Journal of Autism and Developmental Disorders

, 36:807

Residual Language Deficits in Optimal Outcome Children with a History of Autism

Authors

    • Department of PsychologyUniversity of Connecticut
    • Department of PsychologyQueen’s University
  • Jennifer J. Paul
    • Department of PsychologyUniversity of Connecticut
    • Department of PsychiatryUniversity of Colorado Health Sciences Center
  • Deborah Fein
    • Department of PsychologyUniversity of Connecticut
  • Letitia R. Naigles
    • Department of PsychologyUniversity of Connecticut
ORIGINAL PAPER

DOI: 10.1007/s10803-006-0111-4

Cite this article as:
Kelley, E., Paul, J.J., Fein, D. et al. J Autism Dev Disord (2006) 36: 807. doi:10.1007/s10803-006-0111-4
  • 560 Views

Abstract

This study examined whether language deficits persist even in children with optimal outcomes. We examined a group of children with prior diagnoses on the autism spectrum who had IQs in the normal range, were in age-appropriate mainstream classes, and had improved to such an extent that they were considered to be functioning at the level of their typically developing peers. Fourteen such children between the ages of five and nine were matched on age and sex with typically developing children, and were given a battery of 10 language tests to investigate their language abilities. Results indicated that while these children’s grammatical capabilities are mostly indistinguishable from their peers, they are still experiencing difficulties in pragmatic and semantic language.

Keywords

OutcomeHFALanguageSemanticsPragmatics

Poor language and communication skills comprise a defining feature of Autism Spectrum Disorder (ASD; APA, 2000), and most children with ASD receive interventions that focus largely on building these skills. Recently, the prognosis for children with ASD receiving intensive early intervention appears to have improved substantially. It has been reported that such children show significantly reduced problematic behavior, and improved language, cognition, and social interaction skills (Harris & Handleman, 2000; Jocelyn, Casiro, Beattie, Bow, & Kneisz, 1998; Lord, 1996). However, current outcome research on these children has reported only global cognitive measures or classroom placement; no one has yet investigated the language outcomes of these children in detail. Such an investigation seems warranted because global measures can mask both abilities and deficits in specific areas; for example, it is well known that language difficulties remain in children with Down syndrome, particularly in the area of grammar, despite years of intervention (e.g., Fowler, 1990; Fowler, Gelman, & Gleitman, 1994). Moreover, the contributing factors to good versus poor outcomes cannot be firmly established until their links with specific aspects of stable language behavior have been demonstrated. In this paper, we report the first such investigation of the language-related developmental outcomes of a group of “optimal-outcome” children who had been diagnosed with ASD during their preschool years; they are tested in grade school, after the cessation of their intervention.

Research over the past 20 years has revealed that a number of factors contribute to the overall developmental outcomes of children with ASD (Fenske, Zalenski, Krantz, & McClannahan, 1985; Gabriels, Hill, Pierce, Rogers, & Wehner, 2001; Lovaas & Buch, 1997). Most of the relevant studies compare the same children at two different times, which are anywhere from 2.5 years to 7 years apart; thus, the children are typically 3–6 years of age at time one (close to the initial diagnosis and onset of treatment) and 5–13 years of age at time two (i.e., the outcome). Children entering treatment at younger ages have been found to have higher global IQs and/or better school placements at time two than children entering treatment at older ages (Bibby, Eikeseth, Martin, Mudford, & Reeves, 2002; Fenske, et al., 1985; Harris & Handleman, 2000). An advantage for time two IQ has also been found for children whose time one IQs were relatively higher (Gabriels, et al., 2001), and an advantage for time two standardized language scores has been found for children whose initial verbal abilities were stronger (Szatmari et al., 2000). Finally, children who received intensive behavioral treatment for longer hours and/or more months had higher IQs and better school placements than children who received treatment for fewer hours/months (Gabriels et al., 2001; Lovaas, 1987; McEachin, Smith, & Lovaas, 1993). Further scrutiny of these studies reveals that the variability of outcomes in children with ASD encompasses a huge range: some children demonstrate no change in their already low IQ scores from time one to time two whereas others have been found at time two to score within the normal range, and/or have improved to such a great extent that they can be mainstreamed into chronologically age-appropriate classrooms in elementary school (Fenske et al., 1985; Gabriels et al., 2001; Harris & Handleman, 2000; Lovaas, 1987; McEachin et al., 1993). Full understanding of the causes of this variability, though, is hampered by the fact that the measures used to evaluate ASD outcome have been global rather than specific; most have just assessed children’s IQ and/or their school placement. Detailed examination of cognitive functioning at outcome, especially of language functions, is crucial to understanding the degree to which their cognitive impairments have truly resolved, and to providing sensitive measures that can be used in future studies on predictive factors.

The problem of assessing and accounting for language-related developmental outcomes is complicated by findings that different sub-areas of language are differentially affected in ASD. In a recent literature review, Tager-Flusberg (2001) identified three components of language, computational (i.e., grammar), semantic, and pragmatic, and concluded that only the last was consistently impaired in individuals with autism.1 Thus, any assessment of language outcomes in ASD will need to distinguish the pragmatic aspects of language. Moreover, even within the grammatical and semantic components, specific strengths and weaknesses are evident. For example, whereas Tager-Flusberg et al. (1990) and Tager-Flusberg (1994) have found that children with ASD displayed patterns of Mean Length of Utterance (MLU) increase, pronoun case development, and question form development similar to typically developmentally mental-age-matched controls (see also Waterhouse & Fein, 1982), children with ASD have also been found to use a much more limited range of morphological and syntactic forms in their spontaneous speech: They produce fewer prepositions, conjunctions, articles, verb tenses, and auxiliaries, and they use very little complex syntax, including embedded sentences, sentence complements, and relative clauses (Bartolucci, Pierce, & Streiner, 1980; Cantwell, Baker, & Rutter, 1978; Fein et al., 1996; Menyuk & Quill, 1985; Scarborough, Rescorla, Tager-Flusberg, Fowler, & Sudhalter, 1991). Within the lexical realm, children with ASD have exhibited the same type of semantic-prototype organization in categorization at the basic and superordinate levels as language-matched controls, and consistently perform relatively well on standardized vocabulary tests (Kjelgaard & Tager-Flusberg, 2001; Tager-Flusberg, 1985, 2001; Waterhouse & Fein, 1982). However, children with ASD, unlike typically developing controls, do not recall lists of words in meaning-based chunks, nor do they turn them into a sentence to make recall easier (Hermelin & O’Connor, 1970; Tager-Flusberg, 1988; see also Dunn, Gomes, & Sebastian, 1996). Furthermore, their verbal memory deficits are increasingly apparent with material of increasing semantic organization (e.g., digits to sentences to stories) (Fein et al., 1996) Thus, it is possible that children with autism understand and/or store the meanings of words differently from typically developing children (Dunn & Rapin, 1997).

Within the pragmatic realm, where individuals with autism routinely experience their most severe difficulties, consistency across studies is greater (Dunn & Rapin, 1997; Fein et al., 1996; Kjelgaard & Tager-Flusberg, 2001; Prizant, 1996; Tager-Flusberg, 1996). Children on the autism spectrum have particular difficulty in responding to questions, sharing information, and requesting information (Tager-Flusberg, 1996; Wetherby, Prizant, & Schuler, 2000). They have great difficulty with false belief tasks, which are designed to test their understanding of Theory of Mind (Baron-Cohen, Leslie, & Frith, 1985; Happe, 1995; Tager-Flusberg, 1996, 1999). They also have difficulty with the production of narratives, in that children with ASD have been consistently found to use more bizarre language in their narratives than do mental-age-matched controls (Loveland, McEvoy, Tunali, & Kelley, 1990; Losh & Capps, 2003), and frequently use fewer causal connectives and a narrower range of narrative devices (Capps, Losh & Thurber, 2000; Losh & Capps, 2003; Tager-Flusberg, 1996).

In sum, any study of the language-related developmental outcomes of children with ASD will need to assess multiple aspects of grammar, lexicon, and pragmatics in order to present a complete picture of the children’s abilities and deficits. Moreover, a comparison of the research on ASD outcome and language reveals reciprocal gaps in these literatures: As previously mentioned, studies that focus on the developmental outcomes of children with ASD have not investigated any aspects of their language in detail. Studies that have investigated in detail the language of children with ASD have not studied children who have reached the “optimal outcome” status of being fully mainstreamed with no educational supports. A third set of studies, which focus narrowly on the effects of specific types of training, do discuss both type of intervention (e.g., discrete trial learning or milieu learning) and specific aspects of language (e.g., learning spatial prepositions); however, these studies typically only include a very few participants (ns ranging from 1 to 8) and only a narrow range of outcome span (8 days to 8 months) and outcome measures (see Goldstein, 2002, for a review).

The purpose of the current study is to begin to fill these gaps in the research literature. We investigate in detail the language outcomes of a group of children for whom we have both early diagnosis and treatment information. These children were previously diagnosed with ASD between one and five years of age, were treated in intensive behavioral programs for between one and four years, and are now considered to be functioning successfully in a typical school environment. We chose to focus on this “optimal-outcome” group for several reasons: First, children such as these have never been examined in detail for their language strengths and weaknesses. We wished to determine the degree to which their language appears as “recovered” as other aspects of their behavior. From a practical standpoint, it is important to discover whether such children might still need some specific remedial assistance. Second, choosing children who are functioning academically within the normal range minimizes the heterogeneity of the usual autistic sample. While these children cannot be considered to have been ‘modal’ children with autism, their increased homogeneity with regards to diagnosis, treatment, and general outcome can only benefit the comparison with typical children. Third, the increased age and functioning of these children enables us to assess more complex aspects of their grammar and lexicon. For example, we can investigate whether their grammatical development (e.g., mastery of multi-clause syntax) has kept pace with their vocabulary growth, and whether their vocabulary growth has also led to increased understanding of the power of words to categorize (e.g., categorical induction). These are abilities manifested by typical children of four to five years of age.

We tested these children’s language with a battery of psycholinguistic and standardized tasks, which assessed morphological, syntactic, lexical semantic, and pragmatic functioning. The children were given standardized tests to enable comparisons of their abilities with our typical control group, as well as psycholinguistic tasks that had been specifically developed to tap underlying linguistic knowledge in preschool-aged typically developing children while minimizing performance demands. Some tasks examined those properties of language that have been disputed with this population, including morphology and complex syntax within the grammatical realm, and the induction of category properties based on shared labels within lexical semantics. Other tasks tapped areas of language in which deficits or strengths have been found consistently, such as mental verb discrimination (Kazak, Collis, & Lewis, 1997; Lord, 1996; Ziatas, Durkin, & Pratt, 1998), narrative production (Capps et al., 2000), verb argument structure (Tager-Flusberg et al., 1990) and Theory of Mind tasks. Although Theory of Mind tasks do not assess language ability directly, these tasks have been shown to be strongly related to language ability (Happe, 1995). Our control group was chosen to be typically developing children who were matched on age (12 of the 14 were also matched on receptive vocabulary), because the null hypothesis is that these children with a history of autism are indistinguishable from their typically developing peers (e.g., Lovaas, 1987; McEachin, et al., 1993). Each psycholinguistic task we have chosen elicits above chance performance with typically developing children aged four-to-five years; therefore, comparisons will be made with these findings from the literature as well.

Method

Participants

The children with a history of ASD (called here the ASD group) were recruited from the clinical files of the third author, a clinical and research neuropsychologist specializing in autism and related disorders. Extensive clinical files were searched to identify children with clear diagnoses of PDD-NOS, Asperger’s, or Autistic Disorder who had achieved optimal outcomes. All of the children had originally been diagnosed on the PDD spectrum as toddlers and still met criteria for a PDD diagnosis when evaluated by the third author (see Table 1 and Appendix A).
Table 1

Diagnostic information from third author

Child (random initials)

Age at Dx by author in months

Diagnosis given by author at this time

Original diagnosis

Age at current testing in months

T.C.

38

PDD-NOS

Autistic disorder

60

C.F. (f)

13

PDD-NOS

PDD-NOS

65

D.K.

64

Asperger’s

Autistic disorder

77

P.B. (f)

44

PDD-NOS

Autistic disorder

79

M.K.

49

PDD-NOS

Autistic disorder

80

C.B.

58

Autistic Disorder

PDD-NOS

82

N.E.

77

PDD-NOS

Autistic disorder

86

L.X.

52

PDD-NOS

PDD-NOS

91

M.S.

51

PDD-NOS

Autistic disorder

93

T.S.

54

PDD-NOS

PDD-NOS

95

T.M.

73

Asperger’s

Asperger’s

98

U.N.

83

PDD-NOS

PDD-NOS

100

T.B.

90

Autistic Disorder

Autistic disorder

107

Q.K.

92

PDD-NOS (mild)

PDD-NOS

109

f = female

See appendix for a full description of these children’s autistic symptoms and language functioning at the time given here

Table 1 indicates the age at which the third author first evaluated the child and the diagnosis given, as well as their current age. The children were diagnosed according to DSM-IV criteria (using a checklist with all DSM-IV symptoms) using extensive parent interview, child testing, and child observation.

In the Appendix there is a description of the DSM-IV symptoms displayed at that time, and notes on language development. The description in the Appendix indicates whether formal criteria were met for Autistic Disorder or not at the time of the third author’s evaluation. Eight children met formal criteria for Autistic Disorder and six for PDD-NOS. Note that in several cases, formal criteria for Autistic Disorder were met but a diagnosis of PDD-NOS or Asperger’s Disorder was given to the family. The reason for this was that the child had previously been given a diagnosis of PDD-NOS or Asperger’s and had continued to improve so that symptoms were fewer or milder than they had been. It was considered to be counter-productive to give the parents a more severe diagnosis when the child had actually shown significant improvement, and in many cases, the symptoms were present but in relatively mild form. For the youngest child (13 mos.) a diagnosis of PDD-NOS was given; symptoms were not mild, but some symptoms (conversation, stereotyped play, stereotyped language, resistance to change) were not developmentally appropriate to score.

All of the children underwent intensive intervention programs at an early age, consisting of Applied Behavior Analysis (ABA), except for the one child with Asperger’s syndrome (see Table 2 for the children’s age of initiation and termination of treatment, and type of treatment).
Table 2

Participant characteristics

Child

Age at initiation of treatment*

Age at termination of treatment

Type of Program (ABA unless otherwise noted) and hours per week

T.C.

2 years

4 years

10 at school

C.F.(f*)

1 year, 2 months

4 years, 5 months

40 at home/school

D.K.

5 years

Ongoing at home

10 at school, 10 at home

P.B.(f*)

3 years

6 years

30 at school

M.K.

4 years, 5 months

5 years, 6 months

20 at school

C. B.

2 years, 9 months

6 years

20 at school, 15 at home

N.E.

3 years

5 years, ongoing at home

40 at home/school, then a few hours at home

L.X.

3 years, 1 month

5 years, 1 month

40 at home/school

M.S.

3 years

6 years

40 at home/school

T.S.

3 years

7 years

40 for 3 years, 15 for one year

T.M.

4 years

7 years

Special class at preschool

U.N.

3 years, 3 months

6 years, 4 months

40 at home/school

T.B.

4 years, 6 months

6 years

30 at home/school

Q.K.

3 years

6 years

Floor time/PECS/ABA 20 at home

* f = female

Approximate hours of treatment (hours per week × number of weeks)

Range = 1000–6750 h

Mean = 4054.62 h

SD = 2130.16 h

At the time of the study, the children had all been mainstreamed into chronological age-appropriate classrooms, and were considered to be functioning at the level of their peers by parents and teachers. In many cases, the child’s current teacher was not aware of the history of autism/PDD. Two children (see Table 2) continued to receive educational supports in the form of several hours per week of ABA at home to reinforce academics and work on language pragmatics but none received educational supports or ABA in school. At the time of the current study, the children ranged in age from 5 years 6 months to 9 years 1 month (M = 7;3, SD = 14 months). Twelve were boys and two were girls, a sex ratio that is typical for children with this disorder. All of the children resided in suburban or rural areas of Massachusetts.

The ASD group was matched on age and sex with a group of typically developing children (TD) (12 boys, two girls, mean age = 7;4, SD = 15 months; range 5;10 to 9;1). Although it was not our prior intention, the two groups ended up being matched on standard scores of the TACL vocabulary. The TD group was recruited through a primary school in suburban/rural Connecticut; the children’s parents volunteered at a school PTO meeting where the purposes of the study were explained. The TD children were given small toys for their participation. All children were matched within a six-month age range of their counterparts in the experimental group. There were no significant differences between the groups in age (t(26) = −0.50, p = .96). All of these children were in age-appropriate classrooms and none of them were receiving special educational services at school.

Language assessments

The children were given 10 different language tests, described below:

The Test for Auditory Comprehension of Language, Third Edition (TACL-3) (Carrow-Woolfolk, 1985). This test assessed three categories of language understanding: Vocabulary, Grammatical Morphemes, and Elaborated Phrases and Sentences. Children were shown a page with three pictures on it, given a linguistic stimulus, and asked to point to the matching picture.

The Expressive One-Word Picture Vocabulary Test (EOWPVT) (Gardner, 1990). This task requires the child to name a pictured stimulus, providing a standardized measure of expressive vocabulary.

The Stanford-Binet Memory for Sentences Subtest (Thorndike, Hagel, & Satler, 1986) assessed children’s verbal memory by asking children to repeat increasingly complex sentences.

The Wug Test of Productive Morphology (Berko, 1958). The Wug Test used nonsense words and pictures to assess children’s ability to generalize the basic rules of English inflectional morphology. The 22 stimuli included six tests of the plural, nine of the past tense, four of the possessive, two of the present tense, and one of the present progressive. All of the children received the test stimuli in the same order. They were shown a picture of an unusual fictional creature and told, “This is a Wug.” They were then shown another picture of the same type of creature and told, “Now there are two of them. There are two _____.” Children had to complete the statement by saying “Wugs” in order to receive credit for a generalization. Berko (1958) found that typically developing children as young as four years were able to make such generalizations regularly (see also Kim, Marcus, Pinker, Hollander, & Coppola, 1994).

Understanding of Complex Syntax (deVilliers & Roeper, 1995). This task investigated the extent to which the children understood how the syntactic phenomenon of “wh-movement” is affected by the presence of different types of subordinate clauses (e.g., Chomsky, 1982). That is, the wh-question, “When did the girl say she planted the pumpkin?” has two possible readings (i.e., When did she say?/When did she plant?), whereas the wh-question “How did the mother learn what to bake?” has only one (How did she learn?). The presence of the medial wh-word in the latter sentence promotes the reading of the matrix or main verb over the subordinate one. Our test sentences contained 12 embedded sentence complements, either with (six sentences), or without (six sentences) medial wh-words.

Children were told that they would be read some stories and at the end of each story there would be a question. The stories were told in the same order to all of the children. The responses were coded as to whether the children had answered the first or second clause of the sentence. deVilliers and Roeper (1995) reported that three-year-olds responded significantly differently to questions with and without the medial wh-word.

Verb Argument Structure (Naigles, Gleitman, & Gleitman, 1993). This task assessed children’s understanding of verb argument structure by investigating how they enacted ungrammatical sentences, that is, sentences in which there were too many or too few noun arguments. Children were given a wooden ark and a number of wooden animals that they used to act out sentences spoken by the experimenter. The children were given twenty test sentences to enact; twelve were grammatical and eight were ungrammatical. The grammatical sentences (e.g., “The elephant pushes the zebra.”) ensured that children understood the task: those who did not act out at least eighty percent of the grammatical sentences correctly were eliminated from the analyses.2 Of the eight ungrammatical sentences, four included transitive verbs in intransitive frames (e.g., *The lion brings), and four included intransitive verbs in transitive frames (e.g., *The tiger comes the horse). The sentences were read in the same order for all the children.

Coding was performed as previously reported (Naigles et al., 1993; Naigles, Fowler, & Helm, 1992). The children’s enactments of the ungrammatical sentences were coded as either: (a) Frame Compliant, in which the children would enact the sentence according to the meaning of the frame (the immature response), (b) Verb Compliant, in which the children would enact the sentence according to the meaning of the verb (the mature response), or (c) Other, used when the child either seemed to mishear the experimenter (e.g., using a combing motion when they heard the verb come) or used the wrong animals, or when it was unclear what the child was doing.3 Only seven (out of 104) enactments were coded as Other; this is comparable to other results for children in this age group (Naigles et al., 1992). Naigles et al. (1992, 1993) found that children aged two to four years were more likely to behave Frame Compliantly whereas children five years and older were increasingly likely to behave Verb Compliantly.

Categorical Induction (Gelman & Markman, 1986). This task asked children to make a prediction about the property of a depicted natural kind object. The major question concerned whether the prediction (i.e., induction) would be made on the basis of name similarity. Eight natural kinds were examined, four of which were inanimate (e.g., a rock) and four animate (e.g., a rabbit). For each natural kind, children were shown the picture, told its name, and then told one of its properties. Except for the first two children tested (C.B. & U.N.), the children were asked to repeat back the name of the object and its special property to ensure they were paying attention. Then they were asked whether the next items also had that property. Four test cards were presented for each natural kind: (1) the exact same natural kind with a slightly different perceptual appearance (coded as same); (2) one of the same natural kind but of a different color or form (target); (3) a perceptually similar object which was not of the same kind (perceptual); and (4) a distracter item which did not bear any relation to the original natural kind other than whether it was animate or inanimate (distracter). Children were required to answer yes or no for each test card; if they said “maybe” they were asked to take their best guess. When the child refused to commit to an answer (n = 4 instances) these were always scored as incorrect. Both the order within and across test trials was counterbalanced across participants. Gelman and Markman (1986) reported that three-year-old and adults consistently make predictions based on name similarity.

Certainty differences with mental state verbs (Moore, Bryant, & Furrow, 1989). This task ascertained children’s understanding of the certainty differences between think and guess as opposed to know. The materials included two puppets, a cat and a cow, and a number of white, yellow, and blue boxes. A three-trial pretest ensured that the children could distinguish the utterances of the two puppets and use these utterances to select the location of each sticker. All of the children responded correctly in at least two out of the three pretest trials.

During the test trials the child was presented with two boxes on each trial and given clues by the puppets. The puppets used the verbs think, guess, and know to designate the likely location of the sticker. For example, the cat puppet said, “ I guess it’s in the blue box,” whereas the cow puppet said, “I know it’s in the white box.” Children were again asked to select the box holding the sticker. They were then told to put their box choice off to the side and wait until the end of the 12 trials to discover how many stickers they had obtained. The children all received the test trials in the same order. Four tests trials contrasted think and know, four contrasted think and guess, and four contrasted guess and know. Only the think/know and guess/know trials were tabulated for this analysis, as Moore et al. found that the think/guess distinction was not reliably understood until the age of eight. Moore et al. (1989) reported that five-year-old performed well above chance on the think/know and guess/know distinctions.

Theory of Mind tasks (Wimmer & Perner, 1983; Perner, Leekham, & Wimmer, 1987). Theory of Mind tasks are designed to assess whether children understand that other people can have a false belief. Two different tasks were used to assess the children’s Theory of Mind capabilities.

The first of these tasks was called the Unexpected Location task and was a variation of the classic Maxi task (Wimmer & Perner, 1983). Children were introduced to a puppet, Astro, who said that she was going on a trip and needed to take her toy monkey with her. Astro then asked the children to help by putting her toy monkey in a blue box, which would function as her suitcase. She then remarked that she had forgotten snacks for her trip and left for the store. The experimenter then conspiratorially asked the children if they would like to play a trick on Astro, and the children and the experimenter would then move the toy monkey from the blue box to the white box. The children were then asked the target question of where Astro would look for the monkey when she came back from the store. The children were also asked two control questions: where the monkey was now, and whether or not Astro had seen the monkey being moved. In the original task (Wimmer & Perner, 1983), children simply viewed puppets engaging in deception and being deceived. The current version of the task allowed much more involvement of the children. It should be noted, however, that this version of the task nonetheless has been found to yield similar developmental results (Mayeux, 2000).

The second Theory of Mind task was called the Unexpected Contents task (Perner et al., 1987). Children were shown a typical metal “band-aid” box and asked what they thought was inside; every child in this study said “band-aids”. The experimenter then opened the box and showed the children that there were really balloons inside. The children were given a balloon, the box was closed, and the children were asked what was really inside as a control question. The children were then asked two target questions: what they thought was inside before the box was opened, and if the box had been shown to their best friend, what their friend would have thought was inside the box. The children always received the unexpected location task before the unexpected contents task and all the test questions were given in the order noted for all participants. Both tasks were scored for overall scores, and also for only the target questions addressing their understanding of those and other’s mental states.

Narrative Capability (Capps et al., 2000; Tager-Flusberg & Sullivan, 1995). This task asked the children to relate a wordless picture book. The wordless picture book used for this task was “Frog, Where are You?” (Mayer, 1969). This book contains a story about a boy who loses his pet frog and encounters many adventures on the search for his frog, which he finds at the end of the story. The child and the experimenter first looked through the book together, page-by-page, but silently. The experimenter turned the pages for the children at this point to ensure that they did not move through the pictures too quickly. Children were then asked to tell the story from the beginning, in their own words, looking again at the book. While children were telling the story the experimenter would nod his/her head in agreement or give non-descript positive feedback (e.g., “mm-hmm” or “really?”). If the child did not give much information about one page, or hesitated before turning the page the experimenter would ask “Anything else?” No specific leading questions were asked of the children.

The stories related by the children were then transcribed from the videotapes and coded for a number of different variables. General lexical variables such as number of words (tokens), number of different words (types), and the type-token ratio were compared between groups. More grammatical aspects were also examined, such as the number of clauses, number of connectives, number of tense changes, as well as various grammatical omissions such as omitted determiners and pronouns. Finally, the stories were coded for pragmatic variables such as causal attributions, mental state verbs, and dramatic devices such as intensifiers and sound effects (see Table 3 for complete explanations of all Narrative variables examined).
Table 3

Narrative variables

(1)

Word Tokens: Number of overall words in the narrative.

(2)

Word Types: Number of different words in the narrative.

(3)

Type/Token Ratio: Types over Tokens-a measure of lexical fluency.

(4)

Number of Clauses: Number of simple clauses. All clauses contained at least one verb.

(5)

Connectives: Number of words used to connect one clause to another (e.g., because, so, then, and then, finally, etc.) Numbers 1–5 were combined to create a general lexical variable.

(6)

Complex Connectives: Percent of connectives used other than “then” or “and then”.

(7)

Tense Changes: Number of times child changed tense while telling the story, for example, the child may have been telling the story in the past tense and switched to the present tense.

(8)

Verb agreement errors: When the child used the wrong form of the verb, that is, one that did not agree with the number of the noun (e.g., the boy think)

(9)

Over-regularizations: The child used a grammatical rule on an irregular form of a verb or an irregular pluralization (e.g., the frog goed)

(10)

Anaphora: The child used a pronoun without it being clear whom the referent was referring to. For example, if the child was talking about the boy and the dog in one sentence, and then merely said he in the next sentence, it would be unclear whom the child was referring to (unless the child had established the dog was a girl of course).

(11)

Omitted morphemes: this is a composite variable that examined the number of omitted: determiners, pronouns, prepositions, and auxiliaries. A sentence with an omitted determiner would be, “the boy thought dog was gone” where dog is not being used as a proper name so needs a determiner. A missing pronoun was counted when the child omitted either subject or object pronouns, e.g. , “and looked for his frog” where there should be a he in subject position. Omitted prepositions were coded for in instances such as, “ And he went the forest”, where the child should have inserted into or toward before the forest. Finally, an omitted auxiliary was coded for utterances such as “the boy going to find the frog”, where the child needed to insert the auxiliary is or was before the verb. Numbers 6–11 were combined to create a grammatical errors composite variable.

(12)

Intensifiers: Narrative devices used to stress particular aspects of the story, or make the story more dramatic. These include a) hedges (I’m not sure the boy knew what he was doing), negatives (the boy did not know those were deer’s antlers), and repetition for dramatic effect (the boy looked and looked and looked for his frog).

(13)

Emotion and mental state verbs: any of these spoken by the child was calculated.

(14)

Sound effects: things such as shouting, and using onomatopoeia.

(15)

Speech: Any time the child spoke for the boy, dog, or various other animals. Numbers 12–15 were combined to create a narrative devices composite variable

(16)

Misinterpretation: instances where the child incorrectly understood the scene, such as thinking that the deer’s antlers are branches of a tree.

(17)

Repetition: repetition of information that is not done for dramatic effect.

(18)

Unclear reference: When what the child is saying has nothing to do with what is on the page of the storybook, e.g., boy, girl, dog, frog, boy, girl, dog, frog. Numbers 16–18 were combined to create a misinterpretation of the story variable. The last two variables (numbers 19 and 20) were not included in any composite variables.

(19)

Causal discussion: Any time the child discusses the cause of an action or occurrence, e.g., the boy fell down from the tree because an owl flew out

(20)

Goal discussion: Any time the child discusses the goals of the characters, such as the overall goal of the boy looking for his frog

Vineland Adaptive Behavior Scales (Sparrow, Balla, & Cicchetti, 1984). Parents of the ASD group were interviewed with the Vineland Adaptive Behavior Scales to ensure that the children were in fact functioning within the normal range in the Communication and Socialization domains.

Procedure

All of the children were tested in their homes over the course of two sessions. The two sessions for the ASD group occurred on the same day, at least three hours apart. The two sessions for the typically developing children (TD) occurred on two different days, ranging from 2 days to 31 days apart (M = 10.29 days, SD = 7.32). The average session lasted 52.48 min for the ASD group (SD = 9.98), and 45.81 min for the TD group (SD = 5.74) (this difference was not statistically significant). Four different orders of the tests were generated; participants were randomly assigned to one of these four orders.

The children were taken to a quiet room in their home, and sat at a table with the experimenter who administered the tests. Another experimenter was also present to score the standardized tests and manipulate the video camera. Parents were allowed to observe their child’s sessions but were not permitted to participate. As they were accustomed, the children in the ASD group were reinforced liberally for staying on task in the testing sessions with various edible reinforcers. They were also given a token to put on a token board at the end of each test and were verbally praised in a non-directive manner as well (i.e., mild positive praise regardless of the correctness of their answers). The typically developing children were not deemed to need or be used to the same motivational enhancers; thus, only verbal praise was used with these children.

All of the tasks were recorded on videotape. With the exception of the TACL-3 (for which the experimenter noted the correctness of the answers as the tasks proceeded), all tasks were coded from the videotape. Ten percent of the data from each task was coded by two trained observers; reliability ranged from 92% to 100%, with an average of 98% reliability.

Results

The first data analyses examined whether the children in the TD and ASD groups performed differently on each task. Unpaired two-tailed t-tests were performed comparing the children in the two groups (see Johnson & Carey, 1998, for a similar strategy). Twelve of the children in each group could also be matched, within one point of their raw scores, on the receptive vocabulary subtest of the TACL-3. Paired-sample t-tests were also performed for each task on these vocabulary-matched individuals to examine the extent to which the ASD children performed more poorly than their TD language controls. The second set of analyses examined whether the ASD children’s performance on each task was related to their Communication and Socialization subscale scores on the Vineland. As the scores on the psycholinguistic tasks are essentially raw scores, all correlations were performed on both the raw (controlling for age) and the standard scores of the standardized tests and the Vineland. Correlations were performed to examine the relationship between estimated length of treatment and scores on the various tasks. In addition, correlations between all tasks performed by the children were analyzed separately by group to determine if the patterns of relations between the tasks would be similar between groups.

Standardized tests

Table 4 presents the standard scores of the standardized language tests for both groups.
Table 4

Standardized IQ measures

Test standard scores

Mean (SD) ASD group N = 13

Mean (SD) TD group N = 14

Equality of variance

t-test probability

TACL-3 Vocabulary (mean = 10)

10.54 (3.26)

11.14 (2.63)

F = .679 p > .05

t(25) = −.533 p > .05

TACL-3 Grammatical Morphemes (mean = 10)

9.77 (3.61)

12.29 (1.90)

 = 3.662  = .067

t(18)   = −2.292∼ p < .05

TACL-3 Elaborated Sentences (mean = 10)

9.00 (4.62)

12.57 (1.70)

 = 8.228 p < .01

t(15) = −2.707∼ p < .05

Expressive Vocabulary (EOWPVT) (mean = 100)

108.08 (30.14)

118.14 (14.22)

 = 2.754 p > .05

t(25) = −1.123 p > .05

Stanford-Binet Sentence Memory (mean = 50)

47.79 (8.89)

54.86 (8.61)

 = 0.022 p > .05

T(26) = −2.137 p < .05

The ASD group appeared to be performing similarly to their peers on the vocabulary portions of the standardized language tasks. Note, however, that on the TACL-3 Grammatical Morphemes and Elaborated Sentences subtests, and the Sentence Memory task, the ASD group performed more poorly than the TD group, although their standard scores were still within the normal range.

The Wug test of productive morphology

Both groups performed comparably on the Wug task. T-tests revealed no significant differences between the groups (ASD: = 61.69% correct, SD = 24.36; TD: = 71.43% correct, SD = 31.00). Note that in Berko’s (1958) original study, the first graders obtained an average of 68.59% correct; thus, the current sample, with an average age of seven, are roughly comparable.

Complex syntax

Table 5a presents the percent correct on the Complex Syntax task across sentence types.
Table 5 

(a) Percent correct on the complex syntax task. (b) percentage of children answering which clause

Percent Correct

Mean (SD) ASD group

Mean (SD) TD group

Equality of Variance

t-test

Percent correct on the complex syntax task

Across all Sentences

57.74 (28.95)

72.62 (15.13)

F = −3.254 p > .05

t(26) = −1.704 p > .05

Sentences with medial wh-words

36.90 (37.08)

54.17 (29.01)

F = 2.749 p > .05

t(26) = −1.372 p > .05

Without medial wh-words

78.57 (30.96)

90.48 (12.60)

F = 5.593 p < .05

t(24) = −1.333∼ p > .05

Percentage of Children answering which clause

ASD group with no medial wh-words

When did the girl say ___23%______ she planted the pumpkin___77%____?

with medial wh-words

How did Rover learn____58% _____ what to catch___42%___?

TD group with no medial wh-words

When did the girl say ___29%______ she planted the pumpkin__71%_______?

with medial wh-words

How did Rover learn____62%_____ what to catch___38%______?

∼ corrected for unequal variance

T-tests revealed no significant differences between the groups. Table 5a also shows the children’s percent correct for the sentence complements involving medial wh-words and those without medial wh-words. Again, no significant differences were found between groups on either sentence type. Similarly to deVilliers and Roeper’s (1995) findings, both groups had significantly more difficulty with those sentences including medial wh-words than those without (ASD t(13) = −4.301, < .001; TD t(13) = −3.970, < .01).

Children’s answers were then analyzed to determine whether they addressed the question of the first or second phrase. These data are presented in Table 5b. Results demonstrated that when there was no wh-word to restrict interpretation of the question, both the ASD group and the TD group answered the second clause more of the time (i.e., when the girl planted, not when she said she planted). However, when the medial wh-word was present, both groups answered the first clause significantly more (i.e., when the dog learned, not what he caught) (ASD: t(13) = −2.28, p < .05; TD: t(13) = −4.47, p < .001).

Verb argument structure

Both groups of children were significantly more Frame Compliant with the transitive ungrammatical sentences than the intransitive ungrammatical sentences; this is a pattern that has consistently been seen across age groups (Naigles et al., 1992, 1993; Naigles, Fowler, & Helm, 1995; Naigles & Lehrer, 2002), and was significant for both groups (ASD: t(12) = 4.43, p < .01; TD: t(12) = 9.90, p < .001). The ASD group tended to be more Frame Compliant than the TD group with the ungrammatical intransitive sentences (ASD = 29.17% (35.09), TD = 8.3% (12.5); t(20) = 1.90, p = .077), and performed significantly more Frame Compliantly with the ungrammatical transitive sentences (ASD = 82.5% (16.87), TD = 66.67% (16.28); t(20) = 2.23, p < .05).

Categorical induction

For the Categorical Induction task, the first analysis was conducted to compare the two groups’ overall scores (see Table 6a). A t-test comparing the number of correct answers approached significance. As described above, for each natural kind, four objects were presented. Of greatest interest is whether or not the child is able to induce the properties of another object of the same natural kind (the target object). When the target objects alone were examined for differences between the groups, this comparison also approached significance. For the stimuli that were of the same natural kind and only slightly perceptually different (same), the t-statistic also approached significance, with the TD group obtaining more correct answers. The perceptual stimuli, which looked similar to the test stimuli but was of a different category, did not show significant differences between the groups, nor did the distracter stimuli.
Table 6

(a) Overall percent correct for variables in the categorical induction task. (b) Percent correct for animate items

 

ASD Mean (SD)

TD Mean(SD)

Variance

t-test

Sig.

(a) Overall percent correct for variables in the categorical induction task

Across items

74.04 (13.46)

82.03 (5.51)

F= 8.602 p < .01

t(16) = −1.911∼

p = .066

Target

52.68 (28.66)

70.19 (15.76)

F = 2.037 p > .05

t(25) = −1.945

p = .063

Same

83.04 (19.98)

94.64 (6.42)

F = 7.991 < .01

t(16) = −2.069∼

p = .055

(b) Percent correct for animate items

Overall

70.68 (19.64)

81.99 (12.91)

F = 3.842 > .05

t(26) = −1.801

p = .083

Target

44.64 (38.20)

73.21 (22.92)

F = 4.489 p < .05

t(21) = −2.400∼

p < .05

Same

82.14 (24.86)

96.43 (9.08)

F = 13.50 p < .001

t(16) = −2.020∼

p = .060

Because of the reported differences in how autistic children view animate versus inanimate objects (Fay & Schuler, 1980), the data were collapsed into groups divided along this dimension, and then separated into target and same (perceptual and distracter items were not included because of the similarity of the ASD and TD responses in the overall analyses). T-tests between groups showed the same pattern as above with the animate stimuli: the ASD group performed significantly more poorly on the animate target items, and the t-test comparing answers on the animate same approached significance (Table 6b). There were no differences between the groups for any of the inanimate stimuli. Given this pattern, repeated measures t-tests were conducted to determine if the ASD group would show significantly more difficulty with animate as opposed to inanimate items; all were non-significant.

Mental state verbs

The ASD group performed significantly more poorly than the TD group on the Mental State Verb task (t(26) = 3.02, p < .01). The ASD group performed correctly on the questions an average of 58.04% of the time (SD = 26.68%) whereas the TD group performed correctly 86.61% of the time (SD = 23.24%).

Theory of mind

When examining all three of the answers combined for the Unexpected Location task (whether the puppet saw the deception, where the toy monkey is now, and where the puppet will look for the toy monkey when she comes back), there were no significant differences between the groups (see Table 7). When only the target question was examined, however, a different picture emerged; the typically developing children performed significantly better than the ASD group.
Table 7

Percent correct on theory of mind tasks

Test

ASD group Mean (SD)

TD group Mean (SD)

Equality of variance

t-test probability

Unexpected

Location

85.67

95.33

F = 4.858

t(20) = −1.155∼

Overall

(28.40)

(12.10)

p < .05

p > .05

Unexpected

Location

71.00

100.00

F = 57.78

t(13) = −2.280∼

Target

(46.90)

(0.00)

p < .001

< .05

Unexpected

Contents

73.67

93.00

F = 3.811

t(26) = −2.162

Overall

(26.73)

(19.30)

p > .05

p < .05

Unexpected

Contents

73.67

93.00

F = 3.811

t(26) = −2.162

Target

(26.73)

(19.30)

p > .05

p < .05

∼Corrected for unequal variances

The Unexpected Contents task revealed similar results (see Table 7). The ASD group had a mean correct of 2.21 (out of three), with a standard deviation of 0.80, and a range from one to three. The TD group had a mean of 2.79 correct with a standard deviation of 0.58 and a range from one to three. These means were significantly different between the groups (t(23) = −2.16 p < .05), even after correcting for close to unequal variances (F(2,26) = 3.81, p = .062).

Narrative

In accordance with past research (Capps et al., 2000; Loveland et al., 1990; Tager-Flusberg & Sullivan, 1995), three different types of variables were examined with regards to the children’s narrative competence: general lexical variables, grammatical errors, and pragmatic variables. There were no significant differences on any of the general lexical variables such as number of words (tokens), number of different words (types), or the type-token ratio, nor were there any significant differences on any of the grammatical variables; the data from these measures can be requested from the authors. The general lexical variables and grammatical errors were analyzed both individually and as composite variables (see Table 3); neither individual nor composite analyses yielded any significant differences between groups.

Table 8 presents the findings by group for the pragmatic variables. No differences between the groups emerged on many types of narrative devices, including the number of intensifiers used, such as repeating for emphasis (e.g., “he searched and searched and searched for his frog”), hedges (e.g., “I think the boy was worried about his frog”), or negatives for dramatic effect (e.g., “the boy didn’t know that those were deer’s antlers”). There were also no differences between the groups on the number of emotion and mental state verbs used in the narrative; both groups used very few. The ASD group demonstrated significantly more sound effects and attributed significantly more speech to the characters than the TD group; however, this was due to two of the children taking on the persona of the characters and ‘barking’ and ‘ribbiting’ and ‘buzzing’ throughout the task. When these two children were eliminated from the analyses examining sound effects and attribution of speech, these differences were no longer significant. These variables were combined to create a composite variable: number of narrative variables (see Table 3).
Table 8

Narrative task-pragmatic variables

 

ASD group Mean (SD)

TDC group Mean (SD)

Equality of variance

t-test probability

Number of intensifiers

1.00 (1.49)

0.77 (1.30)

F = 0.353 > .05

t(25) = 0.431 p > .05

Emotion or mental state verbs

1.07 (1.39)

1.38 (1.90)

F = 0.728 p > .05

t(25) = −0.493 p > .05

Sound effects

3.43 (5.20)

0.31 (0.48)

F = 15.03 p < .01

t(13) = 2.248∼ p < .05

Speech attributed to characters

3.71 (4.34)

0.62 (1.33)

F = 19.170 p < .001

t(16) = 2.467∼ p <.05

Misinterpretation of story

2.43 (1.91)

1.00 (1.00)

F = 2.829 > .05

t(25) = 2.406 p < .05

Repetition of previous info.

3.57 (5.29)

0.92 (1.55)

F = 8.603 p < .01

t(14) = 1.736∼ p = .09

Unclear reference

6.29 (4.81)

4.23 (2.71)

F = 2.869 p = .061

t(21) = 1.379∼ p = .10

Causal explanations

0.07 (0.27)

1.23 (1.30)

F = 21.744 < .001

t(13) = 3.152∼ p < .01

Discussion of goals

1.64 (1.60)

3.31 (1.97)

F = 0.114 p > 05

t(25) = −2.417 < .05

∼controlled for unequal variance

Group differences emerged most strongly with those pragmatic variables most relevant to telling the story. The ASD group was significantly less likely to give causal explanations for story happenings, and they were significantly less likely to discuss the goals and motivations of the characters. Moreover, the clarity of the reference of the children (i.e., what objects or actions they were discussing), and the number of times they needlessly repeated information yielded differences that approached significance. The ASD group was also significantly more likely to misinterpret the story and thus give incorrect information in their narratives (i.e., thinking that the dog had jumped out the window instead of falling out). These last three variables were also combined into a composite variable: difficulties in interpretation of story (see Table 3).

Paired sample t-tests on vocabulary-matched pairs

As was mentioned earlier, 12 of the ASD children were able to be matched within one point of their raw score on the TACL-Vocabulary subtest with children from the TD group. Paired-sample t-tests were then conducted on these twelve pairs to determine if the children would continue to differ on these tasks when matched on very basic language skills. The ASD group continued to perform more poorly than their vocabulary-matched peers on a number of tasks: the Unexpected Contents task (t(11) = −2.569, < .05), the Mental State Verb task (t(11) = −3.500, < .01), and the number of causal explanations given in the Narrative task (t(11) = −3.130, < .05). In addition, a number of these paired sample tests approached significance, with the ASD group again performing more poorly than their peers: the Elaborated Sentences subtest of the TACL (t(11) = −1.907, = .08), the False Belief target questions (t(11) = −1.915, = .08), and the number of times the story was misinterpreted in the Narrative task (t(11) = −1.980, = .08).

Between-task correlations

The results of the correlations between tasks for the separate groups can be seen in Tables 9 and 10. Both groups demonstrated a general pattern of correlations between tasks within aspects of language, for example, correlations between vocabulary tests and correlations between syntactic tasks. The main difference between the two groups was that the TD group showed correlations between the narrative tasks and other aspects of language competence (as well as within narrative tasks), whereas the ASD group showed within-narrative correlations only. That is, their narrative ability did not seem to be correlated with any other language tasks.
Table 9

Correlations between tasks for ASD group

 

TACL vocabulary standard score

TACL morphology standard score

TACL elaborated sentences standard score

EOWPVT standard score

Wug percent correct

Stanford-Binet sentence memory raw score

TACL morph standard score

.60*

     

TACL elabor. sentences SS

.60*

.64*

    

EOWPVT standard score

.78**

.37

.33

   

Wug percent correct

.22

.42

.20

.55∼

  

S-B sentence Memory RS

.49∼

.50∼

.62*

.45

.45

 

Categorical induction percent correct

.20

.67*

.32

.02

.29

.22

False belief target

.33

.34

.19

−.09

.10

.51∼

Un. contents target

.40

.40

.22

.74**

.49∼

.33

Mental state verb % cor

.51∼

.58*

.38

.25

.24

.44

Complex syntax % cor.

.44

.68*

.66*

.18

.42

.61*

% Frame compliant transitives

−.61∼

−.63∼

−.25

−.68*

−.53

−.004

Percent frame compliant intransitives

.17

.48

−.11

.11

−.16

−.03

Narrative-general lexical var.

−.13

−.14

−.10

−.14

−.02

−.17

Narrative-grammatical errors

−.16

−.27

−.50

−.36

−.49∼

−.45

Narrative-no. of devices

−.15

−.17

.18

−.28

−.002

.35

Narrative-difficulties

−.14

−.05

.24

−.19

−.11

.20

 

Categorical induction percent correct

False belief target question correct

Unexpected contents percent of target questions

Mental state verbs percent correct

Complex syntax percent correct

Percent frame compliant on ungram. transitives

False belief target ques.

.37

     

Unexpected contents

.33

−.03

    

Mental verbs % correct

.61*

.43

.23

   

Complex syntax

.39

.41

.23

.24

  

Ungrammatic. transitives

−.55∼

−.03

−.33

−.52

−.32

 

Ungrammatic. intransitives

.10

.07

.07

−.25

.32

−.45

Narrative lexical

−.12

.13

−.15

−.04

.21

.23

Narrative-grammatical

−.17

.03

−.23

−.11

−.09

.29

Narrative-no. of devices

−.13

.27

−.38

.07

.09

.28

Narrative-difficulties

−.17

−.20

−.19

−.17

−.04

.32

 

Frame com. intransitives

Narrative lexical

Narrative gram. errors

Narrative devices

  

Narrative lexical

−.06

     

Narrative grammatical errors

.16

.73**

    

Narrative devices

−.39

−.38

−.46∼

   

Narrative difficulties

−.27

−.67**

−.57*

.74**

  

∼ < .10; * < .05; ** < .01

Table 10

Correlations between tasks for TD group

 

TACL vocabulary standard score

TACL morphology standard score

TACL elaborated sentences standard score

EOWPVT standard score

Wug percent correct

Stanford-Binet sentence memory raw score

TACL morph standard score

.60*

     

TACL elabor. sentences SS

.05

.50∼

    

EOWPVT standard score

−.68**

−.24

.09

   

Wug percent correct

−.39

.01

.19

.54*

  

S-B sentence memory RS

−.53*

.10

.29

.52∼

.56*

 

Categorical induction percent correct

−.05

−.24

.13

−.08

.21

.07

False belief target un. contents target

−.13

.34

.13

.33

.43

.29

mental state verb % cor

−.11

.22

.06

.13

.34

.58*

Complex syntax % cor.

−.46∼

.09

.58

.31

.69**

.67**

% Frame compliant transitives

−.22

−.05

.06

.11

.20

.01

Percent frame compliant intransitives

.20

.17

−.21

−.13

.05

.02

Narrative-general lexical var.

−.28

.26

.13

.40

.77**

.34

narrative-grammatical errors

−.48∼

−.03

−.05

.53∼

.40

.50∼

Narrative-no. of devices

−.30

.24

.10

.52∼

.50∼

.18

Narrative-difficulties

−.05

−.23

−.57*

−.21

−.05

−.42

 

Categorical induction percent correct

False belief target question correct

Unexpected contents percent of target questions

Mental state verbs percent correct

Complex syntax percent correct

Percent frame compliant on ungram. transitives

False belief target ques. unexpected contents

−.47∼

     

Mental verbs % correct

.61*

 

.70**

   

Complex syntax

.39

 

.30

.24

  

Ungrammatic. transitives

−.55∼

 

.000

−.14

.12

 

Ungrammatic. intransitives

.10

 

.000

.50

−.38

.35

Narrative lexical

−.12

 

.44

.19

.46

−.04

Narrative-grammatical

−.17

 

.006

.07

.15

.14

Narrative-no. of devices

−.13

 

.24

−.05

.20

.24

Narrative-difficulties

−.17

 

.09

−.11

−.49∼

−.20

 

Frame com. intransitives

Narrative lexical

Narrative gram. errors

Narrative devices

  

Narrative lexical

−.003

     

Narrative grammatical errors

−.06

.57*

    

Narrative devices

.34

.62*

.26

   

Narrative difficulties

−.36

−.15

.05

−.24

  

∼ < .10; * < .05; ** < .01

Vineland adaptive behavior scales

These scales were not given to the typically developing children. The children with a history of autism were completely within the normal to high range on the Communication subscale (Mean standard score = 98.92, SD = 13.18, range = 81–132) and all but two of the children scored within the low normal to normal range on the Socialization subscale (Mean standard score = 80.23, SD = 10.62, range = 63–101).

Correlations between the language tasks and the Vineland

Pairwise correlations were performed between the ASD children’s standard scores on the Vineland Communication and Socialization subscales and their scores on each language task; only two reached significance: The Sentence Memory task (r = .71, p < .01) and the number of narrative devices used in the Narrative tasks (r = .57, p < .05) both correlated significantly with the Vineland Communication standard score. However, as mentioned earlier, two children in the ASD group took on the point of view of the characters and ‘barked’ and ‘ribbited’ throughout; when these two children were eliminated from the analysis, the latter correlation was no longer significant. When the raw scores of the Vineland Communication and Socialization subscales were used (controlling for age), the only significant correlation was between the Stanford-Binet Memory for Sentences raw score and the Vineland Communication subscale (= .79, < .05).

Correlations between the language tasks and estimated hours of treatment

Pair-wise correlations were performed between the children’s scores on each language task discussed above and the number of hours of treatment they experienced. No significant correlations were obtained between language scores and the number of hours in treatment.

Correlations between the language tasks and number of early autistic symptoms

Correlations were also run between number of early autistic symptoms from the DSM-IV and the language tasks. The only significant correlations were between the False Belief task and the number of Social autistic symptoms (= −.60, < .05) and the number of Communicative symptoms (= −.56, < .05). The negative correlations indicate that the more autistic symptoms they had originally, the more poorly they did on the False Belief task.

Discussion

This study investigated the language abilities of grade-school-aged children who had a history of ASD but now functioned in the normal range on standardized measures of language and communication skills, and functioned without support in mainstream classes, and compared their abilities to their typically developing peers. The primary question was whether a comprehensive assessment of the grammatical, semantic, and pragmatic aspects of their language would reveal residual deficits in the ASD group. The results were threefold.

First, a solid area of strength for the ASD group was found within the standardized tests. These children did not differ from the TD group on the vocabulary assessments, and although they performed less well than the comparison group on the other subtests, they tested well within the normal range for their age. The significant differences, therefore, were probably due to the above-average performance of the TD group on these tests.

Second, the psycholinguistic tasks revealed both strengths and weaknesses in the ASD group’s language. In particular, the ASD group’s performance was indistinguishable from that of the TD group on the Productive Morphology and Complex Syntax tasks. However, the ASD group did perform somewhat more poorly than the TD group on the Verb Argument Structure task, and on the target questions, particularly when they involved animate stimuli, of the Categorical Induction task. The ASD group performed significantly more poorly than the TD group on the Mental Verb task, the Theory of Mind task, and the pragmatics aspects of the Narrative task.

Third, only one significant correlation emerged between the ASD children’s scores on the Vineland and their performance on the language tasks; only their scores on the Sentence Memory measure correlated significantly with their scores on the Vineland Communication subscale. Children who had initially presented with more severe autistic features now performed more poorly on the False Belief task.

What is the pattern of language abilities and deficits in these children with a history of ASD?

Despite the fact that the children with a history of autism told stories that were just as long and grammatically indistinguishable from those of their peers, they were still experiencing difficulties in the pragmatic realm. For example, when retelling the Frog Story, our ASD group provided incorrect and redundant information, few causal explanations, and few mentions of the goals and motivations of the characters. Our ASD group also was more likely to interpret the Theory of Mind questions solely in terms of their own knowledge state rather than also taking into account the knowledge states of others. In both of these findings, we corroborate those already reported in the literature for ASD children (Fein, 2001; Geller, 1998; Happe, 1995; Lord & Paul, 1997; Tager-Flusberg, 1996, 1999): Thus, children with autism have consistent difficulty with the pragmatics component of language, and our children with a history of autism still manifested that difficulty.

The tasks more related to the grammatical sub-domain elicited generally better performance. These children demonstrated their ability to generalize morphological rules to novel items in the Productive Morphology task as well as their sensitivity to the syntactic ramifications of the presence of a medial wh-word in a multi-clause wh-question in the Complex Syntax task. The ASD group’s good performance on this latter task contrasts with their poor performance on the Mental Verb task, which also involves multi-clause sentences. It seems more likely, though, that the Mental Verb task is tapping the children’s pragmatic and/or semantic knowledge rather than their syntax, as the contrasting stimuli in this task had the same syntactic form but different semantic and pragmatic implications (i.e., that guess and think are less certain than know). Thus, it is more parsimonious to attribute the ASD group’s poor performance on the Mental Verbs task to their already attested pragmatic (or semantic, see below) difficulties, rather than any syntactic difficulties.

The ASD group’s performance with the Verb Argument Structure task also deserves mention. As described earlier, these children were significantly more Frame Compliant with the ungrammatical transitive sentences than the TD; their scores resembled those of typically developing preschoolers (Naigles et al., 1993). It is likely, though, that this level of performance actually reflects lexical rather than syntactic immaturity. As discussed by Naigles et al. (1992, 1993, 1995), this verb argument structure task taps syntactic knowledge insofar as it assesses whether participants understand the relation between sentence frames and sentence meanings; for example, that transitive frames canonically imply causative meanings (e.g., Jackendoff, 1990; Levin, 1993). Our finding that the ASD systematically interpreted the ungrammatical transitive frames causatively indicates that they are in possession of this grammatical knowledge. However, the task also taps lexical knowledge insofar as it assesses whether participants have mastered the verb-specific frame requirements of English; for example, that bring must be transitive and go must be intransitive. This, we conjecture, is the aspect of knowledge that our ASD group were having difficulty with. Previous findings have suggested that this aspect of verb argument structure acquisition is accomplished on a verb-by-verb basis in typically developing children (Naigles et al., 1992); it seems likely that the ASD group were further behind in this lexical process.

Finally, the ASD group also performed more poorly on the other task that tapped lexical semantics, namely, Categorical Induction. That is, these children had considerable difficulty in realizing that a given property of a named natural kind should “transfer” to another named instance of that natural kind (i.e., that if told a white rabbit eats grass, should induce that a brown rabbit eats grass, too). Thus, our findings are more similar to those of Dunn et al. (1996) and Hermelin and O’Connor (1970) than to those of Tager-Flusberg (1985). On the surface, these children’s poor performance with the lexical semantics tasks is at odds with their excellent performance on the standardized vocabulary assessments. However, the children’s performance on the standardized vocabulary tests indicated only that they knew many words, thus demonstrating understanding of the identification function of words and concepts. What they may have had difficulty with are other conceptual functions, such as that which motivates the relations between concepts and that which allows for the induction of hidden properties (e.g., Gelman & Medin, 1993).4

Taken together, those tasks that most closely tap children’s pragmatic language (Theory of Mind, Narrative, Mental Verbs) elicited the poorest performance by our ASD group, whereas the grammatical tasks (Productive Morphology, Complex Syntax) yielded the most consistently high performance of any domain. Tasks that required children to use their lexical knowledge showed mixed findings with good overall vocabulary but impairments in specific aspects of lexical understanding. Thus, the psycholinguistic tasks do appear to cluster together to reveal a pattern of linguistic strengths and weaknesses in these children with a history of autism. The pattern replicates to a large extent that put forth by Tager-Flusberg (2001), who proposed that children with ASD would perform poorly within the pragmatics domain but quite well within the grammatical domain (but see Kjelgaard & Tager-Flusberg, 2001, for a discussion of a subgroup of children with autism with grammatical difficulties). However, whereas Tager-Flusberg (2001) considered the semantics abilities of individuals with autism to be fairly intact, we found that our ASD group showed mixed performance in the lexical semantics domain (see also Dunn & Rapin, 1997).

This conclusion is supported by the findings from the correlations performed between all of the tasks (Tables 9 and 10). Several patterns within the two correlation matrices are similar. For example, both groups yielded comparable numbers of correlations within the lexical domain (e.g., between TACL vocabulary and the EOWPVT, between Categorical Induction and Mental Verbs), within the grammatical domain (e.g., between the Stanford-Binet sentence memory test and the test of Complex Syntax), and across the lexical and grammatical domains (e.g., between TACL morphology and Categorical Induction, between TACL vocabulary and the Stanford Binet sentence memory test). Moreover, both groups yielded few correlations between the language measures and the Theory of Mind-related measures (i.e., one each with the Unexpected Contents test, two each with Mental Verbs). Where the groups differed most involved the narrative measures: Whereas these were inter-correlated for each group, only the TD children’s scores also yielded significant correlations between the narrative measures and other aspects of language5 Thus, it appears that for the typically developing children, pragmatic ability, as manifested by their narratives, is well-integrated with other areas of language. In contrast, the pragmatic abilities of the ASD appear to stand alone.

What predicts ASD children’s performance on the language tasks?

There is little to suggest that the ASD children’s adaptive skills were associated with their performance on the language tasks. Only one of the tasks yielded a significant correlation (once outliers were removed), the Sentence Memory task. The range of Vineland Communication standard scores (81–132) seems wide enough to support more correlations; thus, it is possible that the language tasks and the Vineland are simply tapping different abilities and deficits in the children. Although Vineland Communication measures adaptive use of communication, at the age and competence level of the children tested, it largely measures specific, taught skills, such as grade level for reading, using a dictionary, writing notes, and addressing envelopes and would not be expected to be sensitive to the kinds of deficits found here.

None of the correlations between the language tasks and the estimated amount of treatment were significant. Moreover, the children’s early diagnostic symptoms only predicted their performance on the False Belief task. It is possible that the paucity of correlations is attributable to our small sample size. In order to examine the true effects of early severity and duration and type of treatment on cognitive outcomes, one would need a larger sample with greater variation in treatment, and early measures of cognition. It is also possible that children with more severe early symptoms received earlier and longer intervention, thus suppressing any correlations.

Conclusions

Several limitations to the current study should be mentioned. The sample size was relatively small and homogeneous; although it might be difficult to identify a large sample of equally successful children, such an effort would allow firmer conclusions about this group of children. The sample size was too small to allow examination of results by prior diagnosis, such as Autistic disorder vs. PDD-NOS, although examination of the Vineland scores suggests no differential communication outcome by early diagnosis. As mentioned above, treatment type and SES were also of limited variability. Second, the children received clinical but not research diagnoses (e.g., ADOS or ADI), although the clinical diagnosis was given by a diagnostician who has specialized in this area for many years (see Table 3 for the children’s diagnostic markers). It should also be noted that the children’s diagnoses were given long before the study was even conceptualized, thus making any bias unlikely. Third, for the current study, the children were assessed for language and adaptive functioning; it was clear that most of them would no longer meet criteria for any form of PDD but formal diagnostic procedures were not administered at the time of testing. Fourth, no measure was given to examine the children’s level of psychopathology. It is possible that these children are now manifesting a different disorder. Finally, no nonverbal tests were given to the children. Although our primary interest was the children’s language status, a measure of nonverbal IQ might have been illuminating. A follow-up of these children, including formal diagnostic procedures, a parent-reported measure of psychopathology, and nonverbal IQ measures, is under way.

As mentioned in the introduction, this study is the first to investigate in detail the language abilities and residual weaknesses of children who were previously diagnosed with ASD and who have shown such great improvement that they are mainstreamed into age-appropriate classrooms in elementary school with minimal or no supports. On the one hand, the dramatic improvement of these children can be seen in the mere fact that they were able to participate fully in all tasks, split only over two hour-long sessions, and that their Vineland Communication and Socialization scores were all within the normal range. Moreover, their receptive and expressive vocabulary scores, which were not significantly different from those of the TD, attest to their strong ability to learn the mappings between words and their referents. Scores on the grammatical tests were either indistinguishable from their peers (i.e., Productive Morphology, Complex Syntax) or well within the normal range for their chronological age (Grammatical Morphemes, Elaborated Sentences, Verbal Memory). Taken together, these results paint a picture of children whose knowledge of grammar and vocabulary is appropriate for their age level.

However, their performance was not consistently at this level. Significant residual deficits in some areas of language still remain in these children, particularly with regard to lexical semantics and pragmatics. These results have clinical implications for children with a history of ASD. The language difficulties described above suggest that the children might benefit from periodic in-depth language assessments and continued language therapy where indicated. Standardized scores in the average range and adequate performance academically may have led some school systems to discontinue support services prematurely. The difficulty with complex lexical and pragmatic tasks suggests that social cognition may be an area of continuing difficulty. It is possible that as the children face the more complex social and linguistic demands of the higher grades, their academic and social adjustment may be at risk.

Footnotes
1

Prosody, and phonology in general, are also areas of difficulty for children with autism (Shriberg, Paul, McSweeney, Klin, & Cohen, 2001; Wolk & Giesen, 2000). Because our study did not address these areas, though, space constraints do not permit us to review them.

 
2

This occurred with only one of the ASD group. In addition, there were technical difficulties with the videotape for one of the TD children; thus his enactments are not included.

 
3

For the intransitive verbs in transitive frames, Frame Compliant enactments were those where the child treated the first animal spoken as a causal agent of the second animal’s motion (e.g., “*the zebra comes the horse”, enacted as the child using the zebra to push, force, or carry the horse). Verb Compliant enactments were those where the children had the horse and zebra move separately. For the transitive verbs in intransitive frames, Frame Compliant enactments were those where the children had the animal performing some sort of non-causative action (e.g., “*The elephant brings,” enacted as simply moving the elephant around in some manner without using another object). If children added a direct object, and had the elephant move another animal from one location to another, the enactment was then coded as VerbCompliant. Enactments were coded as Other when they did not fit into one of the two preceding categories.

 
4

Recall that, while the ASD group’s scores on the Categorical Induction task were consistently lower than those of the TD, this difference only reached significance for those items that depicted animate natural kinds (see Table 4). Greater capability with inanimate than animate stimuli has been reported before in connection with ASD: Animate objects may be over-arousing; hence, less attention is paid to them (Frith & Happe, 1998) or they are less inherently reinforcing to children (Hauck, Fein, Maltby, Waterhouse, & Feinstein, 1998).

 
5

When the correlations were conducted using the raw scores of the standardized tests, the TD group’s scores yielded three additional significant correlations between narrative variables and the lexical/grammatical measures; the ASD yielded none.

 

Acknowledgments

We would like to thank the University of Connecticut Small Grant program for supporting this research. We greatly appreciate all the parents and children for their cheerful participation. Thank you to Anna Janovicz, who was extremely helpful in the collection and analysis of the data, Lara Mayeux, Donna Vear, and Tony Sharillo for help in the data collection process, and Jennifer Mangini for her help with reliability coding. We would also like to thank Jill deVilliers and Susan Gelman for their gracious sharing of stimuli for the Complex Syntax and Categorical Induction tasks, respectively.

Copyright information

© Springer Science+Business Media, Inc. 2006