DEVELOPMENTAL DISABILITIESRESEARCH REVIEWS 17:141-150(2011)

LATE TALKERS: D O GOOD PREDICTORS OYOUTCOME EXIST?

Leslie Rescorla*Department of Psychology, Bryn Mawr College, Bryn Mawr, Philadelphia

Both small-scale and epidemiological longitudinal studies of earlylanguage delay indicate that most late talkers attain language scores in theaverage range by age 5, 5, or 7. However, late talker groups typically obtainsignificantly lower scores than groups with typical language histories onmost language measures into adolescence. These findings support adimensional account of language delay, whereby late talkers and typicallydeveloping peers differ quantitatively on a hypothetical language abilityspectrum. Variation in language ability is presumed to derive from variationin skills subserving language, such as auditory perception/processing, wordretrieval, verbal working memory, motor planning, phonological discrimi-nation, and grammatical rule learning. Expressive language screening at18-35 months can serve an important public health function by identifyingchildren whose expressive delay is secondary to autism spectrum disorder,intellectual disability, hearing impairment, receptive language delay, or de-mographic risk. Finally, the review suggests that demographic risk associ-ated with low SES may become more important as a causal factor inlanguage delay as children get older. © 2013 Wiley Periodicals, inc.Dev Disabil Res Rev 2011; 17:141-150.

Key words: late talkers, outcomes, predictors

Adelay in expressive language is one of the most com-mon reasons that young children are referred forevaluation [Whitehurst and Fischel, 1994; Rescorla

and Lee, 2000]. As noted by the US Preventive Services TaskForce [2006], "Speech and language delay affects 5% to 8% ofpreschool children, often persists into the school years, andmay be associated with lowered school performance and psy-chosocial problems" (p. 497).

Because expressive language delay, like a fever, is asymptom found in many conditions, children who are slow totalk are a heterogeneous group. Expressive language delay isoften secondary to another disorder [Whitehurst and Fischel,1994; Rescorla and Lee, 2000]. Late talkers may have a vocaltract malformation or a hearing loss that interferes with lan-guage development. Other young children with delayedlanguage may have a neurological disorder or general intellec-tual disability (ID). StiU other children who are slow to talkhave autism spectrum disorder (ASD), in which case delayedlanguage is accompanied by deficits in social relatedness, play,and behavior. Children may also be slow to talk because theyexperience severe deprivation, neglect, or abuse. We hence-forth use the term late talkers to refer to children 18- to 35-month-old who are slow to talk in the absence of any of theseother conditions presumed to be primary. Some late talkershave an expressive delay only, whereas others are delayed inreceptive language as well.

© 2013 Wiley Periodicals, inc.

Although distinct diagnostic groups within the populationof young children manifesting expressive language delay are easilydelineated, their relative percentages are less well documented.One study that did address this issue was conducted by Busch-mann et al. [2008], who studied 100 German 2-year-olds withdelayed expressive language identified in pediatric practices. Ofthese 100 children (65% boys), 78 were late talkers, 18 had lan-guage delay associated with cognitive impairment, and four hadautism. These three groups did not differ in gender, birth posi-tion, or maternal education. Family history of languageimpainnent was reported for 40% of the language-delayed groupbut for only 4% of a typically developing group. Of the 78 latetalkers, 61 had an expressive delay only, whereas 17 had recep-tive/expressive delay. The late talkers with receptive/expressivedelay had lower nonverbal IQs than the typically developing chil-dren, whereas the late talkers with expressive delays only did not.

For children whose language delay is secondary to ASD orID, developmental progress tends to be slow and language maynever reach nonnal levels. Outcomes also vary among late talkers.The purpose of this article is to review the literature on outcomesof late talkers with the aim of determining what, if any, variablesare reliable predictors of outcome. We wul review two kinds ofstudies: (a) small-scale longitudinal studies of late talker and com-parison samples and (b) large-scale epidemiological studies ofyoung children. The review encompasses children identified withlanguage delays between 18 and 35 months of age. In the small-scale studies, late talkers are typically a weU-defined group thatexcludes children with hearing impairment, neurological disor-ders, ASD, and ID, and, in some studies, even children withreceptive language delays. In the large-scale epidemiological stud-ies, children whose language delay is secondary to a more pdmarycondition may not be excluded or even differentiated.

OUTCOMES FROM SMALL-SCALELONGITUDINAL STUDIES OF LATE TALKERS

Preschool Outcome Studies of Late TalkersIn one of the first studies of late talkers, Fischel et al. [1989]

reported on 22 children identified with specific expressive

*Correspondence to: Leslie Rescorla, Bryn Mawr College, 101 N. Merion Avenue,Bryn Mawr, PA 19010. E-mail: Irescorl@brynmawr.eduReceived 11 SepteEiiber 2012; Accepted 5 October 2012View this article online in Wiley Online Library (wileyonlinelibrary.com)DOI: 10.1002/ddrr.1108

language delay at 24-38 months. The latetalkers scored more than two standarddeviadons (SDs) below average on theExpressive One-Word Picture Vocabu-lary Test [EOPVT; Gardner, 1981] andhad mean vocabulades of <20 words.Five months later, 35% of the group hadscores in the average range on theEOPVT (>85), with outcome predictedby vocabulary size and proportion of con-sonants to vowels at intake. By age ^'^/z,88% scored above 85 on the EOPVT;95% were in this range by 5'/2 [White-hurst etaL, 1992].

Thai et al. [1991] reported thatfour of 10 late talkers idendfied at age 2who were still delayed 1 year later (i.e.,the "truly delayed") had also beendelayed at age 2 in recepdve languageand recognitory gestures. In contrast, thesix "late bloomers" had been comparableto typically developing compadson chil-dren in receptive language and had madeextensive use of gestures to compensatefor communication. For a differentsample. Thai et al. [2005] reportedage 4 outcomes of 20 children whoscored <10th percentile in expressivevocabulary on the CommunicativeDevelopment Inventory [CDI; Fensonet al., 1993] at 16 months. Although thelate talker group scored in the normalrange on language and cognitive tests,they scored significandy lower than agroup of 44 compadson children withtypical language (TL) histodes. The latetalkers also scored lower than the com-padson children on a nonword repetitiontask, particularly for longer sdmuH.

Feldman et al. [2005] reportedsensitivity of only 50% and positive pre-dicdve value of only 64% whenlanguage delay at age 2 (CDI vocabu-lary scores < 10th percentile) was usedto predict language delay at age 3 for113 children (over 50% firom low-income families). Thus, many childrenwith delayed early vocabulary caughtup by age 3 and many children withapparently normal development at age 2were delayed by age 3. Vocabulary sizeat age 2 was a better predictor of out-come than utterance length or sentencecomplexity score.

Lyytinen et al. [2001] traced lan-guage skills using the CDI at 14, 24, 30,and 42 months in 200 Finnish children,106 from dyslexic famihes (DR), and 94from families with no dyslexia history(NDR). By age 2, the DR group hadshorter maximum sentence length butnot smaller vocabulades than the NDRgroup. By 42 months, the DR group hadsignificantly lower scores on expressivevocabulary and inflectional morphology

142

measures but not on receptive languagemeasures. A subsample of 34 late talkers(20 DR and 14 NDR) was idendfiedbased on scores 1 SD below the mean onage 2 expressive language. By 42 months,only the DR late talkers were sdUdelayed in expressive and recepdve lan-guage. For the full sample of 200children, parental educadon and historyof dyslexia, symbolic play and vocabularycomprehension at 14 months, expressivelanguage at 24 months, and receptivelanguage at 30 months were all signifi-cant predicton of expressive languageoutcome at age y/2 years, explaining48% of the total vadance.

Rescoda et al. [2000] reportedage 3 and age 4 outcomes for 34 latetalkers with normal receptive languageand nonverbal ability idendfied at 24—31 months. At age 3, 41% of the latetalkers scored above the 10th percentileon MLU and 34% scored above the10th percentile on the Index of Pro-ducdve Syntax [IPSyn; Scarborough,1990]. At age 4, the percentages were71 ahd 29%. For the same 34 late talk-ers, Rescoda et al. [1997] reported thatthe percentage scodng in the normalrange at age 3 (>16th percentile) was79% on the EOPVT, 58% on the Rey-nell Expressive Language Scale[ReyneU, 1977], 35% on MLU, and24% on the IPSyn. Thus, granimadcalskills showed the most persistent delays.

Rescoda et al. [2000] used regres-sion analyses to identify significantpredictors of outcome. Expressive z-score accounted for 21-34% of the var-iance in the age 3 outcome measures,whereas ReyneU Recepdve z-scores[Reynell, 1977; ReyneU et al., 1990]and nonverbal abiUty scores were notsignificant predictors. The 34 late talkers(age range, lA—'iX months) had verysimilar raw scores on the ReyneU Ex-pressive Language Scale, but becauseReyneU ^-scores are age-based, the olderchildren had lower ar-scores than theyounger children [Rescoda et al., 1997].Therefore, intake age was correlated at—0.1(3 with ReyneU Expressive ar-score,and both measures significantly predictedoutcomes. Because expressive languageskills in typically developing childrenimprove so rapidly between 24 and 36months, the older a late talker is withinthis time pedod, the more he is faUingbehind on a steeply accelerating curve,as noted by Rescoda et al. [1997].

FinaUy, Femald and Marchman[2012] traced lexical development fiom18 to 30 months in 36 late talkers whoscored <20th percentile on the CDI at18 months and 46 typically developing

children who scored >20th percendle.Mean vocabulary scores were vastly dif-ferent between the two groups at 18months (20 vs. 121 words, d = 1.9),and sdU very different by 30 months(419 vs. 569 words, d = 1.0). By 30months, 14 of the 36 late taUcers stillscored <20th percentile on the CDI(39% = delayed), whereas 22 scoredabove this cutoff (61% = late bloomers).Of the 18 children delayed at 30months, 14 had been late talkers at 18months and four had been typicaUydeveloping (9% of the typical groupwere delayed at 30 months). These out-comes yielded sensidvity of 78%,specificity of 66%, positive predictivevalue of 39%, and negadve predictivevalue of 88%.

Femald and Marchman [2012]reported that 18-month-old late talkenhad lower accuracy and speed scores thantypicaUy developing children on looking-while-listening (LWL), an online prefer-endal looking lexical processing task inthe child listens to sentences such as"where is the doggy?" while a picture ofthe named referent and a foil are flashedon a screen. Faster reacdon dme andgreater accuracy predicted steeper accel-eradon in vocabulary growth in the latetalker group from 18 to 30 months.Adding reacdon dme as a predictorincreased posidve predicdve value to55% versus 39% using CDI score alone.Vocabulary plus accuracy yielded a posi-dve predicdve value of 52%. Femald andMarchman [2012] also noted that March-man and Femald [2008] had reportedfaster processing speed and larger vocabu-lary size at 18 months as predicdve ofbetter working memory at age 8.Although Femald and Marchman [2012]acknowledged that their study could notaddress the degree to which endogenousversus expedendal factors accounted forlexical processing abuity, they cited aprevious study [̂ Hurtado et al., 2008]demonstradng that dchness of matemalspeech to the child at 18 months pre-dicted both vocabulary size and lexicalprocessing skill at 24 months.

SummaryPreschool outcomes from small-

scale longitudinal studies of late talkersindicate that: (a) most children scoredin the nomial range on language testsby age 4 or 5; (b) grammatical delaystended to be more protracted than vo-cabulary delays; and (c) significantpredictors of outcome, which vadedacross studies and left much vadanceunexplained, included recepdve lan-guage, gestures/play skills, degree of

DEV DISABIL RES REV- LATE TALKERS: D O GOOD PREDICTORS OF OUTCOME EXIST? • RESCORLA

delay at 2, consonant repertoire, familyhistory of reading problems, and lexicalprocessing.

School-Age Outcome Studies ofLate Talkers

Girolametto et al. [2001] reportedage 5 outcomes for 21 late talkers iden-tified at 24—33 months with scoresbelow the 5th percentile on the CDI.The children participated in an 11-week parent-based intervention pro-gram at age 2, and 13 of the 21 hadreceived subsequent speech-languageservices. At age 5 foUow-up, most ofthe late talkers scored in the normalrange on various language measures,but they scored significandy lower onmost language measures at age 5 thanchildren with TL histories, particularlyon measures tapping more complexskiUs, such as narrating a story.

Paul et al. studied about 30 latetalkers identified between 20 and 34months and a matched comparisongroup [Paul, 1993, 1996; Paul et al.,1997]. Roughly 25% of the late talkerswere delayed in receptive as well as ex-pressive language. At age 3 and age 4, allof Paul's [1993] late talkers scored in theaverage range for receptive and expres-sive vocabulary and for receptivegrammar. Paul [1996] used a criterion of>10th percentile in Developmental Sen-tence Score [DSS; Lee, 1974] to indicaterecovery, which might be considered arather minimal standard for normal func-tioning. By this criterion, recovery wasachieved by 41% of the sample by age3, 57% by age 4, 74% by kindergartenand first grade, and 84% by second grade[Paul, 1996; Paul et al., 1997]. How-ever, the late talken did more poorlythan the typicaUy developing comparisonchUdren at age 7 on the TOLD-P2[Newcomer and HammiU, 1988] Ex-pressive language scale, even if theywere classified as recovered. The recov-ered children were not different fromcontrols on receptive language, reading,spelling, IQ, or phonological skills at age7, whereas the children who were stüldelayed (DSS < 10th percentUe) wereworse than comparison children oneverything except receptive languageand reading/speUing [Paul et al., 1997].

Moyle et al. [2007] reported age 5outcomes for 30 late talkers first identi-fied at age 2 by scores <10th percentileon the CDI. At age 5, late talkers had sig-nificantly lower scores than comparisonchUdren matched on age, SES, gender,and nonverbal cognitive abihty on threeTOLD-P3 subtests [Newcomer andHammiU, 1997]: Oral Vocabulary,

Grammatic Compledon, and SentenceImitation (Cohen's d values of 0.97,1.46, and 1.52, respectively). EllisWeismer [2007] reported age 5'/2 out-comes for 40 late talkers identified at age2 (11% with comprehension delays) and43 typically developing peers. Only threeof the late talkers scored at least 1 SDbelow the mean on TOLD-P3 SpeakingQuotient at age 5y2. However, evenwith these three children excluded, thelate talkers obtained significandy lowerscores than comparison children on boththe Listening and Speaking Quotients ofthe TOLD-P3. Group differences wereparticularly marked in sentence imita-tion. Performance on a fast mapping taskat 2'/2 explained 36% of the variance inMLU at 3V2 [EUis Weismer, 2007];when CDI and Preschool LanguageScale-3 [PLS-3, Zimmerman et al.,1992] scores were added as predictors,65% of the variance was explained. Aswould be expected, prediction wasweaker to age 51/2, with age 2'/2 PLS-3scores, CDI vocabulary, and nonverbalabihty accounting for 51% of the var-iance in TOLD-P3 scores.

Rescorla [2002] reported school-age outcomes for 34 late talkers and 25comparison children matched at intakeon age, socioeconomic status (SES), andnonverbal cognitive ability. By age 6[Rescorla, 2002], only 6% of the latetalkers had scores <10th percentile onat least two TOLD-2 [Newcomer andHammiU, 1988] subtests. Nevertheless,the late talker group means were signif-icandy lower than those for typicallydeveloping comparison children on vo-cabulary, grammar, phonology, andverbal memory tasks, with mostCohen's ds > 0.85. For the combinedlate talker and comparison group sam-ples, vocabulary score on the LanguageDevelopment Survey [LDS; Rescorla,1989] at age 2, grammar skiUs at age 3as measured by MLU and IPSyn, andgrammar skiUs at 5 as measured by thePatterned Ehcitation Syntax Test[PEST; Young and Perachio, 1983]coUectively explained 35% of the var-iance in age 8 scores on the ChnicalEvaluation of Language Fundamentals[CELF-R; Semel et al., 1987].

Rescorla [2005] reported languageand reading outcomes at age 13 for 28late talkers and 25 typically developingchildren from the initial intake samples.As a group, late talkers performed inthe average range on aU standardizedlanguage and reading tasks at age 13,but they scored significantly lower thanSES-matched peers on vocabulary,grammar, verbal memory, and reading

comprehension. Regression analyses forthe combined groups indicated that age2 LDS vocabulary score explained 20%of the variance in age 13 vocabularyand verbal memory and 14% of the var-iance in age 13 grammar. Althoughmuch variance was left unexplained,these results indicated that slow vocabu-lary development at age 2—21/2 wasassociated with a weakness in language-related skiUs into early adolescence rela-tive to typicaUy developing peers.

Rescorla [2009] reported languageand reading outcomes at 17 years of agefor 26 late talkers and 23 typicaUy devel-oping children matched at intake on age,SES, and nonverbal ability. Although latetalkers perfomied in the average range onaU language and reading tasks at age 17,they obtained significandy lower vocabu-lary, grammar, and verbal memory scoresthan peers with TL histories. The largesteffect size was a Cohen's d of 1.08 onstory recaU. With the late talker and com-parison groups combined, age 2 LDSvocabulary score, entered first, explained17% of the variance in the age 17 vocabu-lary/grammar aggregate; ReyneUExpressive and ReyneU Receptive scores,entered next, were not significant predic-tors. Interestingly, Bayley nonverbalscore, added last, explained an addidonal13% of the variance. Results were similarfor the verbal memory aggregates, withthe LDS accounting for 17% and Bayleynonverbal score accounting for an addi-tional 11% of the variance. Thus, 28-30%of the variance in age 17 language scoreswas explained by two age 2 measures.

SummarySchool-age outcomes firom smaU-

scale longitudinal studies of late talkersindicate that: (a) most late talkers scored inthe normal range by age 6 or 7 but con-tinued to have significandy weakerlanguage skiU than typicaUy developingpeers through adolescence; (b) significantpredictors of outcome, which variedacross studies, included vocabulary at age2, nonverbal IQ, and preschool expressiveand recepdve language, but at best onlyabout half the variance in school-age out-comes was explained in these studies.

OUTCOMES FROM LARGE-SCALE EPIDEMIOLOGICALSTUDIES

Outcomes to 18-24 Months forLarge-Scale EpidemiologicalStudies

The Early Language in VictoriaStudy [ELVS; ReiUy et al., 2007] iden-tified 20% of a community sample of

DEV DISABIL RES REV- LATE TALKERS: D O GOOD PREDICTORS OF OUTCOME EXIST? • RESCORLA 143

1,720 2-year-olds as late talkers, basedon CDI expressive vocabulary scores<10th percentile based on U.S. norms[Fenson et al., 1993]. When gender,preterm birth, birth weight, birth order,SES, maternal mental health, matemalvocabulary and education, maternal age,family history of speech-language diffi-culties, and non-English familybackground were used as predictors ofCDI scores, the model accounted foronly 7% of the variance. When 12-month scores on the Communicationand Symbolic Behavior Scales [CSBS;Wetherby and Prizant, 2002] wereadded to the predictive model, the par-tial R^ was 14%, but most of thevariance in 24 months expressive lan-guage remained unexplained.

Zubrick et al. [2007] assessed lan-guage skiUs at 24 months in a sample of1,766 Austrahan children from English-speaking famihes. Late language emer-gence (LLE) was diagnosed based onparental report obtained by maü on sixreceptive and expressive language itemsof the Communication scale of theAges and Stages Questionnaire [ASQ;Bricker and Squires, 1999] (e.g., point-ing to pictures on request, using two-or three-word phrases). The criterion ofscores >1 SD below the U.S. meanidentified 13% of the sample with LLE(238 children). Children with LLEwere more hkely to also be delayed inASQ Gross Motor (6 vs. 1%), FineMotor (8 vs. 3%), Adaptive (21 vs. 6%),and Personal-Social (8 vs. 1%) skiUsthan children with TL, suggesting thatthe LLE group included children withID, ASD, or other developmental dis-abilities. Multivariate logistic regressionwith LLE as the target outcome indi-cated no significant prediction fi-omparental education or mental health,matemal age, SES, parenting style, orfamily functioning. However, signifi-cant odds ratios (ORs) were obtainedfor family history of LLE (2.1), numberof sibhngs (2.1), male gender (2.7), pre-mature birth (1.8), <85% optimal birthweight (1.9), and delays on concurrentASQ Gross Motor (3.1), Fine Motor(2.4), Adaptive (2.6), and Personal-Social (5.5) scores. Zubrick et al. [2007]concluded that the results are consistentwith a models of language that "positstrong role for neurobiological andgenetic mechanisms" rather than mod-els attributing major influence tomatemal and family characteristics.

The Norwegian Mother andChild Cohort Study [Schjolberg et al.,2011] investigated predictors of lan-guage delay at 18 months for 42,107

children as reported on the ASQ[Bricker and Squires, 1999]. A regres-sion model •with numerous predictors(being a boy, low birth weight or gesta-tional age, multiple birth, older sibhngs,low matemal education, maternal dis-tress/depression, and non-Norwegianlanguage background) explained 4—7%of language outcomes, leaving most ofthe variance in language performance at18 months unexplained.

Henrichs et al. [2011] tested pre-dictors of 18-month scores on the 112-word Dutch version of the MacArthurShort Form Vocabulary Checklist[MCDI-N; Zink and Lejaegere, 2003].Translations were available in Englishand Turkish; Moroccan parents whospoke only Arabic were interviewed athome by Arabic-speaking research assis-tants. Matemal age, parenting stress, andchild ethnicity explained 1% of the var-iance in 18 months expressivevocabulary, gestational age and birthweight explained an additional 1%,gender and age at the 18-month evalua-tion explained an additional 4%, andMCDI-N receptive score at 18 monthsexplained an additional 16%, for 22% intotal.

SummaryLarge-scale epidemiological stud-

ies predicting language outcomes at 18-24 months indicate that: (a) demo-graphic and birth variables accountedfor only a modest percentage of thevariance in outcomes at 18 or 24months and (b) adding in earher orconcurrent developmental variablesaccounted for more variance in out-comes, but most variance was stiUunexplained.

OUTCOMES FROM 16 TO 18MONTHS FOR LARGE-SCALEEPIDEMIOLOGICAL STUDIES

Horwitz et al. [2003] studied1,189 Connecticut children with ele-vated rates of demographic risk factors(e.g., 35% poor, 15% more than onelanguage spoken in the home, and 37%non-white). When language delay wasdefined as a CDI short form expressivevocabulary score <10th percentile, ratesof delay were 12.5% at 18-23 months,15% at 24-29 months, and 18% at 30-35 months. Multivariate logistic regres-sion analysis indicated that languagedelay was associated with demographicrisk, but relative risk ratios were <2.0except for being firom a bilingualhousehold (2.78) (e.g., low matemaleducation = 1.26, poverty = 1.33, mi-nority status = 1.28, number of sibhngs

= 0.83, and parenting stress = 1.61, aUquite smaU effects).

EUis and Thai [2008] summarizedage 6 outcomes for a sample of 577children classified at 16 months intothree groups: 461 typicaUy developingchildren; 81 "late producers" (childrenwith expressive delay only, 14% of thesample); and 35 "late comprehenders"(children with receptive and expressivedelay, 6% of the sample). At age 6,2.2% of the 577 children met criteriafor specific language impairment (SLI):1.5% of the typicaUy developing group,3.7% of the late producers, and 8.5% ofthe late comprehenders. Of the 13 chil-dren diagnosed with SLI at age 6, sevenhad been typicaUy developing, threehad been late producers, and three hadbeen late comprehenders at age 16months. This suggests that early recep-tive/expressive delay confers greater riskfor later SLI than does expressive delayonly, but that more children with SLIat age 6 had TL histories as toddlersthan language delays as toddlers.

Westerlund et al. [2006] assessedlanguage development at 18 monthsand 3 years for children seen at Swedishchild health care centers. Parents of 891children completed the 90-word Swed-ish Communication Screening at 18months [SCSI8; Ericksson et al., 2002],with an expressive delay cutoff of <8words. Language delay at age 3 wasdiagnosed if the child manifested lack ofthree-word sentences or failed to com-prehend three of five comprehensionquestions posed by health center nurses.Only half the children delayed at 3 hadbeen delayed at 18 months (sensitivity= 50%). Most of the children identifiedas delayed at 18 months were notdelayed at age 3 (positive predictivevalue = 18%). Most children notdelayed at 3 had not been delayed at 18months (specificity of 90%). The areaunder the curve (AUC) of 77%obtained via receiver operating charac-teristics (ROC) analyses indicated onlyfair prediction. Increasing the 18-monthSCS18 cutoff to <15 words raised sen-sitivity to 66%, but at the cost ofpositive predictive value of only 10%and specificity of only 73%. Further-more, using words comprehended orgestures produced rather than wordsproduced yielded even poorer predic-tion, and combining the three measuresyielded a lower AUC than using pro-duction alone.

Henrichs et al. [2011] reportedoutcome data at 30 months for 3,759 ofthe Dutch chUdren for whom expres-

144sive and receptive vocabulary delay at

DEV DISABIL RES REV- LATE TALKERS: DO GOOD PREDICTORS OE OUTCOME EXIST? • RESCORLA

18 months was deterimned by MCDIscores <10th percentile. Expressive vo-cabulary delay at 30 months wasdefined as a score <10th percentile onthe Dutch translation of the LDS[Rescorla, 1989]. The correlationbetween MCDI-N expressive scores at18 months and LDS scores at 30months was only 0.34, much lowerthan the concurrent correlation of 0.95between the CDI and the LDS in 2-year-olds [Rescorla et al., 2005].

Of the 3,759 children in theHenrichs et al. [2011] study, 85% hadno expressive vocabulary delay at eitherage, 6% were delayed at 18 months andnot delayed at 30 months (late bloom-ers), 6% had late onset expressivevocabulary delay (not delayed at 18months but delayed at 30 months), and3% had persistent expressive vocabularydelay (delayed at both ages). Most chil-dren (71%) delayed at 18 months onthe MCDI-N scored in the normalrange at 30 months on the LDS (posi-tive predictive value = 29%), and mostchildren (70%) delayed at 30 monthshad not scored below the 10th percen-tile at 18 months (sensitivity = 30%).Most children who scored in the nor-mal range at 18 months continued toscore in the normal range at 30 months(negative predictive value = 93%), andmost children with normal skills at 30months also had normal skills at 18months (specificity = 93%). The ROCcurve using MCDI-N expressive vo-cabulary scores at 18 months to predictLDS delay status at 30 months had anAUC of 0.74 (95% CI: 0.71, 0.77; p =< 0.001), indicating only fair predictiveaccuracy.

Henrichs et al. [2011] reportedthat hierarchical hnear regression analy-sis examining prediction of 30 monthsvocabulary score yielded significant butsmall effects (complete model = 18%).Matemal age and education, marital sta-tus, fan-lily income, child ethnicity, andparenting stress explained 5% of thevariance in LDS vocabulary at 30months; gestational age and birthweight explained an additional 0.2%;gender and age at the 18- and 30-month assessments explained an addi-tional 1%; 18 months MCDI-Nexpressive 2r-scores explained an addi-tional 11%; and 18 months MCDI-Nreceptive 2--scores explained an addi-tional 0.5%. Children in the late onsetgroup and the persistent delay grouptended to have lower fan-iily income,less educated mothers, and more paren-tal stress than children in the no delayand late bloomer groups. Late bloomers

and children with persistent delay hadlower gestational ages and lower birthweights than children with no expres-sive vocabulary delay and late onsetexpressive vocabulary delay. The persis-tent delay group showed the lowestnonverbal scores.

SummaryLarge-scale epidemiological stud-

ies predicting language outcomes from16 to 18 months indicate that: (a) lan-guage status at 16-18 months had lowpositive predictive value for later lan-guage delay; (b) most children delayedat later ages had not been delayed at16—18 months (low sensitivity); (c) de-mographic and birth variablesaccounted for only a modest percentageof the variance in outcomes at 30months; and (d) adding in earher lan-guage status accounted for morevariance in outcomes, but most variancewas StiU unexplained.

OUTCOMES FROM 24 MONTHSIN LARGE-SCALEEPIDEMIOLOGICAL STUDIES

Armstrong et al. [2007] reportedfindings through fifth grade for 689children from the NICHD Early ChildCare Research Network data set. Threegroups were defined based on CDIscores at 24 months and Reynell Ex-pressive Language Scale scores [Reynelland Gruber, 1990] at 36 and 54months: 131 late talkers, who scored<10th percentile on the CDI and <85on the Reynell at 36 and 54 months;39 late bloomers, who scored <10thpercentile on the CDI and >85 on theReynell at 54 months; and 558 typicallydeveloping children, who scored >10thpercentile on the CDI at 24 month and>85 on the Reynell at 54 months. Dif-ferences between the three groupspersisted through fifth grade on theWoodcock-Johnson-Revised [WJ-R;Woodcock and Johnson, 1992] PictureVocabulary, Letter Word Identification,and Memory for Sentences subtests,although the late talker group scored inthe average range on the first two subt-ests at all time points. For all measures,the late talker group performed worst,the typically developing group per-formed best, and the late bloomergroup perfonned in between the othertwo groups, with httle change in thegaps between groups over time.

Outcomes fi-om age 2 to age 4were reported for 1,596 of the childrenin the Austrahan ELVS cohort [Reillyet al., 2010]. Based on a cutpoint of>1.25 SDs below the mean on the

Chnical Evaluation of Language Funda-mentals—Preschool [CELF-P2; Wiiget al., 2006] to identify language delay,13% were delayed in expressive lan-guage and 16% were delayed inreceptive language, with 21% scoringbelow this cutpoint on either scale. Af-ter excluding children with lownonverbal IQs, ASD, hearing impair-ment, and non-English-speakingbackground (NESB), 17% were diag-nosed with SLI based on receptive and/or expressive delay. A multivariateregression model with 12 predictors(male, twin, preterm, low birth weight,older siblings, NESB, low SES, familyhistory of language problems, matemaleducation, maternal mental health prob-lem, matemal age, and matemalvocabulary) explained 19% of the var-iance in CELF-P2 receptive languageand 21% in CELF-P2 expressive lan-guage. Adding in language delay at age2 increased the variance accounted forto 24% (receptive) and 30% (expressive),indicating stronger prediction from age2 to age 4 than firom 12 months to age2 but StiU much unexplained variance.The strongest predictor of delayed ex-pressive language at 4 was a foreignlanguage home background, with anOR of 6.96. Being a boy (OR = 1.90),being a fourth child (OR = 2.36), andfan-lily history of language problems(OR = 1.82) were the next best predic-tors of delayed expressive language.

Dale et al. [2003] identified latetalkers in a sample of UK twins at age2 based on expressive vocabulary scores<10th percentile (<15 words) on the100-word MacArthur CommunicativeDevelopment Inventory UK ShortForm [MCDI: UKSF; Dionne et al.,2003]. All data were collected via par-ent reports sent by mail. Afterexcluding children with uncertain zygo-sity, genetic syndromes, ASD, and non-Enghsh-speaking famihes, 802 of the8,386 children (9.6%) were identifiedwith early language delay (ELD). Com-pared with the 7,584 children with TL,the age 2 ELD group had more boys(65 vs. 47%) and lower nonverbal abil-ity, grammar, and "displaced reference"scores. Their mothers also had lowereducational attainment. However, onaU these measures, there was consider-able variation within the ELD group.The largest effect size was for nonverbalabihty (5.7%).

When a score <15th percentileon two of three language measures (vo-cabulary, grammar, and abstractlanguage) was used to identify ELD atage 3, 11% of the sample was delayed

DEV DISABIL RES REV- LATE TALKERS: DO GOOD PREDICTORS OF OUTCOME EXIST? • RESCORLA 145

(835 children), 61% of whom had notbeen delayed at 2; 44% of the 2-year-olds with ELD were still delayed at 3.When the same criterion was used atage 4, 12% of the sample was delayedin language (746 children), only 34% ofwhom had been delayed at 2; 40% ofthe 2-year-olds with ELD were stüldelayed at 4.

Dale et al. [2003] used logisticregression analyses to determine deci-sion statistics for prediction from age 2to ages 3 and 4. Age 2 vocabulary, dis-placed reference, and nonverbal abüityscores as well as gender and maternaleducation were used in a predictionmodel for age 3 and age 4 delay for theage 2 children with ELD. Sensitivitywas 42% at age 3 and 52% at age 4,meaning that half or fewer of the chil-dren delayed at foUow-up werepredicted to be delayed based on themodel. Similarly, only 57% of the chil-dren predicted to be delayed by themodel at age 3 (and 64% at age 4) wereactually delayed at those ages (positivepredictive value). When the full age 2sample was used, sensitivity was verypoor (19% at both ages), as was positivepredictive value (53% at 3 and 58% at4). In sum, language delay at age 2 anda number of additional factors poorlypredicted language delay at a later age,more than half of children with ELD at2 were in the normal range at ages 3 or4, and most children with languagedelay at ages 3 and 4 had normal lan-guage at age 2.

Because the Dale et al. [2003]sample was comprised of twins, the rel-ative contributions of genetic andenvironmental factors associated withlanguage delay could be tested. Bishopet al. [2003] concluded that sharedenvironmental factors accounted formore variance in age 2 ELD (71-72%)than did genes (22-25%), regardless ofwhether the children had a persistent ora transient delay as measured by parentreport at ages 3 and 4.

Rice et al. [2008] reported age 7outcome data for children with LLEand children with normal languageemergence (NLE) fi-om the Zubricket al. [2007] study. At age 2, 128 chil-dren had expressive language delay(19% of the sample), 88 of whom alsohad receptive delays at age 2. Exclusioncriteria included ID, ASD, deafness.Down syndrome, and cerebral palsy. Atage 7, the children with LLE did notdiffer firom the NLE group on SES/de-mographic variables or on nonverbalinteUigenee, and they scored in the nor-mal range on all language measures

146

given. However, they had significantlylower scores than the NLE group onreceptive vocabulary, articulation, andnumerous grammatical scales, with thelargest Cohen's ds on the morphosyntaxmeasures. The percentages of LLE chil-dren who were impaired (—1 SDbelow the mean) ranged fi-om 4 to 23%across the 17 outcome measures, withsignificant group differences for sevenof the measures.

SummaryLarge-scale epidemiological stud-

ies predicting language outcomes firom24 months indicate that: (a) most latetalkers scored in the nomial range byage 6 or 7 but continued to have signif-icantly weaker language skuls thantypicaDy developing peers; (b) few sig-nificant predictors of outcome werefound, and (c) positive predictive valuefirom age 2 was generaUy low.

CONCLUSIONS

Outcomes of Late TalkersAlthough several studies have

reported outcomes for late talkers iden-tified between 18 and 35 months ofage, findings vary as a flinction of ageat intake, age at foUow-up, compositionof the sample, and outcome measuresused. For example, late talkers identifiedat 18 months tend to have smaller per-centages of persistent delay by age 3than late talkers identified at 24 months,as seen in rates of 18% in Westerlundet al. [2006] and 29% in Henrichs et al.[2011] versus 44% in Dale et al. [2003].Furthermore, the older late talkers areat foUow-up, the larger the percentagewho wul be recovered. For example,the percentage of Paul's late talkerswho scored in the normal range forsyntax increased from 41% at age 3 to84% at age 7 [Paul, 1996]. With respectto sample composition, EUis and Thai[2008] reported that 8.5% of late talkerswith slow comprehension had persistentdelay firom 16 months to age 6 versusonly 3.7% of those with Fischel normalcomprehension. FinaUy, vocabularymeasures tend to show better outcomesthan grammar measures, and grammarmeasures vary in their degree of strin-gency. For example, the percentage ofRescorla's 34 late talkers scoring >16thpercentile was 79% on the EOPVT,58% on the ReyneU Expressive Lan-guage Scale, 35% on MLU, and 24%on the IPSyn [Rescorla et al., 1997].

Despite the variations in out-comes reported across both smaU-scaleand epidemiological studies of ELD, a

robust finding across many studies isthat most late talkers attained languagescores in the average range by age 5, 6,or 7 [Fischel et al, 1989; Paul et al.,1997; Girolametto et al., 2001;Rescorla, 2002, 2005, 2009; EUisWeismer, 2007; EDis and Thal, 2008;Rice et al., 2008]. Furthermore, mostlate talkers scored in the average rangeby elementary school even on the tasksthat appear to be the most chaUengingfor them, namely grammatical andverbal memory measures [Rescorla,2002; Rice et al., 2008].

Another very robust finding fromoutcome studies is that late talkergroups consistently obtained signifi-cantly lower scores than groups withTL histories on most language meas-ures, even when the late talkersperformed in the average range. Paul[1996] and Rescorla [2002] wereamong the first to report this finding,which was rephcated in Moyle et al.[2007], EUis Weismer [2007], Rescorla[2005, 2009], Thai et al. [2005], Arm-strong et al. [2007], and Rice et al.[2008]. This finding provides supportfor the dimensional account of languagedelay, according to which late talkersand children with TL development dif-fer quantitatively on a hypotheticallanguage abihty spectrum [Rescorla,Fischel, 2009].

As articulated by Rescorla [2002,2005, 2009], the language abihty spec-trum, like inteUigenee, can beconceptualized as deriving from varia-tion in many discrete skiUs. The distinctyet interrelated abihties hypothesized tosubserve language include auditory per-ception/processing, word retrieval,verbal working memory, motor plan-ning, phonological discrimination, andgrammatical rule leaming. This notionof a speetmm of language abihty is con-sistent with Leonard [1991] and Bishopand Edmundson [1987]. EUis Weismer[2007] has related the notion of a lan-guage endowment speetmm to thebroader theoretical debate regardingcontinuous versus dichotomous charac-terizations of language impairment.

Success in Predicting Late TalkerOutcomes

When predicting language delayas a dichotomous outcome, both falsepositive and false negatives have beencommon. The false positive rate for latetalkers is often very high (i.e., late talkerscoring in the nomial range at out-come), as indicated by generaUy lowpositive predictive values, such as 18%in Westerlund et al. [2006]; 29% in

DEV DISABIL RES REV- LATE TALKERS: DO GOOD PREDICTORS OF OUTCOME EXIST? • RESCORLA

Hendchs et al. [2011]; 44% (age 3) and40% (age 4) in Dale et al. [2003]; 64%in Feldman et al. [2005]; and 39% inFernald and Marchman [2012]. Becausemost late talker outcome studies do notreport intervention infomiadon, it isdifficult to determine the extent towhich low positive predictive values aredue to treatment success versus sponta-neous remission.

High rates of false negative errorshave also been widely reported inlarge-scale epideniiological late talkerstudies, in that the majodty of childrenwith language delays at foUow-up hadTL development at intake. Althoughmost children with typical developmentat intake continue to have normal lan-guage skiUs at foUow-up [e.g., negativepredictive value of 93% in Hendchset al., 2011], children who were notlate talkers at intake generally compdsethe majodty of children delayed at out-come, e.g., 70% in Hendchs et al.[2011], 54% in EUis and Thai [2008],50% in Westedund et al. [2006], and61% (age 3) and 66% (age 4) in Daleet al. [2003]. However, this pattern hasnot generally been found in the smaU-scale late talker studies [e.g., Paul, 1996;EUis Weismer, 2007; Rescorla, 2009],where very few children in the typicallydeveloping compadson groups mani-fested language delays at any foUow-upage. The reason for this differencebetween epideniiological and smaU-scalelate talker studies is that the small-scalestudies selected typicaUy developingcompadson groups to have solidly aver-age or even above average languageskiUs, rather than just being above a10% percentile cutoff. The only small-scale study in which many childrenwith delays at 3 had been typicallydeveloping at age 2 was Feldman et al.[2005], which used a CDI score >10thpercentile to define typical developmentat 2 and which contained many chil-dren from low SES families, known tobe at dsk for language delays withincreasing age [Hart and Risley, 1995].

EUis Weismer [2007] has high-lighted the puzzMng mismatch betweenthe smaU percentage of late talkers whomanifest SLI at age 5 and the 7% ofkindergarten children identified withSLI [TombHn et al., 1997]. As EllisWeismer so cogendy states, "Given thereladvely low proportion of late talkerswho display clinical language impair-ment at school entry, we must continueto ask where those 7% of kindergartenchildren with SLI come fiom if notfrom the ranks of late talkers" (p. 95).This apparent paradox should be a focus

for fijrther research, which can best beexplored in diverse epideniiologicalsamples that are being foUowed from 18or 24 months, such as the Hendchset al. [2011] Dutch sample or the Reillyet al. [2007] Australian sample.

As Hendchs et al. [2011] noted,poor decision statistics may dedve inpart from imposing a fixed cutpoint onan underlying continuum, wherebychildren just missing the cutpoint (i.e.,at the 11th percentile) are classified asnormal. However, Hendchs et al.[2011] also noted that the poor predic-don in their sample was not onlyattdbutable to dichotoniization, becausethe correlation between 18-monthMCDI-N and 30-nionth LDS was only0.34, and ROC analysis, which tests allcutpoints in a continuous fashion,yielded an AUC of only 74%.

Prediction of outcomes can alsoutihze continuous scores. Multipleregression analyses with smaU-scalestudies involving late talker and typi-cally developing groups have generallyyielded moderately strong prediction.For example, EUis Weismer found that30-month fast mapping performance,CDI vocabulary, and PLS scoresexplained 65% of the vadance in age 5language outcome. Rescorla [2009]reported that 30% of the vadance inage 17 vocabulary/grammar scores waspredicted by LDS score and nonverbalability at age 2. Fernald and Marchman[2012] reported that faster reaction timeand greater accuracy in a lexical proc-essing task at 18 months predictedsteeper acceleration in vocabularygrowth among late talkers fiom 18 to30 months. In contrast, regression anal-yses with epideniiological samples havebeen able to account for rather smaUpercentages of vadance in outcomemeasures. For example, Hendchs et al.[2011] found that only 22% of the var-iance in 18 months expressivevocabulary score was explained (6% bydemographic/perinatal factors and 16%by concurrent receptive vocabularyscores). Reilly et al.'s [2007] model pre-dicted only 14% of the vadance in CDIscores at 24 months (7% by demo-graphic factors and 7% by 12-monthcommunication scores on the CSBS).Zubdck et al. [2007], using logisdcregression to predict LLE versus NLE at24 months, reported that demographicfactors were not significant predictorsand that the highest ORs were for con-current ASQ scores.

In summary, the likely reason thatregression results are stronger for smaU-scale studies spanning many years (i.e..

15 years for the Rescorla age 17 results)than for epideniiological studies span-ning short dme pedods is that thesmaU-scale studies, by design, hadsharply contrasdng groups. The latetalkers were selected to have severedelays (generaUy 1.5 SDs below ageexpectations) and the compadson chil-dren were selected to have solidlyaverage or even above average languageskiUs. Therefore, even though the latetalkers eventually performed in the av-erage range, they continued to haveweaker skills than the compadson chil-dren and the vadables that inidaUydifferentiated the two groups continuedto differentiate them over time. Bycontrast, the epideniiological samplesrepresented the fuU spectrum of lan-guage skiUs, with the late talkers at thetail of a continuous distribution (e.g.,the bottom 10%) and the rest of thesample considered "normal." Further-more, the epidemiological samplestended to be much more diverse interms of SES than the smaU-samplestudies, and the follow-ups to date havebeen quite short-term. It may be thecase that when diverse samples such asthe Dutch, UK, and Australian samplesare foUowed up into the elementaryschool years, family SES factors willplay an increasingly important role inaccoundng for vadation in languageskiUs.

Predictors of OutcomeAlthough several studies have

tested predictors of outcome for latetalkers, results have been inconsistent.For example, although Bishop et al.[2003] reported that shared environ-ment impacted language delay andHorwitz et al. [2003] indicated strongdemographic effects on language delay,two large Australian studies found thata host of environi-nental factorsexplained very little vadance in lan-guage skiUs [ReiUy et al., 2007;Zubdck et al., 2007]. It would seemthat family income, marital status, eth-nicity, parendng stress, and matemaleducational level should be among themost important factors for explainingindividual differences in language devel-opment. However, across most studiesreviewed, their predicdve abuity hasbeen quite modest. As noted above,longer temi foUow-ups in these largeepidemiological samples are likely toindicate increasing effects of SES withage.

Reproductive factors such as ges-tadonal age, birth weight, and pednatalcomplications have also explained very

DEV DISABIL RES REV- LATE TALKERS: DO GOOD PREDICTORS OF OUTCOME EXIST? • RESCORLA 147

little variance in later language skiUs,consistent with many earher studies ofperinatal risk. This is probably because,as proposed by Sameroff and Chandler[1975], the "continuum of caretakercasualty" is more potent than the"continuum of reproductive casualty"in determining developmental out-comes. As Sameroff and Chandlernoted, transacdonal effects betweencaregivers and children serve to magnifythe effects of reproductive risk in moredisadvantaged famihes but to minimizethese effects in higher SES families.

FaniUy history of language delaysignificandy increased the odds of beinga late talker at 2 in the Zubrick et al.[2007] sample (OR = 2.1) and at 4 inthe Reilly et al. [2010] sample (OR =1.8), and history of language delay inthe family has also been commonlyreported in smaU-scale late talker stud-ies. Lyytinen et al. [2001] found thatgenetic risk for dyslexia was stronglyassociated with persistent delay in latetalkers identified at 2. Although thesefindings suggest that ELD may have amajor genetic component. Bishop et al.[2003] noted that the magnitude of thegenetic effect was very dependent onthe particular outcome measure used todetermine language delay as weU as onthe cohort analyzed.

Being a boy is one of the bestpredictors of being a late talker in epi-demiological studies. SmaU-scale latetalker samples typicaUy have manymore boys than girls, and late talkers inepidemiological studies also are moreoften boys than girls. Because girls typi-caUy have larger reported vocabulariesthan boys, using the same delay crite-rion for both genders (i.e., <50 wordsor <10th percentUe on the CDI) wiUalmost always yield more boys thangirls. If a gender-specific cutpoint isused (i.e., 10th percentile on the CDIby gender), then genders wiU be equalin the late talker group, but the meanvocabulary score of the girl late talkerswiU be higher than the score of the boylate talkers.

The best predictors of expressivelanguage appear to be earher expressivelanguage skiUs. This was found in bothsmaU-scale studies, such as those byRescoda 2002, 2005, 2009] and EUisWeismer [2007], as yveU as in epidemi-ological studies [Lyytinen et al., 2001;ReiUy et al., 2007, 2010; Henrichset al., 2011]. For example, Henrichset al. [2011] reported that the strongestpredictor of 30 months expressivevocabulary was 18 months expressivevocabulary, which accounted for 11%

of the variance. Clearly, however,much variance was left unexplained.

An excidng possibility in recentlate talker research is that some of theunexplained variance in late talker out-comes may be accounted for bylanguage processing skiUs. As Femaldand Marchman [2012] and Marchmanand Femald [2008] have shown, reac-tion time and accuracy in a lexicalprocessing task at 18 months signifi-candy predicted later vocabularygrowth in both typicaUy developingchildren and late talken, as weU as pre-dicting age 8 working memory. Thesefindings are noteworthy, because verbalmemory deficits are among the mostrobust and enduring weaknesses mani-fested by late talkers, even when theyperform in the normal range at foUow-up. Femald and Marchman [2012]acknowledged that variation in lexicalprocessing could be due to endogenousfactors, but they also highlighted thatvariation in matemal speech to childrenhas been demonstrated to predict laterlexical processing skiU at 24 months. Itis hkely that children vary in theirinborn auditory processing abihty, butthat this abihty is also shaped by subse-quent language environment, such thatchUdren with richer language environ-ments develop in their abihty toefficiendy process lexical informationmore than children with less rich lan-guage environments.

ImplicationsWhen expressive language delay

is secondary to a more primary disabil-ity, such as autism or ID, then thelikelihood of continuing delay is highand the benefits of early interventionare substantial. AdditionaUy, when ex-pressive language delay is the result ofsignificant environmental neglect orabuse, it is important to intervene inorder to protect the child and provide amore supportive and stimulating envi-ronment. To the extent that childrenwith ID, ASD, or environmental depri-vation would not be otherwiseidentified, picking them up in an ex-pressive language screening would havesome pubhc health benefit. One mightalso make this argument for late talkerswho are delayed in receptive language,given that receptive/expressive delay isoften associated with nonverbal deficitsand generaUy presages a poorer out-come than expressive delay alone.

On the other hand, childrenwhose only developmental disability at18-35 months is an expressive languagedelay are at less risk, as this review indi-

cates. Most late talkers catch up tonormative expectations, although someso not do so untU age 5 or later andtheir language skiUs continue to beweaker than those of their peers, on av-erage. Furthermore, the existinghterature strongly suggests that mostchildren with language delays at age 5were not late talkers. Therefore, earlyidentification and intervention with latetalkers wiU not have much impact onpreventing SLI in school-age children.

An implication of this review isthat expressive language screening at 18-35 months can serve an important pubhchealth function because it can identifychUdren whose expressive delay is sec-ondary to another disabihty. This isparticularly important for children withreceptive language delay and weak non-verbal skiUs but who do not have ID orASD and so might not be identified byother screening procedures. ChUdrenwith receptive delays are fairly hkely tohave continuing language delay, as weUas being at risk for later leaming and be-havioral/emotional problems. Therefore,they are the late talken most in need ofearly intervention.

Although not providing definitiveevidence, the current review suggeststhat demographic risk associated withlow SES may become more importantas a causal factor in language delay aschUdren get older. In contrast, individ-ual differences in basic languageendowment seem more important as aprimary causal factor in language delayfrom 18 to 35 months. To be sure,chUdren from very disadvantaged fami-hes, or from famihes where more thanone language is spoken, may showdelays in this 18- to 35-month period,and the more such famihes there are ina large, diverse sample, the more hkelydemographic factors are to emerge assignificant predictors.

In summary, language is a com-plex set of skiUs. A large number ofbiopsychosocial factors are most likelyresponsible for individual differences inthe skiUs that subserve language abihty.Early in development, biological factorsmay have the most important influence,as indicated by the famUial aggregationof language problems. However, inborncharacteristics are influenced by envi-ronmental factors. Therefore, by thetime children start talking, their capacityfor language learning reflects both theirbiological endowment and their linguis-tic environment. As chUdren enter thepreschool period, psychosocial factorssuch as how much (and how weU)

148parents talk to them probably become

DEV DISABIL RES REV. LATE TALKERS: DO GOOD PREDICTORS OE OUTCOME EXIST? • RESCORLA

increasingly important. Thus, childrenwho manifest SLI at age 5 most likelyrepresent a mixture of children, somewho have been delayed since they weretoddlers, as weU as some who are atpsychosocial risk. Children with SLI at5 typicaUy have poorly developed lis-tening skiUs, limited vocabularies, weakgrammatical skiUs, and poorly devel-oped higher level language abilities suchas defining, describing, and narrating.

REFERENCESAmistrong ES, Marchman VA, Owen MT. 2007.

School age cognitive and achievement out-comes for late talkers and late bloomers: dolate bloomers really bloom? Poster presentedat American Speech-Hearing AssociadonAnnual Conference, Boston, November2007.

Bishop DVM, Edmundson A. 1987. Languageimpaired 4-year-olds: Disdnguishing tran-sient from persistent impairment. J SpeechHear Disord 52:156-173.

Bishop DV, Price TS, Dale PS, et al. 2003. Out-comes of early language delay. II. Etiologyof transient and persistent language difFicul-des. J Speech Lang Hear Res 46:561—575.

Bricker D, Squires J. 1999. Ages and StagesQuestionnaires: a parent-completed, child-monitoring system, 2nd ed. Baldmore: PaulH. Brookes.

Buschmann A, Jooss B, Rupp A, et al. 2008.Children with developmental language delayat 24 months of age: results of a diagnosdcwork-up. Dev Med Child Neurol 50:223-229.

Dale PS, Price TS, Bishop DV, et al. 2003. Out-comes of early language delay. 1. Predicdngpersistent and transient language difFiculdesat 3 and 4 years. J Speech Lang Hear Res46:544-560.

Dionne G, Dale PS, Boivin M, et al. 2003.Genedc evidence for bidirecdonal effects ofearly lexical and grammatical development.Child Dev 74:391-412.

EUis EM, Thai DJ. 2008. Early language delayand risk for language impainnent. Perspec-dves 15:93-100.

EUis Weismer S. 2007. Typical talken, late talkers,and children with specific language impair-ment: a language endowment spectrum? In:Paul R, editor. The influence of develop-mental perspectives on research and pracdcein communication disorders: a festschrift forRobin S. Chapman. Mahwah, NJ: LawrenceErlbaum Associates, pp. 83-101.

Ericksson M, Westerkund M, Berglumd E. 2002.A screening version of the Swedish Com-municadve Development Inventoriesdesigned for use with 18-month-old chil-dren. J Speech Lang Hear Res 45:948-960.

Feldman HM, Dale PS, CampbeU TF, et al.2005. Concurrent and predicdve validity ofparent reports of child language at ages 2and 3 years. Child Dev 76:856-868.

Fenson L, Dale PS, Reznick JS, et al. 1993. TheMacArthur Communicadve DevelopmentInventories: user's guide and technical man-ual. Baltimore: Paul H. Brookes Publishing.

Femald A, Marchman VA. 2012. Individual dif-ferences in lexical processing at 18 monthspredict vocabulary growth in typicaUydeveloping and late-talking toddlers. ChildDev 83:203-222.

FischelJ, Whitehurst G, Caulfield M, et al. 1989.Language growth in children with expressivelanguage delay. Pediatrics 82:218-227.

Gardner MF. 1981. Expressive one-word picturevocabulary test. Novato, CA. AcademicTherapy Pubbcadons.

Girolametto L, Wiigs M, Smyth R, et al. 2001.Children with a history of expressive lan-guage delay: outcomes at 5 years of age. AmJ Speech-Lang Pathol 10:358-369.

Hart N, Risley T. 1995. Meaningful differencesin everyday experiences of young Americanchildren. Baldmore: Paul H. Brookes Pub-lishing Co.

Henrichs J, Rescorla L, Schenck J, et al. 2011.Examining continuity of early expressive vo-cabulary development: the Generation Rstudy. J Speech Lang Hear Res 54:854-869.

Horwitz SM, Irwin JR, Briggs-Gowan MJ, et al.2003. Language delay in a communitycohort of young children. J Am Acad ChildAdolesc Psychiatry 42:932-940.

Hurtado N, MArchman VA, Femald A. 2008.Does input influence uptake? Links betweenmatemal talk, processing speed, and vocabu-lary size in Spanish-leaming children. DevSei 11:F31-F39.

Lee L. 1974. Developmental sentence analysis: agramniadcal assessment procedure for speechand language clinicians. Evanston, IL:Northwestern University Press.

Leonard LB. 1991. Specific language impairmentas a clinical category. Lang Speech HearServ Schools 22:66-68.

Lyydnen P, Poikkeus A, Laakso M, et al. 2001. Lan-guage development and symbolic play in chil-dren with and without familial risk for dyslexia.J Speech Lang Hear Res 44:873-885.

Marchman VA, Femald A. 2008. Speed of wordrecognidon and vocabulary knowledge ininfancy predict cognitiven and languageoutcomes in later childhood. Dev Sei11:P9-F16.

Moyle MJ, Ellis Weismer S, Lindstrom M, et al.2007. Longitudinal reladonships betweenlexical and grammatical development in typ-ical and late talking children. J Speech LangHear Res 50:508-528.

Newcomer P, HanimiU D. 1988. Test of languagedevelopiiient-2 (primary). Ausdn, TX:Pro-Ed.

Newcomer P, HamniiU D. 1997. Test of languagedevelopment-3 (primary). Ausdn, TX:Pro-Ed.

Paul R. 1993. Outcomes of early expressive lan-guage delay. J Childhood Commun Disord15:7-14.

Paul R. 1996. Clinical implications of the naturalhistory of slow expressive language develop-ment. AmJ Speech-Lang Pathol 5:5—21.

Paul R, Murray C, Clancy K, et al. 1997. Read-ing and metaphonological outcomes in latetalkers. J Speech Lang Hear Res 40:1037-1047.

ReiUy S, Wake M, Bavin EL, et al. 2007. Predicdnglanguage at 2 years of age: a prospecdve com-munity study. Pediatrics 120:el441-1449.

ReiUy S, Wake M, Ukoumunne OC, et al. 2010.Predicting language outcomes at 4 yean ofage: findings from Early Language in Victo-ria Study. Pediatrics 126:1530-1537.

Rescorla L. 1989. The Language DevelopmentSurvey: a screening tool for delayed lan-guage in toddlen. J Speech Hear Disord54:587-599.

Rescorla L. 2002. Language and reading out-comes to age 9 in late-talking toddlers. JSpeech Lang Hear Res 45:360-371.

Rescorla L. 2009. Age 17 language and readingoutcomes in late-talking toddlers: supportfor a dimensional perspective on languagedelay. J Speech Lang Hear Res 52:16-30.

Rescorla L, Achenbach TM. 2002. Use of thelanguage development survey (LDS) in anadonal probability sample of children 18 to35 months old. J Speech Lang Hear Res45:733-743.

Rescorla L, Alley A. 2001. Validadon of the lan-guage development survey (LDS): a parentreport tool for identifying language delay intoddlers. J Speech Lang Hear Res 44:434—445.

Rescorla L, Dahlsgaard K, Roberts J. 2000. Late-talking toddlers: MLU and IPSyn outcomesat 3;0 and 4;0. J Child Lang 27:643-664.

Rescorla L, Lee EC. 2000. Language impaimientsin young children. In: Layton T, Watson L,editors. Handbook of early language impair-ment in children: Volume I: Nature. NewYork: Delniar Publishing Company, pp. 1—38.

Rescorla L, Ratner NB, Jusczyk P, et al. 2005.Concurrent validity of the language devel-opment survey: associadons with the Mac-Arthur-Bates communicadve developmentinventories: words and sentences. Am JSpeech-Lang Pathol 14:156-163.

Rescorla L, Roberts J, Dahlsgaard K. 1997. Latetalkers at 2: outcome at age 3. J SpeechHear Res 40:556-566.

ReyneU JK. 1977. ReyneU Developmental Lan-guage Scales. Windsor, UK: NFER.

ReyneU JK, Gruber CP. 1990. ReyneU Develop-mental Language Scales, U.S. edidon. LosAngeles: Westem Psychological Services.

Rice ML, Taylor CL, Zubrick SR. 2008. Lan-guage outcomes of 7-year-old children withor without a history of late language emer-gence at 24 months. J Speech Lang HearRes 51:394-407.

Sameroff AJ, Chandler MJ. 1975. Reproducdverisk and the condnuum of caretaker casualty.In: F.D. Horowitz, M. Hetherington, S.Scarr-Salapatek, G. Siegel (Eds). Review ofchild development research. Vol. 4. Chi-cago: Univenity of Chicago Press.

Scarborough HS. 1990. Index of productive syn-tax. Appl Psycholinguist 11:1—12.

Schjolberg S, Eadie P, Zachrisson H, et al. 2011.Predicdng language development at 18months: data fi-om the Norwegian Motherand Child Cohort Study. J Dev BehavPediatrics 32:375-383.

Semel E, Wiig E, Secord W. 1995. CUnical evalu-adon of language fundamentals, 3rd ed. SanAntonio, TX: Psychological Corporadon.

Semel E, Wiig E, Secord W, et al. 1987. Clinicalevaluadon of language fundamentals-revised.San Antonio, TX: Psychological Corporadon.

Thai DJ, MiUer S, Carlson J, et al. 2005. Non-word repedtion and language developmentin 4-year-old children with and \vithout ahistory of early language delay. J SpeechLang Hear Res 48:1481-1495.

Thai DJ, Tobias S, Morrison D. 1991. Languageand gesture in late talkers: a 1-year foUow-up. J Speech Hear Res 34:604-612.

Tomblin JB, Records NL, Buckwalter P,Zhang X, Smith E, O'Brien M. 1997. Theprevalence of specific language impaimientin kindergarten children. J Speech LangHear Res 40:1245-1260.

US Prevendve Services Task Force. 2006. Screen-ing for speech and language delay in pre-school chUdren: recommendadon statement.Pediatrics 117:497-301.

DEV DISABIL RES REV- LATE TALKERS: D O G O O D PREDICTORS OE OUTCOME EXIST? • RESCORLA 149

Westerlund M, Berglund E, Eriksson M. 2006.Can severely language delayed 3-year-oldsbe idendfied at 18 months? Evaluadon ofa screening version of the MacArthur-Bates Communicative DevelopmentInventories. J Speech Lang Hear Res49:237-247.

Wetherby AM, Prizant BM. 2002. Communica-tion and symbolic behavior scales. Balti-more, MD: Paul H. Brookes Publishing.

Whitehurst G, Fischel J. 1994. Early developmen-tal language delay: what, if anything, shouldthe clinician do about it? J Child PsycholPsychiatry 35:613-648.

Whitehurst GJ, Fischel JE, Arnold DS, et al. 1992.Evaluating outcomes with children with ex-pressive language delay. In: Warren SF, ReichleJ, editors. Causes and effects in communicationand language intervention. Vol. 1. Baltimore,MD: Brookes Publishing p 277-313.

Wiig E, Secord W, Semel E. 2006. Clinical evalua-tion of language iundamentals-preschool-2.San Antonio, TX: Psychological Corporation.

Woodcock RW, Johnson MB. 1989. Woodcock-Johnson psychoeducational battery-revised.Chicago: Riverside Publishing.

Young EC, Perachio JJ. 1983. Patterned eUcita-tion syntax test, revised. Tucson, AZ: Com-munication Skill Builders.

Zimmerman 1, Steiner V, Pond R. 1992. Pre-school Language Scale-3. Chicago: Psycho-logical Corporation.

Zink I, Lejaegere M. 2003. N-CDIs: Korte vor-men, aanpassing en hemonnering van deMacArthur Short Fonn Vocabulary Checklistsvan Fenson et al. [N-CDIs: short forms, adap-tation and nonns of the Dutch version of theMacArthur Short Fomi Vocabulary Checklistsof Fenson et al.]. Leuuven, Belgium: Acco.

Zubrick SR, Taylor CL, Rice ML, et al. 2007.Late language emergence at 24 months: anepidemiological study of prevalence, predic-tors, and covariates. J Speech Lang HearRes 50:1562-1592.

150 DEV DISABIL RES REV- LATE TALKERS: D O G O O D PREDICTORS OF OUTCOME EXIST? • RESCORLA

Copyright of Developmental Disabilities Research Reviews is the property of John Wiley & Sons, Inc. and its

content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's

express written permission. However, users may print, download, or email articles for individual use.