delay. According to this dimen-
Phonological aggregate was significant at age 6 years sional view, children vary widely in their language
(d = 0.91), as was the group difference on TOLD Sentence ability, from seriously impaired at the lower tail to ex-
Imitation at age 6 years (d = 1.26). Although negligible tremely gifted at the upper tail. For clinical purposes, a
numbers of children in the late-talker group scored below cutpoint on what is essentially a continuous dimension
the 10th percentile on reading measures, significant group may be needed to select children most in need of inter-
differences for the Reading aggregate were found at vention. However, proponents of a dimensional view
ages 8 and 9 (ds = 0.84 and 0.72, respectively). would argue that drawing such a cutpoint imposes a
Rescorla (2002) reported high correlations among dichotomy on what is essentially a continuum. The stud-
the various language and reading outcome measures ies reviewed above indicated that many children with
given from ages 6 to 9 years. The correlations that were clinically significant early language delay had moved
significant at p < .01 (66 of 78) ranged from .33 to .99, into the normal range at follow-up. Nonetheless, consis-
with 32 being large effects (.50) according to Cohen tent with the dimensional view of language delay, children
(1988). These correlations indicate some shared variance in these studies who had early delays continued at follow-
among diverse language and reading measures. Rescorla up to have lower scores on a variety of different language
also presented regression results, with predictors entered measures than children with typical language histories.
in chronological order. Vocabulary score on the Language Children whose follow-up language status was still
Development Survey ( LDS; Rescorla, 1989) at 2 years of significantly impaired had even lower language scores
age, grammar skills at 3 years of age as measured by than children who had recovered, suggesting that their
mean length of utterance ( MLU) and scores on the Index impairment was more long lasting and severe.
20 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009
Table 1. Means, standard deviations, and ranges for intake measures by group.
When Cohens d (Cohen, 1988) was used to calculate the therapy or reading remediation. Because degree of re-
ES for Reynell Expressive Language Scale z score (using mediation was so varied and because information about
the difference between 1.62 and 0.36 divided by the interventions was incomplete, intervention was not an-
standard deviation of the typically developing group alyzed for this study.
only, because of the extreme restriction in range in the
late-talker group), the ES was large (3.96), as would be
expected given the use of Reynell Expressive Language Procedure
Scale score to define the groups. Of the 26 late talkers, 4
Age 17 participants were seen for a 2-hr session.
were 6 months delayed, and 22 were 713 months delayed
Numerous standardized tests were used to assess vocab-
relative to their CA on the Reynell Expressive Language
ulary, grammar, verbal memory, and reading/writing
Scale at intake, with all late talkers scoring at least
skills. Every child was tested by the author, a licensed
1.2 SDs below age expectations (<10th percentile), and
and certified school psychologist. The examiner was not
15 of the 26 scoring 1.5 SDs or more below age ex-
blind to the childrens language history, having seen
pectations (<7th percentile). Finally, the late talkers had
them at regular intervals since age 2 years.
a mean vocabulary of 25 words on Rescorlas (1989) LDS,
in contrast to a mean vocabulary of 235 words for the
comparison group. Mothers of all late talkers reported Measures
fewer than 50 words or no word combinations on the LDS.
Wechsler Adult Intelligence ScaleThird Edition
By age 4 years, 29% of the late talkers had MLU
(WAIS-III; Wechsler, 1997a). Participants ability to de-
scores, and 71% had IPSyn (Scarborough, 1990) scores at
fine words on the WAIS-III Vocabulary subtest was used
least 1.25 SDs below age expectations (Rescorla et al.,
to assess verbal conceptual ability. In addition, the Digit
2000). This indicates that the percentage of the sample
Span subtest was used to assess recall of strings of digits
diagnosable with SLI at age 4 years varied with which of
in forward and backward order. Finally, the Block De-
these two measures was used. By age 6 years, about 17%
sign subtest was given to assess nonverbal ability.
of the late talkers were judged to have SLI on the basis of
continuing significant delays in expressive language Woodcock Johnson Psychoeducational Battery Tests
based on their TOLD (Newcomer & Hammill, 1988a) per- of AchievementThird Edition ( WJ-III; Woodcock,
formance (Rescorla, 2002). Independent of their partici- McGrew, & Mather, 2001). The LetterWord Identifica-
pation in this longitudinal study, about one third of the tion, Reading Fluency, and Writing Fluency subtests were
late talkers received some speech-language therapy dur- administered to assess, respectively, ability to decode
ing the preschool period, and a few participated in pre- words on a list, to read and comprehend short statements
school programs for children with language impairments; under time pressure, and to write short statements using
2 late talkers were in special education classes for chil- target words under time pressure. In addition, to obtain a
dren with language impairments in the early elementary second nonverbal measure, the WJ-III Math Fluency
school years. Some of these children received diagnoses of subtest was given to assess the ability to perform arith-
SLI from their educational agencies, but diagnostic prac- metic calculations under time pressure.
tices varied across school districts, and full diagnostic Comprehensive Assessment of Spoken Language
information was not available for all children. A few (CASL; Carrow-Woolfolk, 1999). The Syntax Construc-
comparison children also received speech and language tion, Sentence Comprehension, Grammatical Judgment,
Table 2. Age 17 language and reading test scores: Means, standard deviations, ranges, and percentages below the 10th percentile by group,
plus Cohens d between groups.
% below % below
Measure M SD Range 10th percentilea M SD Range 10th percentilea d
Vocabulary/Grammar factor
WAIS-III Vocabulary subtest 13.50 2.64 9, 19 0 15.48 2.33 10, 19 0 0.80
CASL Syntax Construction subtest 103.77 14.84 76, 133 4 111.13 10.09 92, 129 0 0.59
CASL Sentence Comprehension subtest 102.88 11.60 84, 122 0 108.13 10.82 87, 127 0 0.47
CASL Grammatical Judgment subtest 103.36 14.06 74, 143 4 114.78 14.58 82, 138 0 0.80
CASL Ambiguous Sentences subtest 111.65 13.29 88, 129 0 119.43 9.33 101, 137 0 0.69
Verbal Memory factor
WAIS-III Digit Span subtest 10.35 3.14 4, 19 0 11.70 2.98 6, 18 0 0.44
WMS-III Logical Memory subtest 9.36 2.60 5, 14 17 11.87 2.05 9, 17 0 1.08
WMS-III Verbal Paired Associates subtest 10.31 3.72 1, 16 15 11.91 2.95 7, 16 0 0.48
Reading/Writing factor
WJ-III LetterWord Identification subtest 104.35 11.60 87, 126 0 110.39 12.32 86, 131 0 0.51
WJ-III Reading Fluency subtest 99.88 11.52 75, 120 12 104.17 14.21 72, 124 9 0.33
WJ-III Writing Fluency subtest 112.12 16.24 64, 133 4 112.96 10.67 93, 128 0 0.06
Note. WAIS-III = Wechsler Adult Intelligence ScaleThird Edition; CASL = Comprehensive Assessment of Spoken Language; WMS-III = Wechsler
Memory ScaleThird Edition; WJ-III = WoodcockJohnson Psychoeducational Battery Tests of AchievementThird Edition.
a
Refers to the percentage of children in each group scoring below the 10th percentile on each measure (<6 or <80).
22 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009
of the variance in that factor, with the following load- The Reading Mechanics aggregate (a = .86) was com-
ings: WAIS-III Digit Span = 0.59, WMS-III Logical Mem- puted by averaging Wechsler Individual Achievement
ory = 0.77, WMS-III Verbal Paired Associates = 0.83. Test (Wechsler, 1992) Reading and Spelling scaled scores
Cronbachs alpha was only moderate for the three-item with scores from the Test of Word Reading Efficiency
Verbal Memory factor (.57). (Torgesen, Wagner, & Rashotte, 1999) and with standard
Finally, three measures were assigned to the scores for miscues and words per minute computed from
Reading/Writing factor ( WJ-III LetterWord Identifi- the Qualitative Reading InventoryII (Leslie & Caldwell,
cation, Reading Fluency, and Writing Fluency). The 1995).
three subtests assigned to the Reading/Writing factor To address the fourth hypothesis, correlations were
accounted for 55% of the variance in that factor, with the calculated between LDS score, Reynell Expressive Lan-
following factor loadings: WJ-III LetterWord Identifi- guage Scale and Reynell Receptive Language Scale scores,
cation = 0.69, WJ-III Reading Fluency = 0.76, and WJ-III and the three age 17 factor score outcomes. In addi-
Writing Fluency = 0.78. As with the Verbal Memory fac- tion, Bayley Nonverbal score was tested as a predictor
tor, Cronbachs alpha was only moderate for the three- to examine whether age 2 nonverbal cognitive ability
item Reading/Writing factor (.60). also predicted age 17 outcomes. Follow-up regressions
Before conducting the main analyses, t tests on were then run to examine how much unique variance in
WAIS-III Block Design and WJ-III Math Fluency scores age 17 outcomes were accounted for by these four age 2
were used to determine whether the groups differed predictors.
significantly on nonverbal measures. To address the
first hypothesis, differences between the late-talker
and comparison groups were tested using t tests on the Results
Vocabulary/Grammar, Verbal Memory, and Reading/
Writing factors. Because 2 late talkers were not given Group Differences at Age 17 Years
the WMS-III Logical Memory subtest, they would have Late talkers and comparison children did not differ
been excluded from all follow-up analyses had a multi- significantly on either WAIS-III Block Design or WJ-III
variate analysis of variance been used. Because five sep- Math Fluency. Both groups scored somewhat above
arate t tests were run, p < .01 was used as the significance average on Block Design, with the late talkers having
level. slightly lower scores ( late talkers: M = 12.04, SD = 2.63;
To address the second hypothesis, correlations among typically developing: M = 13.39, SD = 3.35), t(47) = 1.58,
the age 17 Vocabulary/Grammar, Verbal Memory, and p = .12. On Math Fluency, both groups scored in the
Reading/Writing factorsas well as among all 11 age 17 average range ( late talkers: M = 102.73, SD = 16.24;
language-related measureswere obtained. typically developing: M = 101.91, SD = 12.31), t(47) = 0.19,
To address the third hypothesis, four of the age 13 p = .85. Thus, as had been the case 15 years earlier, these
aggregates analyzed by Rescorla (2005) were used as two demographically matched groups of children were
predictors of the three age 17 outcomes in hierarchical comparable in their nonverbal cognitive abilities.
regressions. As described in Rescorlas study, the age 13 As can be seen in Table 2, the late talkers generally
Vocabulary aggregate (Cronbachs a = .77) was computed obtained scores in the average range or above (i.e., scores
by averaging scores on the Wechsler Intelligence Scale 8 or 90) on the language and reading/writing mea-
for ChildrenThird Edition ( WISC-III; Wechsler, 1991) sures administered. However, on many of these mea-
Vocabulary subtest and the Test of Adolescent and Adult sures, late talkers scored somewhat below comparison
LanguageThird Edition ( TOAL-3; Hammill, Brown, children. To measure the magnitude of these differences,
Larsen, & Wiederholt, 1994) Listening Vocabulary and Cohens (1988) d was calculated by subtracting the mean
Reading Vocabulary subtests. The age 13 Grammar ag- score for the late-talker group from the mean score for
gregate (a = .71) was computed by averaging scores for the the comparison group and dividing by the pooled SD for
TOAL-3 Listening Grammar and Reading Grammar sub- the sample. As shown in Table 2, the largest ds were
tests, the Clinical Evaluation of Language Fundamentals found for WAIS-III Vocabulary (d = 0.80), CASL Gram-
Third Edition (CELF-III; Semel, Wiig, & Secord, 1995) matical Judgment (d = 0.80), and WMS-III Logical Mem-
Formulated Sentences subtest, and the Test of Language ory (d = 1.08). Table 2 also shows the percentage of
Competence (TLC; Wiig & Secord, 1989) Ambiguous Sen- children in each group obtaining scores below the 10th
tences subtest. The Verbal Memory aggregate (a = .74) percentile (scores 6 or 80). These percentages indicate
was composed of the WISC-III Digit Span subtest, the that few late talkers scored below the 10th percentile on
CELF-III Recalling Sentences subtest, and a Pseudoword any of the age 17 language and reading outcome mea-
Repetition task modified by Scarborough (1990) from a sures. The most challenging tasks for the late talkers
procedure reported by Taylor, Lean, and Schwartz (1989). appeared to be WMS-III Logical Memory and WMS-III
Table 3. Correlations among age 17 language, memory, and reading/writing outcome measures.
Measure 1 2 3 4 5 6 7 8 9 10 11
1. WAIS-III Vocabulary subtest .54** .32* .56** .64** .30* .18 .41** .68** .39** .36*
2. CASL Syntax Construction subtest .46** .49** .57** .41** .25 .49** .37** .36** .43**
3. CASL Sentence Comprehension subtest .47** .48** .36* .26 .43** .31* .05 .21
4. CASL Grammatical Judgment subtest .69** .51** .31* .47** .49** .19 .28
5. CASL Ambiguous Sentences subtest .54** .35* .56** .52** .21 .41*
6. WAIS-III Digit Span subtest .18 .30* .44** .03 .19
7. WMS-III Logical Memory subtest .46** .02 .35* .20
8. WMS-III Verbal Paired Associates subtest .25 .05 .35*
9. WJ-III LetterWord Identification subtest .28 .32*
10. WJ-III Reading Fluency subtest .40*
11. WJ-III Writing Fluency subtest
Note. Values in bold grouped together indicate correlations among tests from the same factor.
*p < .05. **p < .01.
24 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009
Table 4. Predicting age 17 Vocabulary/Grammar factor scores from rs < .15). In summary, these results indicate that the skills
age 13 Vocabulary, Grammar, Verbal Memory, and Reading tapped by age 13 CELF-III Recalling Sentences and
Comprehension scores. Pseudoword Repetition differed from those tapped by
age 17 WMS-III Logical Memory and WMS-III Verbal
Predictor B SE B b p Paired Associates, but these latter age 17 Verbal Memory
Vocabulary 13 0.17 0.05 .45 .001
tests were significantly associated with age 13 TOAL-3
Grammar 13 0.16 0.07 .33 .026 Listening Grammar and age 13 CELF-III Formulated
Verbal Memory 13 0.08 0.05 .18 .155 Sentences.
Reading 13 0.00 0.01 .05 .733 When the four age 13 aggregates were used to pre-
2 dict the age 17 Reading/Writing factor, the model ac-
Note. R = .67, p < .001.
counted for 67% of the variance ( p < .001; see Table 6).
Only the age 13 Reading Comprehension aggregate had
a significant standardized beta (p < .001), indicating good
When the four age 13 aggregates were used to pre- convergent and discriminant predictive validity.
dict the age 17 Verbal Memory factor, the model ac-
counted for only 42% of the variance (p < .001; see Table 5).
Contrary to prediction, only the age 13 Grammar aggre-
Prediction From Age 2 to Age 17 Years
gate had a significant standardized beta ( p < .05), in- Correlations among the three age 2 language mea-
dicating poor convergent and discriminant predictive sures were all large by Cohens (1988) standards: LDS
validity. To further explore prediction of age 17 Verbal Reynell Expressive (r = .89, p < .001), LDS Reynell Re-
Memory, correlations between the individual tests mak- ceptive (r = .59, p < .001), and Reynell Expressive
ing up this factor and age 13 subtests were examined. All Reynell Receptive (r = .69, p < .001). None of the three
four measures comprising the age 13 Grammar aggre- intake language measures were significantly correlated
gate as well as all three tests comprising the age 13 with Bayley Nonverbal score (all rs < .10). None of the
WISC-III Verbal Memory aggregate were significantly four intake measures were significantly correlated with
correlated with age 17 WAIS-III Digit Span: TOAL-3 Hollingshead SES scores (which had restricted range),
Listening Grammar (r = .55, p < .001); TOAL-3 Reading and SES was not correlated with any age 17 outcome, so
Grammar (r = .61, p < .001); CELF-III Formulated Sen- SES was not further analyzed.
tences (r = .41, p < .05); TLC Ambiguous Sentences (r = .45, All four age 2 predictors were significantly correlated
p < .01); WISC-III Digit Span (r = .72, p < .001); CELF-III with the age 17 Vocabulary/Grammar factor: LDS (r = .43,
Recalling Sentences (r = .49, p < .01); and Pseudoword p < .01), Reynell Expressive (r = .45, p < .01), Reynell
Repetition (r = .59, p < .001). Age 13 TOAL-3 Listening Receptive (r = .38, p < .01), and Bayley Nonverbal (r = .42,
Grammar and age 13 CELF-III Formulated Sentences, p < .01). They were also all significantly correlated with
both in the age 13 Grammar aggregate, were also sig- the age 17 Verbal Memory factor: LDS (r = .41, p < .01),
nificantly correlated with age 17 WMS-III Logical Memory Reynell Expressive (r = .37, p <.05), Reynell Receptive
(rs = .58 and .39, ps = .001 and .03) and WMS-III Verbal (r = .36, p < .05), and Bayley Nonverbal (r = .41, p < .01).
Paired Associates (rs = .52 and .41, ps < .01 and < .05). On These predictive associations were further analyzed using
the other hand, age 13 CELF-III Recalling Sentences and stepwise multiple regressions. For both regressions, the
Pseudoword Repetition, both in the age 13 Verbal Memory order of entry was LDS score, Reynell Expressive Lan-
aggregate, were not significantly correlated with WMS-III guage score, Reynell Receptive Language score, and Bayley
Logical Memory or WMS-III Verbal Paired Associates (all Nonverbal score. Because none of the correlations between
Table 5. Predicting age 17 Verbal Memory factor scores from Table 6. Predicting age 17 Reading/Writing factor scores from
age 13 Vocabulary, Grammar, Verbal Memory, and Reading age 13 Vocabulary, Grammar, Verbal Memory, and Reading
Comprehension scores. Comprehension scores.
Predictor B SE B b p Predictor B SE B b p
Vocabulary 13 0.03 0.06 .07 .694 Vocabulary 13 0.07 0.05 .19 .143
Grammar 13 0.24 0.09 .51 .011 Grammar 13 0.05 0.07 .10 .492
Verbal Memory 13 0.05 0.07 .11 .509 Verbal Memory 13 0.03 0.05 .08 .501
Reading 13 0.00 0.02 .03 .857 Reading 13 0.05 0.01 .58 .000
26 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009
Digit Span, which was highly correlated from age 13 to endowment), the argument advanced here is that dif-
age 17 years. ferential endowment accounts in large part for the con-
Finally, parent-reported vocabulary score on the siderable stability of individual differences in language
LDS significantly predicted Vocabulary/Grammar and skills as well as IQ across development.
Verbal Memory skills 15 years later, accounting for 17% According to the argument advanced here, children
of the variance in both factors. Once the variance as- who are very well endowed on most of these skills will
sociated with the LDS had been taken into account, the have superior or advanced language ability, whereas chil-
Reynell Expressive and Receptive Language Scales did dren with weaker endowment of many of these skills will
not account for a significant amount of variance in either have delayed language acquisition initially and weaker
of these age 17 outcomes. Thus, significant continuity in language skills than their typically developing peers as
language functioning over 15 years was found in this they mature. Children with the weakest endowment on
study, even when age 2 language skills were measured these skills will be delayed in language acquisition and
by a simple parent-report checklist that takes about will continue to manifest clinical deficits in language
10 min to complete. functioning into adolescence and probably beyond. Fur-
An unexpected finding that emerged from these re- thermore, it is postulated that late talkers have some-
gression analyses was that age 2 Bayley Nonverbal score what below-average ability on these constitutionally
was also a significant predictor of age 17 Vocabulary/ based language-related abilities. Toddlers who have re-
Grammar and Verbal Memory skills, accounting for 13% ceptive as well as expressive delays and toddlers whose
and 11% of the variance, respectively, beyond that ac- expressive delays last long enough for them to be diag-
counted for by the LDS. This suggests that some of the nosed with SLI at age 4 years are both further away from
variance in age 17 Vocabulary/Grammar and Verbal the mean on this ability spectrum than late talkers with
Memory skills was due to nonlanguage skills measur- no receptive problems whose delays resolve by age 4 years.
able at age 2 years. These nonlanguage skills include the However, even these less impaired late talkers manifest
various nonverbal cognitive abilities tapped by Bayley significant differences in language skills up to age 17 years
items involving blocks, puzzles, pegs, drawing, and ob- relative to peers with typical language histories.
ject hiding, as well more general test-taking skills, such The fact that the LDS was a significant predictor
as attention, cooperation, and persistence. of age 17 Vocabulary/Grammar and Verbal Memory fac-
In summary, the results of this study are generally tor scores supports the notion that the LDS at age
consistent with the notion that language ability consti- 2431 months is an excellent early measure of the hy-
tutes a dimensional spectrum along which individuals pothesized general language ability. However, contrary
differ. Although the methods of the present study did not to findings reported at 13 years of age, intake Bayley
allow disentangling nature versus nurture factors, the Nonverbal score was also a significant predictor for both
general argument proposed here is that these individual of these age 17 factors. This suggests that by the time the
differences in language ability derive in large part from participants in this study reached late adolescence,
constitutionally based variation in endowment. On the some nonverbal ability factors that the Bayley tapped at
basis of the analogy with intelligence, it can be argued 2431 months also began to account for a significant
that this constitutionally based variation in endowment percentage of variance in the skills tapped by the age 17
derives from variation in many discrete skills subserving Vocabulary/Grammar and Verbal Memory measures.
language. Although the present study does not directly The age 2 language measures used in this study
indicate what this set of distinct yet interrelated skills were parent-report or clinical instruments tapping major
might be, it might be conjectured that likely skills are au- language features, such as vocabulary size, use of word
ditory perception, word retrieval, verbal working memory, combinations, and comprehension of words and phrases.
motor planning, phonological discrimination, and gram- They were not experimental measures designed to tap
matical rule learning. This notion of a spectrum of lan- the language-related skills presumed to underlie language
guage ability is consistent with Leonards (1991) argument ability, such as auditory perception, word retrieval, verbal
that SLI represents the tail of a normal distribution of working memory, motor planning, phonological discrimi-
language abilities, as well as with Bishop and Edmundsons nation, or grammatical rule learning. However, such
(1987) formulation of a spectrum of impairment. abilities are increasingly being measured in toddlers
The argument that children are born with differing and tested as predictors of later language outcomes. For
endowments for language is similar to the argument example, Newman, Ratner, Jusczyk, and Jusczyk (2006)
that children are born with differential endowments for reported that the ability to segment speech in laboratory
intelligence. Although environmental factors clearly af- tasks administered before 12 months of age was as-
fect both language development and IQ (and environ- sociated with vocabulary size at 24 months and with
mental factors also covary to some degree with genetic scores on language tests at ages 46 years.
This study suggests that most late talkers with Bishop, D. V. M., Price, T. S., Dale, P. S., & Plomin, R.
(2003). Outcomes of early language delay: II. Etiology of
normal receptive language skills will perform in the av-
transient and persistent language difficulties. Journal of
erage range by national standards in most language Speech, Language, and Hearing Research, 46, 561575.
skills by the time they enter school and will continue to
Carrow-Woolfolk, E. (1999). Comprehensive Assessment of
do so through adolescence. On the other hand, the find- Spoken Language. Circle Pines, MN: American Guidance
ings suggest that early expressive language delay may Service.
indicate some subclinical weakness in the component Cohen, J. (1988). Statistical power analysis for the behavioral
skills that serve language. Continuing weaknesses in sciences (2nd ed.). New York: Academic Press.
vocabulary, grammar, and verbal memory may be evi-
Dale, P. S., Simonoff, E., Dishop, D. V. M., Eley, T. C.,
dent in adolescence, relative to peers with typical lan- Oliver, B., Price, T. S., et al. (1998). Genetic influences
guage histories from the same backgrounds. Although on language delay in two-year-old children. Nature Neuro-
this study did not assess intervention effects, it is pos- science, 1, 324328.
sible that providing late talkers with activities to im- Dollaghan, C. A. (2004). Taxometric analyses of specific
prove language processing, phonological discrimination, language impairment in 3- and 4-year-old children. Jour-
verbal memory, and word retrieval might help to pro- nal of Speech, Language, and Hearing Research, 47,
464475.
mote stronger language skills and thereby reduce or
eliminate the gap between them and peers with typical Fenson, L., Reznick, S., Thal, D., Bates, E., Hartung, J.,
language histories. Pethick, S., & Reilly, J. (1993). The MacArthur Commu-
nicative Developmental Inventory. San Diego, CA: Singular
Publishing Group.
Acknowledgments Girolametto, L., Pearce, P. S., & Weitzman, E. (1996). In-
teractive focused stimulation for toddlers with expressive
This research was supported by grants from the Bryn Mawr vocabulary delays. Journal of Speech and Hearing Research,
College Faculty Research Fund, National Institute of Child 39, 12741283.
Health and Human Development Grant 1-R15-HD22355-01, Girolametto, L., Wiigs, M., Smyth, R., Weitzman, E., &
and National Institute on Deafness and Other Communication Pearce, P. S. (2001). Children with a history of expressive
Disorders Grant R01-DC00807. I wish to thank the parents language delay: Outcomes at 5 years of age. American
and children whose participation made this research possible. Journal of Speech-Language Pathology, 10, 358369.
Gopnik, M., & Crago, M. B. (1991). Familial aggregation of
a developmental language disorder. Cognition, 39, 150.
References
Hall, K., & Tomblin, J. B. (1978). A follow-up study of chil-
Aram, D., Ekelman, B., & Nation, J. (1984). Preschoolers dren with articulation and language disorders. Journal of
with language disorders: 10 years later. Journal of Speech Speech and Hearing Disorders, 43, 227241.
and Hearing Research, 27, 232244. Hammill, D. D., Brown, V. L., Larsen, S. C., & Wiederholt,
Aram, D., & Hall, N. E. (1989). Longitudinal follow-up of J. L. (1994). Test of Adolescent and Adult LanguageThird
children with study of the relationship between specific Edition. Austin, TX: Pro-Ed.
28 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009
Hart, B., & Risley, T. R. (1995). Meaningful differences in the Rescorla, L., & Lee, E. C. (2000). Language impairments in
everyday experiences of young American children. Baltimore: young children. In T. Layton & L. Watson ( Eds.), Handbook
Brookes. of early language impairment in children: Volume 1: Nature
Hollingshead, A. (1975). Four-Factor Index of Social Status ( pp. 138). New York: Delmar Publishing Company.
[ Working paper]. New Haven, CT: Author. Reynell, J. K. (1977). Reynell Developmental Language
Lee, L. (1974). Developmental sentence analysis: A grammat- Scales. Windsor, England: NferNelson.
ical assessment procedure for speech and language clini- Reynell, J. K., & Gruber, C. P. (1990). Reynell Developmental
cians. Evanston, IL: Northwestern University Press. Language Scales, U.S. edition. Los Angeles: Western Psy-
Leonard, L. B. (1987). Is specific language impairment a use- chological Services.
ful construct? In S. Rosenberg ( Ed.), Advances in applied Rice, M. L. (2003). A unified model of specific and general
psycholinguistics: Disorders of first language development language delay: Grammatical tense as a clinical marker
(Vol. 1, pp. 139). New York: Cambridge University Press. for unexpected variation. In Y. Levt & J. Schaeffer ( Eds.),
Leonard, L. B. (1991). Specific language impairment as a Language competence across populations: Toward a defini-
clinical category. Language, Speech, and Hearing Services tion of specific language impairment ( pp. 6395). Mahwah,
in Schools, 22, 6668. NJ: Erlbaum.
Leonard, L. B. (1998). Children with specific language im- Rice, M. L., & Wexler, K. (1996). Toward tense as a clinical
pairment. Cambridge, MA: MIT Press. marker of specific language impairment in English-speaking
Leslie, L., & Caldwell, J. (1995). Qualitative Reading children. Journal of Speech and Hearing Research, 39,
InventoryII. New York: Harper Collins. 12391257.
Moyle, M. J., Ellis Weismer, S., Lindstrom, M., & Evans, Scarborough, H. S. (1990). Index of Productive Syntax. Ap-
J. (2007). Longitudinal relationships between lexical and plied Psycholinguistics, 11, 112.
grammatical development in typical and late talking chil- Semel, E., Wiig, E., & Secord, W. (1995). Clinical Evaluation
dren. Journal of Speech, Language, and Hearing Research, of Language FundamentalsThird Edition. San Antonio,
50, 508528. TX: The Psychological Corporation.
Newcomer, P. L., & Hammill, D. D. (1988a). Test of Lan- Semel, E., Wiig, E., Secord, W., & Sabers, D. (1987).
guage Development. Austin, TX: Pro-Ed. Clinical Evaluation of Language FundamentalsRevised.
Newcomer, P. L., & Hammill, D. D. (1988b). Test of Lan- San Antonio, TX: The Psychological Corporation.
guage DevelopmentPrimary, Second Edition. Austin,
Stothard, S. E., Snowling, M. J., Bishop, D. V. M., Chipchase,
TX: Pro-Ed.
B. B., & Kaplan, C. A. (1998). Language-impaired pre-
Newcomer, P. L., & Hammill, D. D. (1997). Test of Language schoolers: A follow-up into adolescence. Journal of Speech,
DevelopmentPrimary, Third Edition. Austin, TX: Pro-Ed. Language, and Hearing Research, 41, 407418.
Newman, R., Ratner, N. B., Jusczyk, A. M., & Jusczyk, Taylor, H. G., Lean, D., & Schwartz, S. (1989). Pseudoword
P. W. (2006). Infants early ability to segment the conver- repetition ability in learning-disabled children. Applied
sational speech signal predicts later language development: Psycholinguistics, 10, 203219.
A retrospective analysis. Developmental Psychology, 42,
643655. Tomblin, J. B., Zhang, X., Buckwalter, P., & OBrien, M.
(2003). The stability of primary language disorder: Four
Paul, R. (1993). Outcomes of early expressive language years after kindergarten diagnosis. Journal of Speech,
delay. Journal of Childhood Communication Disorders, Language, and Hearing Research, 46, 12831296.
15, 714.
Torgesen, J., Wagner, R., & Rashotte, C. (1999). Test of
Paul, R. (1996). Clinical implications of the natural history Word Reading Efficiency ( TOWRE). Austin, TX: Pro-Ed.
of slow expressive language development. American Journal
of Speech-Language Pathology, 5, 521. Wechsler, D. (1991). Wechsler Intelligence Scale for Children
Third Edition. San Antonio, TX: The Psychological
Paul, R., Murray, C., Clancy, K., & Andrews, D. (1997). Corporation.
Reading and metaphonological outcomes in late talkers.
Journal of Speech, Language, and Hearing Research, 40, Wechsler, D. (1992). Wechsler Individual Achievement Test.
10371047. San Antonio, TX: The Psychological Corporation.
Rescorla, L. (1989). The Language Development Survey: A Wechsler, D. (1997a). Wechsler Adult Intelligence Scale
screening tool for delayed language in toddlers. Journal of Third Edition. San Antonio, TX: The Psychological
Speech and Hearing Disorders, 54, 587599. Corporation.
Rescorla, L. (2002). Language and reading outcomes to age 9 Wechsler, D. (1997b). Wechsler Memory ScaleThird Edition.
in late-talking toddlers. Journal of Speech, Language, and San Antonio, TX: The Psychological Corporation.
Hearing Research, 45, 360371. Whitehurst, G. J., & Fischel, J. E. (1994). Early develop-
Rescorla, L. (2005). Age 13 language and reading outcomes in mental language delay: What, if anything, should the
late-talking toddlers. Journal of Speech, Language, and clinician do about it? Journal of Child Psychology and
Hearing Research, 48, 459472. Psychiatry, 35, 613648.
Rescorla, L., Dahlsgaard, K., & Roberts, J. (2000). Late- Wiig, W. H., & Secord, W. (1989). Test of Language Competence
talking toddlers: MLU and IPSyn outcomes at 3;0 and 4;0. Expanded Edition. San Antonio, TX: The Psychological
Journal of Child Language, 27, 643664. Corporation.
30 Journal of Speech, Language, and Hearing Research Vol. 52 1630 February 2009