34

REGULAR ARTICLES ..

Treatment Effects on Speech Intelligibility andLength of Utterance in Children with SpecificLanguage and Intelligibility Impairments

PAUL YODER, STEPHEN CAMARATA, & ELIZABETH GARDNERVanderbilt University

This purpose of this randomized group experiment was (a) to test the post-treatment (i.e.,immediately after treatment) and follow-up (i.e., 8 months after the end of treatment) efficacy ofa treatment designed to facilitate both sentence length and speech intelligibility (i.e., broadtarget recast), and (b) to explore whether pretreatment speech accuracy predicted response totreatment in children with severe phonological and expressive language impairment. Theresults support the conclusion that broad target recast facilitated follow-up speech intelligibilityin children whose speech accuracy was relatively low prior to treatment.

Preschoolers who have grammatical impair-ments (e.g., lower utterance length than ex-pected for age) and speech intelligibility im-pairments are at particular risk for continuedlanguage impairment. In this paper, speech re-fers to intelligibility of speech and grammarrefers to use of proper word order, use of func-tion words such as helping and linking verbs,and use of word endings such as possessiveand plural that affect sentence length. Thesevariables are important because childrenwhose speech cannot be understood are lesslikely to elicit language-facilitating interac-tions from significant others (Camarata &

Yoder, 2002). Reduced language-facilitatinginteraction from others is particularly prob-lematic for such children because they mightneed more exposures to adult expansions andother language-facilitating utterances to ben-efit from such interactions as compared to typ-ically developing children (Fey & Proctor-

Williams, 2000).A relatively high degree of co-occurrence

of speech disorders and language disordershas been reported (e.g., Aram & Kamhi,

1982). Additionally, outcomes for children

with simultaneous speech and language im-pairments are generally worse than for chil-dren with only one of the impairments (Aram& Hall, 1989). We refer to children with both

speech and language impairments with non-verbal intelligence within normal limits as

having a specific speech and language im-pairment (SSLI).

There has been relatively little focus on thesimultaneous treatment of intelligibility andsentence length in children with SSLI. Seg-regating speech intelligibility and sentence

length treatment might not be required andmight be unnecessarily inefficient. If a treat-ment that addressed both sentence length andspeech intelligibility goals in the same sessionwere effective, it would be an important ad-dition to the armament of available speech andlanguage treatments and might improve theoverall efficiency of treatment for childrenwith SSLI.

Conversational recasts might be an effec-tive way to facilitate sentence length and

speech intelligibility within the same treat-

35

ment session. Two types of recasts are of in-terest herein: speech and sentence length. Aspeech recast is an adult utterance that im-

mediately follows a child utterance, gives aneutral or positive evaluation of the meaningof the child’s utterance, and is an exact or re-duced imitation of the word(s) that the childattempted to say only using adult pronuncia-tion of the attempted word(s). Information

about accuracy of pronunciation (i.e., phono-logical information) is the only informationadded to the child’s production. For example,if the child says, &dquo;This a wion [lion],&dquo; the

speech recast might be &dquo;Yes, a lion.&dquo; No newword order, word ending, or vocabulary is

added to the child’s utterance (Camarata,1996). Sentence length recasts are similar tospeech recasts except that new vocabulary orgrammatical information is added to the pre-ceding child utterance. For example, if the

child says, &dquo;This lion,&dquo; the sentence lengthrecast might be &dquo;Yes, that is a lion&dquo; (Camar-ata, Nelson, & Camarata, 1994).

Speech recasts are hypothesized to provideinformation relevant to inducing rules abouthow speech sounds are combined to producewords (i.e., phonological rules) at moments

when the child is most likely to process theinformation. Given a speech recast is tempo-rally proximal to and retains the meaning ofthe child’s prior utterance, it might be easierfor the child to process that the adult pronun-ciation is different from his or her own (Nel-son, 1989). It is important that speech recastsdo not add new vocabulary or grammaticalinformation to the child’s preceding utterance.Many children with speech and language im-pairments have difficulty processing certain

types of grammatical information (Leonard,McGregor, & Allen, 1992) and certain typesof speech information (Tallal, Stark, & Mel-

lits, 1985). When a particular recast adds bothgrammatical and phonological information tothe child’s utterance, it is likely that the childwill only attend to one type of information.Experimental work indicates that children’s

grammatical development, not their phonolog-ical development, changes when the treatmentis composed of recasts that add both gram-matical and phonological information to the

child’s utterance (Camarata et al., 1994; Feyet al., 1994). Speech recasts, therefore, are de-signed to add only phonological information.

Empirical support exists for the efficacy ofspeech recasts to facilitate the acquisition andgeneralization of specific speech sounds andintelligibility in children with severe intelli-gibility disorders who have no evidence oforal motor impairment, the majority of chil-dren with speech impairments (Schriberg,Kwiatkowski, Best, Hengst, & Terselic-Weber,1986). Using specific speech sounds as tar-

gets, both infrequently and never-used speechsounds were acquired in intervention sessionsand generalized to spontaneous speech sam-ples with the mother or the clinician who didnot treat the child. These results were main-

tained 9 months after treatment and showed

replication across participants and behaviors(Camarata, 1993). These participants had in-telligibility disorders of unknown etiology.Similarly, children with co-occurring intelli-gibility and sentence length disorders havedemonstrated faster response to speech recaststhan to imitation training as evidenced by few-er sessions until their first spontaneous and ap-propriate use of target speech sounds in com-municative utterances with a nontraining adult(Camarata, 2004). Children with simultaneousdeficits in speech, language, and cognitive do-mains have demonstrated increased speech ac-curacy and improved overall speech intelligi-bility when provided with sound-specificspeech recasts (Koegel, Camarata, Koegel,Ben Tall, & Smith, 1998) and with speech re-casts targeting overall intelligibility (Camara-ta, 2004; Smith & Camarata, 1999). In thislatter study, the possible confound of in-

creased familiarity with increased number ofintervention sessions was controlled by tran-scribing the sessions in a random order.

Theoretical support for the efficacy of sen-tence length recasts is similar to that for

speech recasts. Sentence length recasts are

thought to be effective in facilitating sentencelength because of their temporal proximity tothe child’s previous utterance and retention ofthe meaning of the child’s previous utterance.These characteristics might aid the child inmaking necessary comparisons between his

36

utterance and the adult’s utterance. Such com-

parisons might make the added vocabularyand grammatical information in the adult’s ut-terance salient. Repeated exposure to manysalient examples of word order or word end-ing targets might aid the child in inducing theunderlying grammatical rules that affect sen-tence length (Nelson, 1989).

Sentence length recasts have been shown tobe effective in facilitating early language tar-gets that would result in longer utterances inseveral internally valid experiments. Nelson,Carskaddon, and Bonvillian (1973) found us-ing sentence length recasts that added infor-mation about the verb phrase resulted in great-er verb phrase complexity (e.g., adding thehelping verb and the -ing word ending) andincreased mean length of utterance (MLU) inchildren with typical development in the ex-perimental group as compared to a random-ized control group. In children with specificlanguage disorders, Camarata et al., (1994)found that using sentence length recasts thatspecifically targeted particular word order orword ending rules (e.g., add the -ed to the endof verbs for past tense) resulted in generalizedspontaneous use of the targeted word order orword ending rules.The previously reported studies all have

used either speech or sentence length recasts,but not both in the same session. Addition-

ally, the previously reported studies used re-casts for a limited set of very specific gram-matical or speech-sound goals. The treatmenttested in the present study differs from priorrecast treatments in two ways. First, we in-tended to recast most of the children’s utter-

ances that afforded developmentally appro-priate recasts, not just those affording recaststhat model a limited set of goals. Using abroad definition of what constituted a recast-

able child utterance potentially allows moregrammatical targets and speech sounds to bemodeled. Given the broad nature of the out-

comes that are of interest to parents and

teachers, we selected broad outcomes for this

study (i.e., MLU and speech intelligibility).It is probable that many grammatical targetsand speech sounds would need to improve

for treatment effects to be demonstrated on

these broad outcomes.

Second, both types of recasts were usedwithin the same sessions. Clinicians attempt-ed to speech recast all of the children’s in-terpretable but poorly articulated utterancesthat afforded developmentally appropriatespeech sounds to be modeled. Clinicians alsoattempted to use sentence length recasts forinterpretable and relatively well articulatedutterances that afforded developmentally ap-propriate grammatical goals to be modeled.Interventionists could ideally adjust the rel-ative emphasis on intelligibility and sentencelength on an utterance-by-utterance basis. Wecall this treatment broad target recasts (BTR).

Speech and language treatments that pro-duce effects detected months after the end of

treatment clearly are more important thanthose that produce effects seen only imme-diately after treatment. A recent review of

language treatments for children with pri-mary language disorders indicated, however,that such follow-ups are rare (Yoder & Mc-

Duffie, 2002). When they are present, theyare usually very short term (e.g., 3 weeks).There is a need to test for more long-termeffects of speech and language treatments onintelligibility and sentence length outcomes.

In this study, we distinguish intelligibilityfrom speech accuracy. Intelligibility has beendefined as the degree to which the speaker’sintended message is recovered by the listener(Kent, Weismer, Kent, & Rosenbek, 1989).Speech accuracy is the extent to which the

child accurately produces the speech soundsin the words he or she uses as compared tothe adult version of the words. For example,a child saying &dquo;ba wo&dquo; for &dquo;ball roll&dquo; mightbe intelligible in that the adult understood themeaning of the child’s production, but the

speech accuracy of the production if measuredin percent consonants correct (PCC) would be25% (i.e., &dquo;b&dquo; correct, the final &dquo;I&dquo; in ball

and roll incorrect, and &dquo;r&dquo; incorrect in roll).This example illustrates that assessments of

speech accuracy will not necessarily tell us

about a child’s intelligibility. Studies of chil-dren with intelligibility problems indicate thatthe measures of speech accuracy (e.g., per-

37

centage consonants correct) explain only anaverage of 16% of the variance in speech in-telligibility (e.g., percentage word attempts forwhich words are transcribed), even thoughboth measures are derived from the same

speech samples (Shriberg & Kwiatkowski,1982; Shriberg et al., 1986). It is reasonable

to consider intelligibility as the more sociallyimportant outcome. However, we recognizethat speech accuracy and speech intelligibilityare related constructs.

Initial speech accuracy might predict chil-dren’s response to BTR. By predicting re-

sponse to BTR, we mean that the size of thedifference between BTR and the control groupon post-treatment, follow-up intelligibility, orMLU is predicted to vary as a function of pre-treatment speech accuracy. This prediction isa statistical interaction between pretreatmentspeech accuracy and treatment group predict-ing the post-treatment or follow-up outcome.Children who generally have poor speech ac-curacy will poorly articulate proportionallymore of their utterances. If a child utteranceis poorly articulated, but interpretable, the

BTR model dictates that the interventionist

should provide a speech recast rather than asentence length recast. If frequency of speechrecast affects intelligibility, then the intelligi-bility of the children with worse speech ac-curacy should benefit most from BTR. In con-

trast, if a child’s utterance is articulated rela-

tively well, the BTR model dictates that in-terventionists should use a sentence lengthrecast. If frequency of sentence length recastsimpacts MLU, then the MLUs of the childrenwith relatively good speech accuracy shouldbenefit most from BTR because these children

will receive a higher proportion of recasts thatfacilitate sentence length.The purpose of this study was to (a) test the

main effects of BTR on post-treatment and

follow-up MLU and intelligibility, and (b) testthe statistical interaction between pretreatmentspeech accuracy and treatment group predict-ing the post-treatment and follow-up out-

comes. That is, we tested whether treatment

group differences on post-treatment and fol-

low-up outcomes varied as a function of pre-treatment speech accuracy.

METHOD

ParticipantsFifty-two preschoolers (mean chronologicalage in years = 3.65; SD = .71) with SSLIparticipated in the research. The children metthe following criteria establishing the presenceof specific language impairment: (a) meanlength of utterance (MLU) was at least 1.3 SDbelow chronological age expectation or a stan-dard score of 80 or below on the expressivescale of the Preschool Language Scale-3rd edi-tion (PLS-3; Zimmerman, Steiner, & Pond,1992), (b) nonverbal IQs on the Leiter Inter-national Performance Scale-Revised (Leiter-R ; Roid & Miller, 1997) above 80 (i.e., theminimum for the borderline IQ range + theSE for the test), and (c) passing a 25 db

threshold hearing screening. Additionally,children scored no higher than a T score of 37(i.e., corresponding to 1.3 SD below the meant of 50, SD = 10) on the Arizona ArticulationProficiency Scale (AAPS; Fudala & Reynolds,1986) to document speech accuracy impair-ments.

To increase the probability that the childrenwould be appropriate for the language treat-ment being tested, the children had to haveinitial MLUs below 2.5, and use at least 10different words in a 20-min language samplewith an examiner. English had to be the onlylanguage spoken in the home. Children withevidence of oral motor disorders were exclud-

ed because the treatment presumably affectedintelligibility problems due to causes other

than oral motor disorders. Oral motor disor-

ders were screened using the Oral SpeechMechanism Screening Exam-Revised (St. Lou-is & Ruscello, 1987).

Participants were 38 boys (73%) and 14

girls (27%) with an equal number in each °

treatment group. Seventy-one percent wereEuro-American with 18 and 19 in the controland BTR groups, respectively. The remaining15 were African-American (7), &dquo;other&dquo; (6), orAsian (1). There were no differences amongtreatments in racial composition of the sam-ple. Parents’ formal education levels were 1-2years of college on average, but ranged from1 Oth grade to graduate school and did not dif-

38

Table 1

Means and Standard Deviations for Participant Descriptor Variables

Note. PLS-3 = Preschool Language Scale-Revised 3~d Edition; MLU = mean length of utterance; *There were nostatistically significant differences between groups on any of these variables.

fer between groups. The parents’ occupationalstatus was above the national average and didnot differ significantly between groups. Table1 shows the means and SDs on the partici-pants’ chronological ages, nonverbal intelli-

gence, productive and receptive language, ar-ticulation percentile rankings, MLU, and in-telligibility. The average expressive standardscore was 71.3 (SD = 14.8), and the averagearticulation T score was 24.5 (SD = 5.6). TheAAPS T score has a population mean of 50and a population standard deviation of 10. Thesample mean score was, on average, more

than 2.5 SD below age-expected levels. Anaverage of 51 % of the children’s utterance at-

tempts were fully intelligible by trained, butunfamiliar, transcribers, given up to three lis-tenings per utterance. These data indicate thatthe sample had co-occurring and severe ex-pressive language and articulation impair-ments.

Attrition. Only one participant did not

meet the attendance criterion for BTR after

enrolling in the study. Prior to enrolling, 12participants declined to participate in the

study. Seven (58%) of these had been random-ly assigned to the BTR group. The most fre-quent reasons given for declining was too

much travel and too many other things to dofor these families. The remaining five familiesgave no reason for declining. There were noobvious differences in selection criteria be-

tween participants who declined and thosethat remained in the study.

Research DesignA randomized group experiment was used totest the efficacy of BTR on MLU and intelli-gibility at post-treatment (i.e., 6 months afterstudy entry) and follow-up (14 months afterstudy entry) periods. Random assignment wasimplemented using a computer program ex-plicitly designed to allow equal probability ofbeing assigned to either group. The effective-ness of random assignment was evaluated

through tests of pretreatment group equiva-lence on 12 pretreatment variables and two

nonproject treatment attendance variables. Anexploratory, correlational design element wasadded to this randomized group experiment totest whether pretreatment speech accuracy

predicted response to BTR at the post-treat-ment and follow-up periods.

ProceduresOverview. Upon entry into the study, chil-dren were administered the Leiter-R, PLS-3,AAPS, the oral mechanism screening, and aninitial 20-min language sample. If selection

criteria were met, children were administered

a second 20-min language sample to providea more stable estimate of their initial intelli-

gibility and MLU. After pretreatment testing,children were randomly assigned to one of

39

two groups: BTR or control. Children in theBTR group received three 30-min treatmentsessions per week for 6 months. Children in

the control group were free to participate incommunity-based treatments but were not

provided BTR. The extent to which childrenreceived outside-of-project treatment wasmeasured to determine whether potential con-founds with group assignment could accountfor post-treatment or follow-up differences be-tween treatment groups. Implementation ofthe BTR treatment was examined in detail bycoding the three treatment sessions that oc-

curred at the 3rd, 4th , and Sth month of treat-ment. At post-treatment (6 months after studyentry) and at follow-up (14 months after entryinto the study), two 20-min language sampleseach were administered to measure intelligi-bility and MLU. The three measures of pre-treatment speech accuracy were derived fromthe AAPS and language samples. Outcomeswere derived from the post-treatment and fol-

low-up language samples.

Measures

Arizona Articulation Proficiency Scale

(AAPS; Fudala & Reynolds, 1986). TheAAPS is a standardized single-word articula-tion test. The child is shown a picture andasked &dquo;What is this?&dquo; If the child does not

respond, the examiner provides the correct

word and says, &dquo;Now you say it.&dquo; In addition

to using T scores (a type of standard score) toqualify and describe participants, AAPS rawscores were used to quantify elicited single-word speech accuracy that is weighted for de-velopmental level of the target phoneme beingtested. Standard scores are adjusted for chro-nological age; raw scores are not. As we in-tended to measure it, the construct of speechaccuracy is referenced to adult pronunciation,and thus, for the purposes of this study, werenot adjusted by children’s chronological age.Raw scores are more appropriate than stan-dard scores for our purposes of quantifyingspeech accuracy level. Although raw scoresare not intervally scaled, the statistics used inthis report do not assume the variables to beon an interval scale or ratio scale (Tabachnick

& Fidell, 1996). Internal consistency on thecurrent sample for the raw score is .89.

Twenty-minute language samples. This

procedure was administered to derive mea-

sures of intelligibility and sentence length atthe three measurement periods. Additionally,the pretreatment language sample was used toderive percentage vowels correct and percent-age consonants correct. These measures were

derived from the average of two, 20-min lan-

guage samples. The same sets of toys wereused for all children and examiners were

taught to use a prescribed, responsive inter-action style during the language sampling ses-sions. To avoid boredom, separate sets of toyswere counterbalanced across measurement pe-riods.

We instituted a strict definition of a word

to achieve reliability on the intelligibility mea-sure. Conceptually, we transcribed any audi-ble (whispered or vocalized) approximation ofa word that is in the English dictionary andfor which there was immediate nonlinguisticsupport, immediate conversational support, orevidence of productive use elsewhere in thesession. Over- or under-extended meaningswere acceptable as long as they had immedi-ate nonlinguistic support. An acceptable ap-proximation must be similar to the adult pro-nunciation in one of two ways. If the syllablestructure was intact, an accurate vowel or anaccurate initial consonant had to be present.In the absence of the same number of sylla-bles as the adult pronunciation of the word, atleast one accurate consonant-vowel or vowel-

consonant combination had to be present.The language samples were transcribed, in-

dicating bound morphemes such as regularpast tense (e.g., /ed/) using the format pre-scribed in Systematic Analysis of LanguageTranscripts (SALT; Miller & Chapman,1993). Pilot work on the potential effect ofdifferential transcriber familiarity indicatedthat familiar transcribers’ MLU and intelligi-bility estimates were over 1 SD larger thanunfamiliar transcribers’ (Yoder & Gardner,2004). The language samples, therefore, weretranscribed by someone other than the child’sclinician. Transcribers were equally unfamiliarwith control and BTR participants. Using the

40

SALT program, we derived MLU in mor-

phemes and the proportion of utterance at-

tempts that were fully intelligible from pre-treatment, post-treatment, and follow-up lan-guage samples. The SALT program derivesMLU from complete and fully intelligible ut-terances. The intelligibility measure was de-rived from all of the child’s utterances.

After transcription was completed, theSALT file was imported into ComputerizedProfiling (CP, Long & Fey, 1995). Trainedspeech-language pathologists indicated thesounds that were said for each of the tran-

scribed words. The phonology module withinCP (i.e., PROPH+) was used to derive per-centage vowels correct and percentage con-sonants correct at pretreatment. These vari-ables were viewed as measures of pretreat-ment conversational speech accuracy.

Treatment GroupsBroad target recasts group (BTR). BTR ses-sions were conducted three times a week for30 mins a session for 6 months. The interven-tionist was a woman with an undergraduatedegree in psychology. She was trained to 95%compliance to the BTR model and was mon-itored weekly by a master’s level speech-lan-guage pathologist. Therapy sessions were ad-ministered in a small therapy room using a 1:

1 format. A table and chair were available inthe room, but much of the interaction occurredon the floor. The therapist followed the child’sattentional and play lead and asked questionsabout what the child was attending to and do-ing in an attempt to get the child to speak.When the child did speak, the interventionistdetermined whether the child utterance afford-

ed a recast of a developmentally appropriategrammatical target or speech sound. Theoret-ically, the interventionist used sentence lengthrecasts for well articulated utterances (i.e., fewor no speech accuracy errors) and speech re-casts for understood, but poorly articulated

(i.e., many speech accuracy errors) utterances.The target recast rate was 4 per min. Partici-

pants were allowed to miss three treatmentsessions and were asked to make up any ad-ditional sessions missed. Participants whomissed treatment in excess of nine sessions

were withdrawn from the study. Only one par-ticipant that began treatment did not meet thiscriterion.

Control group. Children in the control

group were not administered the BTR treat-

ment. A wait-list control was not used because

a major question of the study was long-termefficacy of the treatment, thus requiring a longinterval between the treatment phase endingand the follow-up assessment. The partici-pants in the control group were not prohibitedfrom seeking and obtaining treatment outsidethe study.

Fidelity of treatment. To describe the BTR

treatment implementation, the middle 15 minsof the videotaped treatment sessions from

each of the 3~d, 4th, and 5th months of treatmentwere coded in detail for every child in theBTR group. The behaviors coded from these

videotaped sessions are listed in Table 2.

Nonproject Treatment DescriptionBecause families were free to enroll in com-

munity-based treatments (i.e., those offered bynonproject personnel) and because it is pos-sible that children in the control group soughtnonproject treatment more often than the BTRgroup, nonproject treatment could potentiallyrelate to between-group differences in the out-comes. Although this is rarely measured in

treatment studies, it was included herein be-

cause of this potential impact. Assuming thatnonproject treatment can be facilitative of de-velopment and that BTR also facilitates de-velopment, greater nonproject treatment couldpotentially result in underestimating the BTRtreatment effect size. Although control groupparents were not asked to abstain from non-

project treatment, we can statistically controlbetween-group differences on nonprojecttreatment involvement if there are between-

group differences in the amount of nonprojecttreatment and such differences are related to

outcomes. Therefore, we asked parents to re-

port the number of hours their children at-

tended speech, behavior, physical, and occu-pational therapy for speech and language de-lays in an early intervention program, daycare,clinic, or home. Because enrollment and at-tendance in these therapies were expected to

41

vary during the study, parents were asked toreport participation in nonproject treatments 4times over the course of the project (pretreat-ment, 6 months after study entry, 10 monthsafter study entry, and 14 months after studyentry). It should be noted that the last 2 pe-riods occurred after the post-treatment assess-ments. From this information, we derived (a)the average number of therapy hours from aspeech-language pathologist in all settings permonth and (b) the average number of hoursof any type of educational services from pro-fessionals or paraprofessionals other thanspeech-language pathologists. We also exam-ined whether these two variables assessed atthe last 2 measurement periods could accountfor differences in follow-up results.

ReliabilityWe estimated the interobserver reliability ofthe variables that were derived from transcrip-tions. The interobserver reliability estimate

was the intraclass correlation coefficient (Suen& Ary, 1989). For all coded variables, includ-ing those derived from phonemic transcrip-tion, reliability was estimated from the vid-eotapes of the samples. The same orthograph-ic transcription should not be used to estimatereliability for percentage vowels and conso-nants correct because doing so bypasses esti-mating measurement error due to different or-thographic transcriptions. We sampled be-tween 17% (i.e., 8 for percentage vowels cor-rect at pretreatment) and 39% (i.e., 20 for

MLU and intelligibility at pretreatment) of the52 participants for reliability estimates. Sam-ples used for reliability estimation were se-lected randomly and reliability transcriptionand coding was conducted independently fromthe primary transcription and coding.

At pretreatment, the intraclass correlationcoefficients were .98, .51, .43, and .91 for

MLU, percentage utterance attempts that were

fully intelligible, percentage vowels correct,and percentage consonants correct, respective-ly. At post-treatment, intraclass correlation co-efficients were .77, and .74 for MLU and per-centage utterance attempts that were fully in-telligible, respectively. At follow-up, intra-

class correlation coefficients were .97, and .89

for MLU and percentage utterance attemptsthat were fully intelligible, respectively. Intra-class correlation coefficients were between .71 1

(rate of child recastable utterances) and .97 (M= .92; SD = .08) for fidelity of treatment var-iables. Decreased reliability results in a re-

duced probability for detecting a significantdifference in the analyses. Significant findingsoccurred despite reliability under .70 at Time 1.

RESULTS

Preliminary AnalysesPrior to addressing the research questions, weconducted preliminary analyses to (a) exam-ine pretreatment differences between treat-

ment groups, (b) test group differences on

nonproject treatment attendance, (c) describethe amount of recasts provided to documentdosage of treatment, and (d) document theamount of growth children experienced re-

gardless of treatment group.Test of between-group differences on select

pretreatment variables. A test of between-

group differences on select pretreatment var-iables was completed to determine whetherrandom assignment yielded equivalent groupsat the pretreatment period. Twelve pretreat-ment variables were tested for group differ-

ences to determine whether random assign-ment was effective in creating equivalentgroups prior to treatment onset. There were nostatistically significant differences on any ofthese 12 variables, indicating that random as-signment was successful at creating equivalentgroups at the pretreatment period.

Test of between-group differences on non-project treatment participation. We tested the

significance of between group differences onthe average number of hours of speech-lan-guage therapy per month and the averagenumber of therapeutic hours of other profes-sionals and paraprofessionals per monthacross all four measurement periods. Therewere both variance, F(2, 50) = 14.5; p =.0001, and mean differences, d = -.63;t(25.3) = 2.29; p = .03, favoring the controlgroup on the number of hours with other pro-fessionals and paraprofessionals. Consideringboth BTR and nonproject treatments, there

42

Table 2

Description of BTR Implementation across the 3rd, 4th, and 5Ih Months of Treatment

Note. BTR = broad target recast.

was a statistically significant difference in thetotal amount of treatment favoring the controlgroup, t(28.5) _ -3.18, p = .004. This be-tween-group difference in total treatment wasdue to nonproject treatment provided by otherprofessionals and paraprofessionals (i.e., not aspeech-language pathologist). Additionally,we tested whether the treatment groups were

different on the two nonproject treatment at-tendance variables at 10 and 14 months after

study entry. The 10-month measurement pe-riod was between the post-treatment and fol-

low-up measurement periods for MLU and in-telligibility. The 14-month measurement pe-riod was the follow-up measurement period ofMLU and intelligibility. There was no statis-tical evidence of group differences in nonpro-

ject treatment attendance at these periods (allp values > .16).We examined whether the group difference

in average amount of nonspeech/languagetreatment across all four measurement periodscould account for the results of the tests of theresearch questions. Attendance to nonprojecttreatment was not associated with outcomes at

either post-treatment or follow-up periods.Additionally, nonproject treatment attendancevariables measured for the period betweenpost-treatment and 8 months after treatmentended were not associated with MLU or in-

telligibility at the follow-up period. Nonpro-ject treatment attendance, therefore, cannot

account for differences between groups or in-

teractions between group and pretreatmentspeech accuracy.

Description of the BTR treatment imple-

mentation. Children assigned to the BTR

group attended an average of 74 treatment ses-

sions (SD = 5.6). Individual variability in at-tendance to BTR did not account for variancein post-treatment or follow-up outcomes. Ta-ble 2 shows the central tendency and vari-ability of the variables measured from the

BTR sessions. These fidelity of treatment datadocument that the treatment was implementedas intended. The target rate of recasts (i.e., 4/min) was slightly surpassed, and there wasmuch variability in the rate of recasts. This

occurred in part because the rate of recastable

utterances, which serve as the antecedents to

staff-delivered recasts, varied widely. On av-erage, 82% of the children’s recastable utter-

ances were recasted. The proportion of staff-implemented recasts that were sentence lengthrecasts was positively related to the children’sintelligibility at post-treatment (r = .45; p =

.05), suggesting that more intelligible childrenreceived proportionally more sentence lengthand proportionally fewer speech recasts.

Growth in MLU and intelligibility, regard-less of treatment group. The main effect of

time was tested using a repeated measuresANOVA with time as a 3-level within-sub-

jects factor (pretreatment, post-treatment, fol-low-up). We used the Greenhouse-Geisser ad-justment to the degrees of freedom to correctfor potential violations to the sphericity as-

sumption. There was a strong effect of timeon MLU, F(1.43, 50) = 67.18; p < .001; ~2 2= .62, and a strong effect of time on intelli-

gibility, F( 1.77, 50) = 10.89; p < .001; ~2 =.24. These data indicated significant and large

43

growth in MLU and intelligibility across pre-test through follow-up.

Tests of the Research QuestionsTests of the main effect of BTR. Using fourindependent t tests, we tested whether therewere BTR versus control group mean differ-ences at the post-treatment and follow-up pe-riods on MLU and intelligibility. There wereno statistically significant differences betweengroups on either outcome at either measure-ment period (all p values > .38).

Statistical interactions between pretreat-ment speech accuracy and treatment predict-ing post-treatment and follow-up outcomes.As an exploratory analysis, we examinedwhether pretreatment speech accuracy (an at-tribute variable measured on a continuous

scale) interacted with the treatment group(BTR vs. no BTR) in predicting individualdifferences in MLU or intelligibility at post-treatment (i.e., short-term) and follow-up (i.e.,long-term) measurement periods. A multipleregression analysis was chosen for its abilityto determine proportion of the variance in acriterion variable related to continuous

(speech accuracy) and categorical (treatmentgroup) variables and their interaction. Using acontinuous form of pretreatment speech ac-curacy variable retained maximum informa-tion about individual differences in initial

speech accuracy and maximized statistical

power (Cohen, 1983). Four multiple regres-sions in which the criterion variable was eitherMLU or intelligibility at either the post-treat-ment or follow-up period were tested. The

same three predictor variables were includedin the regression model for each of the fouranalyses: (a) the pretreatment variable, (b) thetreatment group, and (c) the product term for(a) and (b) (Aiken & West, 1991). The pre-treatment variable is grand mean centered toreduce multicollinearity between the pretreat-ment variable and the product term (Aiken &

West, 1991). The treatment group variable is

dummy coded (e.g., 0, 1) to aid interpretation(Aiken & West, 1991). The test of the signif-icance of the product term tests whether themagnitude of the between-group differencevaries as a function of the pretreatment vari-

able (i.e., a statistical interaction between

group and the pretreatment variable). Statis-

tically significant interactions are interpretedusing the Johnson-Neyman technique, whichindicates the values on the pretreatment vari-able below which or above which the treat-

ment groups differ on the dependent variable(Aiken & West, 1991). Only interactions withat least two participants in each treatment

group were interpreted.All statistical assumptions were met for in-

teractions between pretreatment variable and

group predicting post-treatment and follow-upoutcomes (i.e., linearity, homogeneity of re-siduals, no undue influence, and low multi-

collinearity). Table 3 shows the details of thestatistically significant interactions. BTR fa-cilitated short-term and long-term generalizedMLU in children who began treatment pro-ducing less than 49% or 50% consonants cor-rect, respectively. Similarly, BTR facilitatedlong-term generalized intelligibility in chil-dren who began treatment with less than a rawscore of 46 on the AAPS. These effect sizesfor the entire group are weak to moderate. The

total number of children in the region of sig-nificance in which BTR is superior has 9, 10,and 11 children for the post-treatment MLU,follow-up MLU, and follow-up intelligibility,respectively. Regressions are based on the en-tire sample of 52 children and the variablesanalyzed are continuous rather than dichoto-mous. Figure 1 shows the statistical interac-

tion of pretreatment AAPS raw score and

group predicting follow-up intelligibility. Thefigure shows the two regression lines repre-senting the association between grand-meancentered AAPS raw score at pre-treatment and

follow-up intelligibility for the two groups. Aregression line is a line that best fits the in-

telligibility scores at each of the pretreatmentAAPS raw score values. It is similar to a run-

ning mean intelligibility score at each level ofthe pretreatment AAPS raw score. The verti-cal line at -10.25 indicates that the group’spredicted intelligibility scores are statisticallydifferent in children with pretreatment grand-mean centered AAPS raw scores of -10.25and below. The mean for pre-treatment AAPSis 55.89. The uncentered pretreatment AAPS

44

1

.9cs~

~~r

3Ut

b4i2l

~~§°3£Sa

~£~)

ê-fl0()

~fV)

~3

v

.£

~~~_N

i2lU

~~~Q

~

#

3vQo

~~

~i-s 5c~

iS§

~e~~3

M~ ~’S fl ils

ai

u

UaII)

u

24.§

O0

&dquo;5u

«C0N

11rn

Q)u~2a’0-9tos::~

1äII)

e11

~a

w~

Figure 1.Regression lines for the statistical interactionbetween pretreatment AAPS raw score and

group predicting follow-up intelligibility.

raw score at which average groups’ intelligi-bility scores were statistically different was 45and under (i.e., 55.89 - 10.25 = 45.64).

To test for potential alternative explanationsfor the statistical interactions, we computedthe effect size for the difference between con-trol and BTR groups for all pretreatment var-iables within the children in the lower regionof significance. Only six variables had effectsizes over a Cohen’s d of .50. None of these

pretreatment variables correlated significantlywith the post-treatment MLU or either follow-

up outcome. Pretreatment differences between

groups within the children in the region of sig-nificance, therefore, could not explain the sig-nificant statistical interactions.

DISCUSSION

This study was conducted to determine if BTRwas effective in facilitating MLU and intelli-gibility 8 months after the end of treatmentand to determine whether pretreatment levelof speech accuracy was a predictor of long-term response to BTR in a sample of pre-schoolers with severe expressive language andspeech intelligibility impairments but age-ap-propriate nonverbal intelligence and normal

45

hearing. Children with both speech and lan-guage impairments are at particular risk of fu-ture language deficits and subsequent academ-ic risk because poor speech accuracy is

thought to reduce the frequency of potentiallyfacilitating input from others.The results show no evidence that the BTR

treatment was effective for children across the

entire range of speech accuracy represented inthe present study. Instead, the BTR treatmenteffects were dependent on the pretreatmentspeech accuracy levels of children. Only chil-dren with relatively low pretreatment speechaccuracy benefited from the BTR treatment.

This conditional treatment effect occurred forboth MLU (at the post-treatment and follow-

up periods) and intelligibility (at the follow-up period).

Absence of BTR Main EffectsSeveral factors are potential explanations forwhy there was no main effect for BTR. Thisstudy was not strictly a replication study. Mostprevious recast studies used target-specific re-casts and either sentence length or speech re-casts, but not both. We used broad target re-casts and combined both types of recast in onesession. Additionally, the control group hadgreater non-SLP treatment attendance than theBTR group. Either of these explanations couldaccount for the absence of main effects for

BTR. Finally, the absent main effect of BTRcould be a function of the children’s need formore targeted stimulation of a limited set ofparticular goals.

Greater Interpretability of the StatisticalInteraction on IntelligibilityAlthough there were three statistically signif-icant interactions, there are two reasons whywe consider the BTR effect on the intelligi-bility of children with relatively low pretreat-ment speech accuracy the most reliable (i.e.,most likely to replicate). First, the children af-fected by BTR on intelligibility were predict-able before treatment began. Because childrenwith initially low speech accuracy were ex-pected to receive proportionally more speechrecasts and because speech recasts are the as-pect of the BTR thought to address intelligi-

bility, we expected them to benefit most fromBTR. This finding was seen in the results. Wedid not see the predicted relationship for

MLU, because contrary to the findings, we ex-pected children with relatively high initial

speech accuracy to benefit from BRT with re-gards to MLU outcomes. Second, the BTRtreatment effect on MLU became nonsignifi-cant after controlling for pretreatment predic-tors of MLU (i.e., receptive language level;Yoder, Camarata, & Gardner, in preparation).As expected, the BTR effect on intelligibilityof initially low speech accuracy children re-mained significant after controlling for pre-treatment predictors of later intelligibility(Yoder et al., in preparation).

Rationale for Interpreting a TreatmentEffect on Follow-up Intelligibility WhenPost-treatment Effects are Not SeenSome readers might wonder how one can in-terpret the conditional treatment effects of

BTR on follow-up intelligibility when there isno analogous effect on intelligibility at post-treatment. This type of treatment effect hasbeen called a sleeper effect because the effectis not seen until months after the treatment

ends (Clarke & Clarke, 2004). Such effectscan occur when small changes occur at thepost-treatment period in the children that elicitfrom their caregivers facilitating input duringthe interval between post-treatment and fol-

low-up periods. These inputs, in turn, can af-fect follow-up child outcomes. In the presentstudy, it is possible that our measures of in-telligibility at post-treatment were not suffi-

ciently sensitive to detect treatment effects onsome children’s intelligibility, which in turnelicited speech recasts from parents in the

form of confirmation requests. Other reportshave found that parents of children who are

very unintelligible tend to use recasts after

partially intelligible utterances (Yoder, Hoosh-yar, Klee, & Schaffer, 1996). We were not ableto test whether this mechanism accounted for

the conditional BTR effect on follow-up in-telligibility in the present study. Similar ef-fects, however, were seen from a prelinguistictreatment on language development that didnot become manifest until 6 and 12 months

46

after the treatment ended (Yoder & Warren,2001b). This indirect effect occurred throughparental linguistic mapping of children’s pre-linguistic communication (Yoder & Warren,2001 a).

The Paucity of Evidence aboutTreatment Efficacy on IntelligibilityEven though the effect size is small to mod-

erate (R2 = .09) and the effect is conditionalon initial speech accuracy, the finding thatBTR has a long-term effect on intelligibilityis potentially important. No other studies withthis population have examined treatment ef-fects on long-term follow-up intelligibility;therefore, we cannot compare this effect sizewith those in comparable studies. Addition-ally, almost every other study examining theefficacy of speech therapy examines speechaccuracy, rather than speech intelligibility, asthe outcome. The almost complete lack of ev-idence of treatment efficacy on intelligibility,rather than speech accuracy, is striking giventhe social importance of intelligibility. BTR’seffect on intelligibility in children with ini-

tially low speech accuracy is even more note-worthy because it occurred 8 months after

treatment ended. To our knowledge, this is thefirst time it has been shown that conversation-al intelligibility is affected by treatment 8months after the end of treatment. Document-

ing this finding is the most important contri-bution of the current study to the literature.The finding that BTR affected the speech

intelligibility of only the most developmen-tally immature speech producers might con-tradict many reader’s expectations. Speechtherapists have been conducting very struc-

tured articulation therapy for many decades.A continuum of support model would pre-scribe using BTR to address generalization ofsounds recently acquired in the clinic or toaddress the phonological needs of less im-paired children regardless of developmentallevel (Camarata, 1995, 1996). The presentstudy’s findings do not support such a modelin that BTR (which would be considered rel-atively low support and highly naturalistic ininteraction style) was not effective in childrenwith relatively high levels of pretreatment

speech accuracy, but provided support that

BTR affected intelligibility in children with

relatively poor pretreatment speech accuracy.

Limitations of the Study ’

Three primary limitations existed in the pre-sent study. First, we carefully coded and thusdescribed three sessions per BTR participant(4% of the sessions). Coding more sessionswould be more likely to provide a more stableestimate of the typical implementation of theBTR treatment. Second, we did not have in-formation available before the study to knowwhich raw scores defined the margins of thesubgroup in which BTR would be effective.This particular subgroup could have been ov-ersampled and improved the probability thatour subsample of affected children represent-ed the population of children affected byBTR. Instead, the variables expected to pre-dict BTR response were kept continuous andallowed the distribution of scores to be deter-

mined by chance. As usual, the tails of the

distribution on the predictors of BTR responsehad fewer people than the middle of the dis-tribution. Regions of statistical significanceand significance level of statistical interactionsare determined by the entire range of scores,not just those in the regions of significance(Aiken & West, 1991). Although replicationof the interaction is needed, results of this

study should not be disregarded as unimpor-tant simply on the basis of relatively few par-ticipants occurring in the region of signifi-cance. The number of children in the regionof significance in this study is similar to stud-ies that examine predictors of language treat-ment response in children with disabilities

(Cole, 1995; Yoder, Kaiser, & Alpert, 1991;Yoder et al., 1995; Yoder & Warren, 1998).Third, theory allowed us to predict which chil-dren BTR would affect with regards to intel-ligibility, but not which children BTR wouldaffect with regards to MLU. Finally, we didnot have a norm-referenced measure of ex-

pressive language. We could not determine ifany of the control group children caught upwith their age-appropriate expressive lan-guage without treatment. Possibly includingsuch &dquo;late bloomers&dquo; in the study influences

47

our understanding of the appropriate types ofchildren to which the findings might general-ize, but does not influence the internal validityof the study.

Strengths of the StudyThe quality of this efficacy study was note-worthy for several reasons. For example, theecological validity of the outcomes was

strong. They were obtained from languagesamples instead of standardized tests. The for-mer is the more frequent language use contextfor preschoolers. Additionally, length of utter-ance and intelligibility are outcomes at a

broad enough level of description that parentsand teachers are highly likely to notice a dif-ference. Another strength of the study is theuse of an 8-month follow-up period to deter-mine whether treatment effects occurred manymonths after the end of treatment. Finally,there are three reasons why internal validityof the study was strong: (a) random assign-ment to treatment groups was used, (b) pre-treatment equivalence on a number of vari-ables that can affect variance in outcomes was

demonstrated, and (c) nonproject treatment at-tendance did not account for differences be-tween treatment groups or statistical interac-

tion between pretreatment speech accuracyand group on outcomes. Although nonprojecttreatment participation can relate to treatmenteffects, few studies examine this factor.

Measuring nonproject treatment: A noveldesign element. We considered it problematicto require control group parents to refrain

from all other treatment for 14 months (6months of treatment phase + 8 months of fol-low-up). Because group differences can anddid occur after the random assignment to

treatment groups, it was important to controlfor such differences. We did so by (a) testingto see if the confound occurred, (b) testing tosee if the difference was associated with the

outcomes, and (c) statistically controlling forthem, if needed. The greater treatment atten-dance (i.e., differences on nonproject, non-speech/language treatment) in the controlgroup could not account for the reported sig-nificant results for two reasons. Nonprojectnonspeech/language treatment was not asso-

ciated with any outcome at any period and thecontrol group was the group with the most

total and nonproject treatment. Unless treat-ment negatively affected outcome, it is un-

likely that the nonproject treatment attendanceexplained why BTR children with initiallylow speech accuracy had better intelligibilityat follow-up.

Future studies should collect nonprojecttreatment attendance, as it has been positivelyassociated with later outcomes in other studiesof preschoolers with disabilities (Stone &

Yoder, 2001). Future research also should de-

velop predictive measures of the type and

quality of nonproject treatment and how thesemight modify within-project treatment effica-cy. Measuring only attendance in this study isnot a statement that attendance is sufficientinformation. The field presently almost nevermeasures any aspect of nonproject treatment.Attendance to such treatment is only a placeto start investigating this important, but rarelystudied, class of potential threats to internalvalidity.

Future ResearchFuture studies are needed to determine if BTRis effective when used in the classroom byclassroom teachers who are not carefully su-pervised. One potentially important attributeof BTR treatment is the rate of speech recast-ing, which averaged 2.6/min in the presentstudy. If classroom teachers or other interven-tionists are spreading the number of recastsamong 4 to 5 children, the number of recasts

per child per min is likely to be less and theefficacy of the recast might be compromised.Additionally, it is less likely that children withSSLI will process a recast of another child’s

utterance as well as a recast to their own ut-

terance because presumably recasts are effec-tive in part because they maintain the speak-er’s meaning.

ConclusionThis study was the first to identify that initialsingle word speech accuracy predicts whichchildren respond to broad target recasts as anew treatment for long-term conversationalintelligibility. This treatment was only effec-

48

tive for children with initially very low singleword speech accuracy (i.e., below 46 rawscore on the AAPS).

REFERENCES

Aiken, L., & West, S. (1991). Multiple regression:Testing and interpreting interactions. NewburyPark: Sage.

Aram, D., & Hall, N. (1989). Longitudinal follow-up of children with preschool communicationdisorders: Treatment implications. School Psy-chology Review, 18, 487-501.

Aram, D., & Kamhi, A. ( 1982). Perspectives on therelationship between phonological and lan-

guage disorders. Seminars in Speech, Lan-

guage, and Hearing, 3, 3-22.

Camarata, S. (1993). The application of naturalisticconversation training to speech production inchildren with speech disabilities. Journal ofApplied Behavior Analysis, 26(2), 173-182.

Camarata, S. (1995). A rationale for naturalistic

speech intelligibility intervention. In M. E. Fey,J. Windsor, & S. F Warren (Eds.), Languageintervention: Preschool through the elementaryyears (pp. 63-84). Baltimore: Brookes.

Camarata, S. (1996). On the importance of inte-grating naturalistic language, social interven-tion, and speech-intelligibility training. In L. KKoegel, R. L. Koegel, & G. Dunlap (Eds.),Positive behavioral support: Including peoplewith difficult behavior in the community. (pp.333-351). Baltimore: Brookes.

Camarata, S. (2004). A comparison of imitative andconversational recast intervention for phono-logical disorders in children. Manuscript in

preparation.Camarata, S. M., Nelson, K. E., & Camarata, M.

N. (1994). Comparison of conversational-re-casting and imitative procedures for traininggrammatical structures in children with specificlanguage impairment. Journal of Speech andHearing Research, 37(6), 1414-1423.

Camarata, S., & Yoder, P (2002). Language trans-actions during development and intervention:Theoretical implications for developmentalneuroscience. International Journal of Devel-opmental Neuroscience, 20(3-5), 459-465.

Clarke, A., & Clarke, A. (2004). Human resilience:A fifty year quest. Philadelphia: Jessica Kings-ley.

Cohen, J. (1983). The cost of dichotomization. Ap-plied Psychological Measurement, 7(3), 249-253.

Cohen, J. (1988). Statistical power analysis for the

behavioral sciences (2nd ed.). Hillsdale, N.J.:L. Erlbaum Associates.

Cole, K. N. (1995). Curriculum models and lan-guage facilitation in the preschool years. In M.E. Fey, J. Windsor, & S. F Warren (Eds.), Lan-guage intervention: Preschool through the el-ementary years.

Fey, M. E., Cleave, P L., Ravida, A., Long, S. H.,Dejmal, A., & Easton, D. (1994). Effects ofgrammatical facilitation on the phonologicalperformance of children with speech and lan-guage impairments. Journal of Speech andHearing Research, 37, 594-607.

Fey, M. E., & Proctor Williams, K. (2000). Re-casting, elicited imitation and modeling in

grammar intervention for children with specificlanguage impairments. In D. V. M. Bishop(Ed.), Speech and language impairments in

children: Causes, characteristics, interventionand outcome (pp. 177-194). Philadelphia, PA:Psychology Press.

Fudala, J., & Reynolds, W. (1986). Arizona Artic-ulation Proficiency Scale, 2nd ed. Los Angeles:Western Psychological Services.

Ingram, D. (1989). Phonological disability in chil-dren. London: Cole and Whurr.

Kent, R. D., Weismer, G., Kent, J., & Rosenbek, J.(1989). Toward phonetic intelligibility testingin dysarthria. Journal of Speech & HearingDisorders, 54, 482-499.

Koegel, R. L., Camarata, S., Koegel, L. K., BenTall, A., & Smith, A. E. (1998). Increasingspeech intelligibility in children with autism.Journal of Autism and Developmental Disor-ders, 28(3), 241-251.

Leonard, L. B., McGregor, K. K., & Allen, G. D.(1992). Grammatical morphology and speechperception in children with specific languageimpairment. Journal of Speech and HearingResearch, 35(5), 1076-1085.

Long, S., & Fey, M. E. ( 1995). Computerized pro-filing, Version 8.0. Cleveland, OH: Case West-ern University.

Miller, J., & Chapman, R. (1993). SALT: Systematicanalysis of language transcripts. Madison, WI:University of Wisconsin.

Nelson, K. (1989). Strategies for first languageteaching. In M. Rice & R. L. Schiefelbusch

(Eds.), The teachability of language (pp. 263-310). Baltimore: Brookes.

Nelson, K. E., Carskaddon, G., & Bonvillian, J. D.(1973). Syntax acquisition: Impact of experi-mental variation in adult verbal interaction withthe child. Child Development, 44(3), 497-504.

Roid, G., & Miller, L. (1997). Leiter International

49

Performance Scale-Revised. Wood Dale, IL:Stoelting, Co.

Shriberg, L. D., & Kwiatkowski, J. (1982). Pho-nological disorders III: A procedure for assess-ing severity of involvement. Journal of Speech& Hearing Disorders, 47, 256-270.

Shriberg, L. D., Kwiatkowski, J., Best, S., Hengst,J., & Terselic-Weber, B. ( 1986). Characteristicsof children with phonologic disorders of un-known origin. Journal of Speech and HearingDisorders, 51 (2), 140-161.

Smith, A., & Camarata, S. (1999). Increasing lan-guage intelligibility of children with autismwithin regular classroom settings using teacherimplemented instruction. Journal of PositiveBehavior Intervention, 1, 141-151.

St. Louis, K., & Ruscello, D. (1987). Oral speechmechanism screening exam-Revised. Austin,TX: Pro-Ed.

Stone, W. L., & Yoder, P J. (2001). Predicting spo-ken language level in children with autism

spectrum disorders. Autism, 5(4), 341-361.

Suen, H. K., & Ary, D. (1989). Analyzing quanti-tative behavioral observation data. Hillsdale,NJ: Lawrence Erlbaum Associates.

Tabachnick, B., & Fidell, L. (1996). Using multi-variate statistics (3rd ed.). New York: HarperCollins.

Tallal, P., Stark, R. E., & Mellits, D. (1985). Therelationship between auditory temporal analy-sis and receptive language development: Evi-dence from studies of developmental languagedisorder. Neuropsychologia, 23(4), 527-534.

Yoder, P, Camarata, S., & Gardner, E. (2005). Pre-dictors of later length of utterance and intelli-gibility in children with co-morbid speech andlanguage impairment. Manuscript in prepara-tion. Vanderbilt University, Nashville, TN.

Yoder, P, & Gardner, E. (2004). Familiarity effectson estimates of MLU and intelligibility in pre-schoolers with phonological and grammaticalimpairments. Submitted for publication.

Yoder, P J., Hooshyar, N., Klee, T., & Schaffer, M.( 1996). Comparison of the types of child utter-ances mothers expand in children with andwithout Down syndrome. American Journal ofIntellectual Disability Research, 40, 557-567.

Yoder, P J., Kaiser, A. P, & Alpert, C. L. (1991).An exploratory study of the interaction be-tween language teaching methods and childcharacteristics. Journal of Speech & HearingResearch, 34( 1 ), 155-167.

Yoder, P J., Kaiser, A. P, Goldstein, H., Alpert, C.,Mousetis, L., Kaczmarek, L., & Fischer, R.

(1995). An exploratory comparison of milieuteaching and responsive interaction in class-room applications. Journal of Early Interven-tion, 19(3), 218-242.

Yoder, P J., & McDuffie, A. (2002). Treatment ofprimary language disorders in early childhood:Evidence of efficacy. In P Accardo (Ed.), Dis-orders of language development (pp. 151-177).Baltimore: York.

Yoder, P J., & Warren, S. F (1998). Maternal res-ponsivity predicts the prelinguistic communi-cation intervention that facilitates generalizedintentional communication. Journal of Speech,Language, and Hearing Research, 41(5),1207-1219.

Yoder, P J., & Warren, S. E (2001 a). Intentionalcommunication elicits language-facilitating ma-ternal responses in dyads with children whohave developmental disabilities. AmericanJournal on Mental Retardation, 106(4), 327-335.

Yoder, P J., & Warren, S. E (2001b). Relative treat-ment effects of two prelinguistic communica-tion interventions on language development intoddlers with developmental delays vary bymaternal characteristics. Journal of Speech,Language, and Hearing Research, 44( 1 ), 224-237.

Zimmerman, I., Steiner, B., & Pond, R. (1992). Pre-school Language Scale- 3rd Edition. San An-tonio : Psychological Corporation.

This research was supported by the Scottish RiteFoundation of Nashville, a P50 grant from NIDCD(P50DC03282) and the J. F. Kennedy Center Coregrant from NICHD (HD15052). We are deeplygrateful to the parents who trust us with their pre-cious children. We also appreciate the hard andskillful work of Amiee Owen and Lisa Stepp, whoconducted the treatment and transcribed the lan-

guage samples. Finally, we appreciate the supportfrom the larger P50 assessment core staff, whichhelped recruit participants and transcribe languagesamples and AAPS results (i.e., Mary Camarata,Stephanie Cofer, and Jennifer Croslin-Smith). With-out all of this support, we could not do this re-

search.

Correspondence concerning this article should beaddressed to Paul Yoder, Peabody Box 328, Van-derbilt University, Nashville, TN 37203. E-mail:paul.yoder@vanderbilt.edu