Referential communication was examined in youth with Down syndrome or fragile X syndrome in comparison to each other and to MA-matched typically developing children. A non-face-to-face task was used in which the participant repeatedly described novel shapes to listeners. Several dimensions of referential communication were especially challenging for the syndrome groups (i.e., they displayed below-MA performance), although there were differences in the dimensions that each syndrome group found to be most challenging. Independently assessed expressive language ability contributed to variations in referential performance, especially for participants with Down syndrome.
For a discourse to be successful, each speaker must describe his or her intended referents in a way that makes them readily identifiable (Clark, 1996). Failure to do so makes it impossible for a listener to construct an accurate representation of the intended meaning, which can lead to a breakdown of the interaction (Graesser, Mills, & Zwaan, 1997). Individuals with mental retardation are less likely than are typically developing individuals of the same mental ages (MAs) to produce utterances with identifiable referents (Brownell & Whiteley, 1992; Longhurst, 1974; Rueda & Chan, 1980). The result is discourse that is less comprehensible to others than that of same-age (Kernan & Sabsay, 1987) or even MA-matched peers (Hemphill, Picardi, & Tager-Flusberg, 1991). Nevertheless, there is considerable within-group variability in referential talk among those with mental retardation (Abbeduto & Hesketh, 1997). In this study, we focused on the possibility that referential problems vary with etiology by making a comparison between the two most common genetic causes of mental retardation: Down syndrome and fragile X syndrome. These data are necessary for developing language interventions that meet the unique needs of the individual with mental retardation.
Referential talk draws on linguistic, cognitive, and social–cognitive skills that are impaired in those with Down or fragile X syndrome (Abbeduto & Short-Meyerson, 2002). Most individuals with Down syndrome have impairments in numerous cognitive skills that result in IQs in the range of mild to moderate mental retardation (Chapman & Hesketh, 2000). The range of cognitive impairments is broader in those with fragile X syndrome, with virtually all males and half of all females with the full mutation having a diagnosis of mental retardation (Hagerman, 1999). Below age-level mastery of the linguistic system is almost invariably associated with Down syndrome (Chapman & Hesketh, 2000) and is characteristic of many males and females with fragile X syndrome (Murphy & Abbeduto, 2003). Performance in the social–cognitive domain (e.g., in reasoning about mental states) is also delayed relative to typical same-age peers, both in Down syndrome (Zelazo, Burack, Benedetto, & Frye, 1996) and fragile X syndrome (Cornish et al., 2005; Garner, Callias, & Turk, 1999). These domains have all been found to underlie the referential talk of typical speakers (Glucksberg, Krauss, & Higgins, 1975; Ninio & Snow, 1996; Ricard, 1993), and, thus, the impairments that characterize Down and fragile X syndromes in these domains suggest that these populations will find the referential process challenging.
Despite the commonalities between Down syndrome and fragile X syndrome, there are differences between the two that could lead to different referential problems. First, the development of syntax is more delayed in Down syndrome than in fragile X syndrome (Abbeduto et al., 2003). Second, individuals with Down syndrome display deficits in theory of mind (i.e., reasoning about mental states) that are more severe than their deficits in other areas of cognitive functioning (Zelazo et al., 1996). In contrast, persons with fragile X syndrome, as a group, are as accurate in reasoning about mental states as are their cognitive-level matched typically developing peers (Cornish et al., 2005; Garner et al., 1999). Third, auditory memory seems to be more impaired than visual memory and other aspects of cognition in those with Down syndrome (Marcell & Weeks, 1988; Seung & Chapman, 2000), whereas no such asynchrony has been described for individuals with fragile X syndrome (Dykens, Hodapp, & Finucane, 2000). Fourth, maladaptive behaviors occur at lower rates in Down syndrome than in fragile X syndrome (Dykens et al., 2000). Most notable among the maladaptive behaviors of fragile X syndrome are social anxiety (Bregman, Leckman & Ort, 1988; Mazzocco, Baumgardner, Freund, & Reiss, 1998) and attentional difficulties (e.g., Bregman et al., 1988; Cornish, Sudhalter, & Turk, 2004; Dykens, Hodapp, & Leckman, 1989; Mazzocco, Pennington, & Hagerman, 1993). These syndrome differences could lead individuals with Down syndrome and those with fragile X syndrome to find different dimensions of the referential process problematic.
Little evidence is available, however, about the extent and nature of the referential problems in Down or fragile X syndrome. Studies of early intentional communication by children with Down syndrome suggest that they have difficulty using nonverbal behaviors, vocalizations, and single words to draw their partner's attention to an object of interest (Greenwald & Leonard, 1976; Smith & von Tetzchner, 1986), which could be seen as a precursor to intentional referential talk (Abbeduto & Hesketh, 1997). It is unclear, however, whether these early referential problems continue in individuals who are older and more linguistically advanced.
Even less is known about referential communication in fragile X syndrome. Several studies have documented a higher rate of verbal perseveration (i.e., self-repetition) in males with fragile X syndrome than in individuals with Down syndrome or autism (Belser & Sudhalter, 2001; Ferrier, Bashir, Meryash, Johnston, & Wolff, 1991; Sudhalter, Cohen, Silverman, & Wolf-Schein, 1990). The perseverative behaviors studied have included inappropriate repetition of noun phrases, which could reflect a disruption of the process of introducing referents into discourse. Unfortunately, studies to date have failed to distinguish repetitions of noun phrases from other forms of self-repetition that may reflect problems having little to do with referential talk (Pavetto, 2001). Moreover, there have been no studies of referential communication in females with fragile X syndrome (Murphy & Abbeduto, 2003), although there is evidence that even affected females with normal-range IQs have problems with social discourse (Simon, Keenan, Pennington, Taylor, & Hagerman, 2001).
In the present study we had two goals. The first goal was to determine the extent to which individuals with Down syndrome or fragile X syndrome effectively manage the various dimensions of the referential process and whether there are syndrome differences in this regard. In addressing this goal, we employed a framework in which referential talk is seen as an act of collaboration with the listener (Clark, 1996). Thus, we compared the two syndromes to typically developing MA-matched children and to each other in terms of their use of (a) effective forms of referential description, (b) collaborative behaviors designed to ensure that their referents can be identified by the listener, and (c) messages tailored to the knowledge that the listener brings to, and accrues during, the discourse. Our second goal was to determine how referential talk is shaped by a speaker's levels of language, theory of mind, auditory memory, and maladaptive behavior and whether there are differences between the syndromes and MA-matched typically developing children in this regard. In addressing these goals, we focused on dimensions of referential performance that provide the foundation for successful discourse because impairments on these dimensions could have serious consequences for everyday social interaction. Both males and females participated, including in the fragile X syndrome group, because no qualitative differences in behavioral profiles between the sexes have been documented to date (Hagerman, 1999; Murphy & Abbeduto, 2003).
Three groups participated: adolescents and young adults with Down syndrome (n = 25), adolescents and young adults with fragile X syndrome (n = 18), and typically developing 3- to 6-year-olds (n = 25). Participants with Down or fragile X syndrome were recruited locally through newspaper advertisements, mailings to special educators, and notices sent to families in a university research registry. Families in the syndrome groups were also recruited nationally through postings on the Internet and in newsletters of nation-wide developmental disabilities organizations. Typically developing children were recruited locally and were not receiving special education services. The sample overlaps with that of Abbeduto et al. (2003).
Characteristics of the sample are provided in Table 1. The participants were selected such that the three groups were matched groupwise on nonverbal MA, F(2, 65) = .02, p = .99, with nonverbal MA determined by administering three subtests (described subsequently) from the Stanford-Binet, 4th edition (Thorndike, Hagen & Sattler, 1986). The Down syndrome and fragile X syndrome groups did not differ in nonverbal IQ according to the Stanford-Binet subtests, t(27,95) = .61, p = .55, or chronological age (CA), t(41) = .59, p = .56.
Across the Down and fragile X syndrome groups, 60% of participants had mothers (or female guardians) with a college degree or higher, with no difference across syndromes in this regard, χ2(2, N = 68) = .15, p = .70. All but 2 of the mothers of the typically developing children had a college degree; however, no significant differences were found as a function of maternal education for the syndrome groups on any dependent measure and, thus, this variable was not considered further.
Ninety-one percent of the participants were White, and race (White or not White) did not differ across groups, χ2(2, N = 68) = 4.14, p = .13. The distribution of males and females differed across groups, χ2(2, N = 68) = 12.16, p = .002, with the most males in the group with fragile X syndrome and the fewest in the typically developing group.
No participant had more than a mild hearing loss (defined as a mean pure tone threshold across the frequencies of 500, 1000, and 2000 Hz of 30 dB or worse in the better ear). No participant had a diagnosis of autism (see Abbeduto et al., 2003, for details of diagnostic procedures).
The parents/guardians of all participants with Down syndrome reported etiology as trisomy 21. Medical records confirming the karyotype were available for 20 participants with Down syndrome. Reports of DNA confirmation of the fragile X full mutation were available for all but one participant, who had only cytogenetic confirmation. Four males with fragile X syndrome were mosaic. Three families had two children with fragile X syndrome participate.
Putative Predictors of Referential Talk
The following measures were administered as part of a more comprehensive protocol. Each participant completed these measures on the same day as the referential task.
Nonverbal IQ and MA were determined from the Bead Memory, Pattern Analysis, and Copying subtests of the Stanford-Binet, 4th edition (Thorndike et al., 1986). These subtests require few verbal instructions and the participant responds nonverbally. Nonverbal MA was used as a covariate in the group comparisons of referential skill and to evaluate the contributions of cognitive ability to referential performance (see Table 1).
Auditory short-term memory
We administered the Digit Span subtest from the Wechsler Scales of Intelligence for Children, 3rd edition—WISC-III (Wechsler, 1991). In this subtest, the examiner says digit sequences at the rate of one digit per second. The participant must immediately repeat the sequence verbatim. The task begins with two-digit sequences and the sequences increase in length over trials. The number of correctly repeated sequences was used to evaluate the contribution of auditory short-term memory to referential performance (see Table 1).
Measures of receptive and expressive language were obtained. Receptive language was assessed by the Test for Auditory Comprehension of Language (Carrow-Woolfolk, 1985). In this test, the participant points to the one picture of three that matches the meaning of a spoken stimulus on each trial. This test measures understanding of vocabulary and syntax. The total test age-equivalent was used to evaluate the contribution of receptive language to referential performance (see Table 1).
Expressive language was assessed by the Oral Expression Scale of the Oral and Written Language Scales (Carrow-Woolfolk, 1995). This test measures numerous dimensions of expressive language, from single-word use to sentence formulation. For each item, the experimenter reads aloud a stimulus that requires the participant to answer a question, complete a sentence, or generate a new sentence. Drawings provide contextual support. The Oral Expression Scale age-equivalent was used to evaluate the contribution of expressive language to referential performance (see Table 1).
Theory of mind
We administered a false belief task that was based on tasks commonly used to study theory of mind (Yirmiya, Erel, Shaked, & Solomonica-Levi, 1998). Success on such tasks is thought to require the knowledge that the human mind does not simply copy the world, but instead represents and interprets it (Tager-Flusberg, 2001). This knowledge typically emerges in the preschool years with the ability to reason about the beliefs of another person (i.e., first-order beliefs). More complex manifestations of theory of mind, such as reasoning about another person's beliefs about yet another person's beliefs (i.e., second-order beliefs), emerge later, on average.
The examiner told a brief story while enacting it with miniature props. The story involved an object changing location. Some characters held false beliefs about the object's location (first-order beliefs) or about what other characters believed about the object's location (second-order beliefs). The participant always knew the location of the object and which characters had witnessed the object's change of location. Four test questions assessed recognition of the characters' false beliefs. The proportion of test questions answered correctly was used to evaluate the contribution of theory of mind to referential performance (see Table 1).
We asked one parent of each participant in the syndrome groups to complete the Child Behavior Checklist/4–18 (Achenbach, 1991). Higher scores on this checklist reflect more severe problems. We computed a Total Problems T score (see Table 1) and T scores for the five subscales expected to distinguish the groups: Withdrawn, Anxious/Depressed, Attention Problems, Thought Problems, and Social Problems. These scores were used to evaluate the contribution of maladaptive behavior to referential performance.
Each participant completed a non-face-to-face referential task that was portrayed as a “talking and matching game.” The participant, who played the role of speaker, and a researcher, who played the role of listener, were separated by an opaque partition and communicated only through spoken language. The task was designed so that we could examine the form of the descriptions that the participant produced, his or her use of collaborative behaviors, and his or her sensitivity to differences in shared knowledge.
The participant and listener were given an identical set of four cards. Each card contained a drawing of a different novel shape. The shapes could not be adequately described by simple, universally accepted labels (e.g., “square”). Instead, each shape was designed such that it had to be described by reference to its parts (e.g., “a square with a point”) or by reference to a real object that it resembled (e.g., “a house”). The participant's task was to describe each shape so that the listener could select it from among his or her four shapes. The use of difficult-to-label shapes allowed us to determine whether participants designed their descriptions with their listener, the need for collaboration, and attention to shared knowledge in mind. This would not have been possible if we had used familiar entities that “automatically” elicited the same label from most people.
The task was divided into two blocks of six trials. Each shape was described on three trials in each block. In the first block, the game was played with Listener 1 (Listener 1 trials). In the second block, the game was played with a different listener (Listener 2 trials). The decision to present each shape on multiple trials was intended to parallel the fact that most naturally occurring conversations contain references to different entities as well as repeated references to each entity that is the topic of conversation. This made it possible to evaluate the ways in which the participant took into account the accumulation of shared knowledge over the course of the interaction.
Each participant completed the task twice, each time under a different condition. In the naïve listener condition, Listener 2 was unaware of the descriptions that the participant had used with Listener 1. In the knowledgeable listener condition, Listener 2 was knowledgeable about those descriptions, and the participant was aware of this fact. Different shapes were used for each condition. This manipulation was designed to mimic the fact that in everyday conversation, we sometimes speak with people with whom we share little history and other times, with people with whom we have considerable shared history. Such differences in the shared knowledge brought to the conversation can have consequences for how speakers design their descriptions, and we were interested in evaluating this aspect of description design.
Eleven shapes were created. Each was a line drawing created by combining two or more geometric shapes (e.g., circle and triangle) using draw-and-paint computer software. Each shape was approximately 6.5 cm × 8.3 cm and was drawn in black ink onto a 12.7 cm × 12.7 cm white card. In a pilot study, typically developing children of the same ages as those included in the study described each shape to another person. Each shape generated several different descriptions, and no description occurred with noticeably greater frequency than any other description.
Three shapes were used for practice trials. The remaining shapes were randomly divided into two quartets. For each participant, one quartet was randomly assigned to the knowledgeable listener condition and the other, to the naïve listener condition.
For each trial, two shapes were designated as the referents. This designation was random, except that (a) each shape occurred as the referent three times in each block of six trials, (b) no shape occurred as a referent on more than two consecutive trials, and (c) no pair of shapes occurred on consecutive trials. The 81.3 cm × 104.1 cm cardboard partition separating the participant and listener ensured that the participant could see neither the listener nor the listener's shapes.
Each participant was tested in both listener knowledge conditions, with each condition conducted in a different session. Depending on participant availability, the two sessions were conducted on different days or on the same day with a break between. Order of presentation of the listener knowledge conditions was random, with the frequencies of the two orders not being significantly different across diagnostic groups, χ2 (2, N = 68) = 3.67, p = .16.
Each session began with the participant and Listener 1 seated across the table from each other. The experimenter gave both the participant and Listener 1 the practice shapes, explained the game, and placed the partition between the participant and listener. Two practice trials were then conducted. On each practice trial, the experimenter designated two of the cards as referents by moving them closer to the participant and prompting a referential description (“Tell [Listener 1] about these two drawings so that she or he can pick the same ones”). On the first practice trial, Listener 1 responded to each of the speaker's descriptions of the shapes by signaling understanding (e.g., “Ok, I think I got it”). On the second practice trial, Listener 1 feigned noncomprehension of the first description and prompted another description (e.g., “I'm not sure about that one. Tell me a little bit more about it”). This was done so that the participant recognized that his or her descriptions were not always immediately understandable and, thus, consideration of the listener's needs was necessary. The experimenter provided feedback as needed to ensure that the participant understood the task.
The practice trials were followed by the experimental trials, each of which began with the experimenter designating two of the four shapes as the referents and instructing the participant to talk about each shape in turn. After each of the speaker's descriptions, the listener waited a few seconds before making a selection and signaling understanding. The experimenter provided minimal prompts as needed to solicit a description from the participant but offered no feedback concerning the adequacy of the descriptions or the accuracy of the listener's selections.
During the six Listener 1 trials, the participant provided descriptions to Listener 1. In the naive listener condition, Listener 2 waited outside the testing room throughout the Listener 1 trials. At the end of the Listener 1 trials, Listener 1 left and Listener 2, the “naive” listener, was introduced. Listener 2 then assumed Listener 1's position as the listener for the six Listener 2 trials. In the knowledgeable listener condition, Listener 2 was introduced at the start of the Listener 1 trials as someone who would watch and listen, but not play until later. Listener 2 sat next to the participant while the participant and Listener 1 played the game. At the end of the Listener 1 trials, Listener 1 left and Listener 2 assumed Listener's 1 position and role for the six Listener 2 trials.
Verbatim transcripts of the talk were prepared from audiotapes of each session using SALT (Systematic Analysis of Language Transcripts), a computer program for analyzing language transcripts (Miller & Chapman, 2000). For purposes of transcription, a trial began with the listener's “OK, I'm ready” and ended with his or her signal of understanding after the participant's description of the second shape for the trial. Each transcript was checked for accuracy against a videotape of the session by a second transcriber, who inserted relevant nonverbal behaviors and suggested revisions to the transcribed talk. All of the suggestions from the second transcriber were then evaluated and incorporated as deemed appropriate by the first transcriber. If necessary, the primary and secondary transcribers discussed points of disagreement before a final decision was reached.
The participant's description of each shape was coded from the final (i.e., checked) transcript. Each coded transcript was checked by another (secondary) coder, with disagreements decided by the primary coder. The dimensions coded and the dependent measures derived from them are described subsequently. High levels of intercoder agreement were achieved by the end of training (i.e., 80% to 90% or better for all dimensions across 3 randomly selected participants, 1 from each group). Following training, we also determined percentage of agreement (and Cohen's kappa) between two coders who independently coded the same 6 transcripts, each from a different randomly selected participant (2 per group) and involving a total of 288 descriptions (see below).
Ninety-eight percent of the participants' descriptions had a referential intent. The remaining utterances were off task or were repetitions of all or part of the experimenter's instruction or prompt.
Referential descriptions were classified as naming or componential. Naming descriptions were those that indicated that the shape was, or resembled, a specific real-world object (e.g., “a house,” “it looks like an ice cream cone”). Componential descriptions were those that included only the features or components of the shape (e.g., “first it has a line that goes up a lot and then down”). Ricard (1993) has shown that a componential strategy is seldom used by school-age and older typically developing individuals and is generally a less effective strategy for ensuring that the listener can identify the intended referent. The proportion of all referential descriptions that were naming descriptions was calculated (interrater agreement = 99%, κ = .91).
In order to determine whether each participant met the basic requirements of effective collaboration in referential discourse, we further analyzed the naming descriptions in order to determine the extent to which the participants (a) used unique descriptions, (b) expressed their descriptions through referential frames, and (c) were consistent in their descriptions.
Unique descriptions were those that the participant used to refer to only one of the four shapes during a session. Unique descriptions contrast with cases in which a naming description is extended to two or more shapes, rendering it ambiguous from the perspective of the listener. For example, “the muffin” is informative if extended to a single shape, but not if it is used to refer to two or more shapes. The use of unique descriptions reflects the participant's recognition of the basic informational needs of a listener. We calculated the proportion of naming descriptions that were unique (summed across conditions). Interrater agreement was 93%, κ = .88.
Because the novel shapes had no conventional names, any naming description was only a clue to the speaker's referential intent. This observation led us to determine whether each naming description was expressed through what we termed a referential frame. Referential frames were utterances in which the participant indicated that there was a resemblance between the shape and a real-world object while cautioning the listener to take the description as a clue rather than as a literal statement; for example, “it is kind of like an ice cream cone” rather than “it is an ice cream cone.” Thus, referential frames are a means of scaffolding the listener's comprehension. We calculated the proportion of naming descriptions expressed through a referential frame (summed across experimental conditions), and interrater agreement was 99%, κ = .98.
A consistent naming description was one that was essentially the same as a previous description used for the same shape. A description was considered the same if it maintained the “core” of a previous description; for example, “bread” but not “bun” would maintain the core of “a piece of bread.” This measure reflected the extent to which the participant recognized that the listener's acceptance of an initial description constitutes an agreement that constrains subsequent talk about the referent. We calculated the proportion of naming descriptions that were consistent (summed across experimental conditions). Interrater agreement was 95%, κ = .91.
In order to determine whether a participant designed his or her descriptions in accordance with changes in the listener's knowledge across trials and conditions, we determined trial length. This variable was operationalized as the mean number of words produced per trial. We expected that fewer words would be used as the interaction progressed over trials because of the accumulation of shared knowledge. Moreover, because the knowledgeable Listener 2 began Listener 2 trials with knowledge not available to the naive Listener 2, we predicted that fewer words would be used when interacting with the knowledgeable Listener 2 than with the naïve Listener 2
Diagnostic Group Comparisons of Referential Performance
Sample size varied across analyses because some dependent measures were meaningful only for some participants. We ensured that the diagnostic groups remained matched despite changes in the sample by including nonverbal MA as a covariate in all analyses. Proportions were transformed prior to analysis using an arcsine transformation; however, descriptive statistics are reported for the untransformed (and unadjusted) data. Because we expected the syndrome groups to do no better than and perhaps worse than the typically developing group, comparisons of the former with the latter following significant omnibus tests were one-tailed tests. All other tests were two-tailed.
We first analyzed the proportion of naming descriptions (n = 68) in a univariate ANCOVA, with group as a between-participants variable. The vast majority of referential descriptions were naming rather than componential descriptions. The proportions of naming descriptions ranged from .73 (SD = .36) for the typically developing participants to .65 (SD = .39) for the participants with Down syndrome. The main effect of group was not significant. We conducted a MANCOVA for the participants who produced one or more naming descriptions (ns = 23, 15, and 23 for the Down syndrome, fragile X syndrome, and typically developing groups, respectively). Group was the between-participants factor and the three dependent variables reflected the effectiveness of collaboration (i.e., the proportions of unique, referential frame, and consistent descriptions). The effect of group was significant, Wilks' Lambda F(6, 110) = 6.20, p ≤ .0005, and a partial λ2 of .25. The dependent variables are plotted by group in Figure 1.
Follow-up univariate F tests to the MANCOVA were evaluated using the Holm sequential procedure (Holm, 1979; Levin, Serlin, & Seaman, 1994) to prevent inflation of Type I error. For this procedure, the largest F had to reach an alpha of .017 to be significant (i.e., α/3), the next largest F, an alpha of .025 (α/2), and the smallest, an alpha of .05. All three univariate tests were significant by the Holm procedure: referential frames, F(2, 57) = 10.38, p ≤ .0005; consistent descriptions, F(2, 57) = 4.14, p = .02; and unique descriptions, F(2, 57) = 3.32, p = .04. Post-hoc comparisons of the groups for each dependent measure (using Fisher's LSD technique to maintain familywise α at p ≤ .05; Levin et al., 1994) indicated that (a) each syndrome group produced proportionally fewer unique descriptions than did the typically developing group, p ≤ .05, one-tailed; (b) the participants with Down syndrome produced proportionally fewer descriptions in referential frames than did either the fragile X, p ≤ .05, two-tailed, or typically developing group, p ≤ .0005, one-tailed; and (c) the participants with fragile X syndrome produced proportionally fewer consistent descriptions than did the typically developing participants, p ≤ .005, one-tailed.
We next conducted an ANCOVA on trial length (n = 68) to determine whether participants altered their descriptions in accordance with changes in shared knowledge. The design was a Group (Down syndrome, fragile X syndrome, typically developing) × Listener Knowledge (naive, knowledgeable) × Trial Block (Listener 1 trials, Listener 2 trials) factorial, with repeated measures on the last two factors. No effects were significant at the .05 level, suggesting that this variable was not, contrary to our hypothesis, an index of referential skill.
Examination of Putative Predictors of Referential Discourse Performance
Multiple regression was used to examine the relationship between each dependent measure of interest and the putative predictors. Because the number of predictors (including interactions) was large relative to the sample size, we adopted a conservative approach to model building that limited the number of predictors while still testing whether the pattern of relationships among variables varied across diagnostic groups. Proportional dependent measures were subjected to an arcsine transformation prior to analysis. Trial length was not included as a dependent variable in the regressions because, as noted previously, we did not find it to be a useful measure of referential skill.
The analysis for each dependent measure proceeded in three steps. In Step 1, two dummy variables were entered to represent the three diagnostic groups (Cohen, Cohen, West, & Aiken, 2003). One dummy variable indexed whether the participant had Down syndrome and the second, whether the participant had fragile X syndrome. A significant coefficient for one of these dummy variables signified that the indexed group differed from the typically developing group on the dependent variable. In Step 2, the following predictors were entered simultaneously: nonverbal MA, Test for Auditory Comprehension of Language age-equivalent, Oral Expression Scale age-equivalent, number correct on the false belief task, and number of correctly recalled sequences in digit span. In Step 3, we evaluated possible interactions between the predictors that were significant at Step 2 and each of the dummy variables representing the diagnostic groups. Each interaction term was represented by a single variable (i.e., the product of the dummy and predictor variable). A significant coefficient for an interaction term indicated that the relationship between the significant Step 2 predictor (e.g., nonverbal MA) and the dependent variable (e.g., proportion of unique descriptions) differed for the diagnostic group indexed by the dummy variable relative to the typically developing group.
This approach to testing interactions was reasonable because we expected that the interactions would be ordinal rather than disordinal (i.e., the relationships would vary in strength but not direction across groups), which means that significant interactions were likely to be associated with significant main effects as well. Because we expected each dependent measure to be related positively (or not at all) to each predictor, we evaluated main effects with one-tailed tests. All interactions involved two-tailed tests.
In analyzing the proportion use of naming descriptions, we found that the addition of the five predictors (main effects) at Step 2 did not lead to a significant change in the R2. No individual predictor was associated with a significant beta and, thus, the test of interactions at Step 3 was not necessary. The Step 1 model yielded an adjusted R2 of .02, F(2, 65) = 0.27, p = .76.
Analysis of the proportion of unique descriptions indicated that the addition of the five predictors at Step 2 was associated with a significant change in the R2, F(5, 51) = 3.80, p = .003. Only one of the five predictors yielded a significant beta at Step 2: Oral Expression Scale age-equivalent, β = .55, t = 1.99, p = .03 (one-tailed). In Step 3, we entered the two interaction terms for group and Oral Expression Scale score, which resulted in a significant change in the R2, F(2, 49) = 4.88, p = .01. The only terms associated with significant betas at Step 3 were the dummy variable indexing Down syndrome (relative to the typically developing group), β = −1.42, t = 2.42, p = .01 (one-tailed), and the variable reflecting the interaction of the Oral Expression Scale score and the dummy variable indexing Down syndrome, β = 1.28, t = 2.70, p = .01. This latter interaction reflects the fact the slope of the regression between the dependent variable and Oral Expression Scale score was greater for the Down syndrome group than the typically developing group (i.e., an effect of 1.83 for the former and .55 for the latter). The final Step 3 model yielded an adjusted R2 of .37, F(9, 49) = 4.76, p ≤ .0005.
In analyzing the proportional use of referential frames, we found that the addition of the five predictors at Step 2 led to a significant change in the R2, F(5, 51) = 4.72, p = .001. The only one of the five predictors that yielded a significant beta at Step 2 was the Oral Expression Scale age-equivalent, β = .49, t = 2.01, p = .03 (one-tailed). In Step 3, we entered the two interaction terms for group and Oral Expression Scale score, which resulted in a significant change in the R2, F(2, 49) = 4.17, p = .02. None of the variables in the Step 3 model, however, yielded a significant beta, although the final (Step 3) model yielded an adjusted R2 of .48, F(9, 49) = 6.91, p ≤ .0005. The implication of the Step 3 model may be that the relationship between the dependent variable and Oral Expression Scale score varied between the syndrome groups rather than between each of them and the typically developing group.
In analyzing the proportional use of consistent descriptions, we found that the addition of the five predictors at Step 2 was not associated with a significant change in the R2, F(5, 51) = 1.05, p = .40, with no predictor yielding a significant beta, and thus, no further analyses were conducted. The Step 1 model yielded an adjusted R2 of .09, F(2, 56) = 3.97, p = .02.
We also conducted regression analyses to examine the relationship between Child Behavior Checklist/4–18 scores and the three dependent measures. These analyses involved only the Down syndrome and fragile X syndrome groups and were conducted in the same way as those already described, with one set of analyses including Child Behavior Checklist/4–18 Total scores and another set involving the subscale scores as predictors. Child Behavior Checklist/4–18 scores did not contribute significantly to the prediction of any dependent variable.
Diagnostic Group Differences and Similarities in the Effectiveness of Referential Collaboration
The first goal was to determine the extent to which individuals with Down or fragile X syndrome effectively manage the referential process and whether there are syndrome differences in this regard. In general, we found that several, although not all, dimensions of the referential process were especially challenging for these youth (as evidenced by below-MA level performance), with both differences and similarities between the syndromes.
We found that the participants with Down syndrome or fragile X syndrome, like their typically developing MA-matched peers, generally preferred a naming strategy to a componential strategy. In general, the naming strategy is more effective and is the dominant strategy for school-age typically developing children and for adults (Ricard, 1993). Thus, the present findings suggest that selecting the general form of a referential description entails skills that emerge between the developmental levels of 3 and 6 years, regardless of diagnostic category.
Although participants with Down syndrome or fragile X syndrome expressed their referential descriptions in an appropriate form, they were less able to meet other basic requirements of effective collaboration in referential discourse. Indeed, below-MA performance was observed on three measures—unique descriptions, consistent descriptions, and referential frames—although the two syndromes displayed different profiles of relative difficulty on these measures.
Youth with Down syndrome or fragile X syndrome were less likely than were the typically developing children to create unique (i.e., one-to-one) mappings between their descriptions and the shapes. Instead, those from the syndrome groups tended to overextend their naming descriptions, using the same description for multiple shapes. This finding suggests that individuals from both syndrome groups do not fully consider the informational needs of the listener.
The failure of a speaker to create unique mappings can have negative consequences for social interaction because such a mapping can render the speaker's messages ambiguous. Ideally, a listener should respond to such a message by requesting clarification; however, young listeners, whether typically or atypically developing, often fail to request clarification and, thus, allow the confusion to continue (Abbeduto & Short-Meyerson, 2002). Even if the listener requests clarification, there is evidence that speakers with mental retardation have difficulty responding to such requests (Brady, McLean, McLean, & Johnston, 1995; Erbas, 2005). In short, a failure to create unique mappings may compound the difficulties that youth with Down or fragile X syndrome have interacting with peers and, thus, may be another barrier to their meaningful inclusion in the community.
We also found that individuals with fragile X syndrome were less likely than typically developing children to use consistent core descriptions as a shape recurred. Instead, they often invented new descriptions for recurring referents. This finding suggests that, as a group, youth with fragile X syndrome do not fully appreciate that a listener's acceptance of an initial description creates an agreement to continue using that description for subsequent references to that entity.
Although the use of a new description for a previously discussed referent does not preclude comprehension by the listener, such a description can have negative consequences for an interaction. The violation of such tacit “agreements” about the structure of social interaction can lead others to judge the speaker as inept and rude and can negatively color their future interactions (Grice, 1975). Changing the description of a referent can also increase the listener's processing burden and, thus, the risk of comprehension failures.
The greater referential inconsistency of youth with fragile X syndrome compared to typically developing children is surprising in light of the oft-cited claim that verbal perseveration is a hallmark of fragile X syndrome (Sudhalter et al., 1990). Indeed, perseveration in the present study would have led to a higher proportion of consistent naming descriptions and, thus, more “mature” referential performance, at least on this dimension. The fact that the youth with fragile X syndrome in this study were less perseverative (i.e., more inconsistent) suggests that there is a need for a more nuanced characterization of the language problems of this syndrome.
Syndrome differences also emerged in the use of referential frames, with the participants with Down syndrome using proportionally fewer referential frames than either of the other two groups. This finding does not reflect simply a tendency for the participants with Down syndrome to produce shorter descriptions because no group differences were found for description length (in words). Thus, this finding suggests that youth with Down syndrome are less likely to recognize the utility of providing scaffolding through use of a referential frame.
Although the failure of speakers with Down syndrome to use referential frames does not preclude comprehension of their referential descriptions, it can increase the listener's processing burden and that, in turn, could increase the likelihood of misunderstandings. Moreover, this tendency to require the listener to shoulder more of the conversational burden than is typical may lead other people to avoid future interactions with the speaker who has Down syndrome. Interestingly, other evidence has begun to emerge suggesting that despite their relatively high levels of sociability, individuals with Down syndrome have quite limited social skills (Kasari & Bauminger, 1998).
Unexpectedly, we did not find any variation in length of the referential descriptions produced by our participants either across diagnostic groups or experimental conditions. We had anticipated that descriptions would be pared down as shared knowledge accumulated (Clark, 1996). Our null result may reflect the fact that the participants in this study were developmentally less mature, on average, than those in studies that have documented a length effect (e.g., Ricard, 1993).
In summary, the group comparisons clearly demonstrate that referential talk consists of several separate, but interrelated dimensions. The implication for clinical practice is that assessments of referential skills must be comprehensive and fine-grained so that the individual's facility with each dimension of the referential process is evaluated and targeted for intervention as needed. Currently, however, no standardized means of assessing these important dimensions of the referential process are available and, thus, the problems we have uncovered may go undiagnosed and untreated.
Putative Predictors of Effectiveness of Referential Collaboration
Our second goal In this study was to determine how referential talk is shaped by a speaker's levels of language, theory of mind, auditory memory, and maladaptive behavior and whether there are differences between the syndromes and MA-matched typically developing children in this regard. In fact, our measures of these domains accounted for relatively little of the variation in referential talk.
Only scores on the Oral Expression Scale, which measures the range of lexical and grammatical options in an individual's expressive repertoire, made a unique contribution to referential performance, being significantly related to the use of unique descriptions and referential frames. It is, perhaps, not surprising that the Oral Expression Scale was related to performance in the referential task in that a limited expressive repertoire will mean that there are fewer “tools” to draw on to create a unique mapping between a referent and description or to construct a referential frame to scaffold the listener's comprehension. Indeed, the slope relating the measure of use of unique descriptions to Oral Expression Scale scores was greater for the Down syndrome group than for the typically developing children, indicating that limitations in expressive language are particularly important for individuals with Down syndrome, who generally have especially severe expressive impairments (Chapman & Hesketh, 2000).
The findings also suggest, however, that referential talk requires more than a repertoire of words, grammatical forms, and other linguistic tools. First, expressive language skill (measured by the Oral Expression Scale) did not account for variation in the use of naming descriptions or consistent naming descriptions. Second, the youth with fragile X syndrome displayed a high rate of inconsistent descriptions despite displaying MA-consistent levels of vocabulary and grammar. Third, the difference in use of unique descriptions between the Down syndrome and typically developing groups remained significant even after accounting for variation in Oral Expression Scale scores.
Together, these findings suggest that although variations in referential performance are related to differences in expressive language ability, especially for individuals with Down syndrome, referential discourse poses its own set of skills to be learned and processing demands to be met. This conclusion implies that current approaches to language intervention, which have traditionally focused on increasing vocabulary and encouraging the use of more complex grammatical forms (Brady & Warren, 2003), may have minimal effects on the abilities to (a) create unique mappings between descriptions and referents, (b) maintain the core of a referential description over a discourse, and (c) use referential frames to scaffold comprehension, at least for youth in the developmental range included in this study. Thus, the scope of language intervention may need to be expanded.
Unexpectedly, social cognition, auditory memory, and maladaptive behavior, which are defining dimensions of the phenotypes of Down syndrome and fragile X syndrome (Dykens et al., 2000), did not make unique contributions to variations in referential performance. This may reflect limitations of the measures or of the sample size. It is interesting to note, however, that despite claims that theory of mind provides the foundation for communication and much of social interaction (Tager-Flusberg, 2001), several previous researchers have also found weak or no correlations between various aspects of language use and performance on false belief tasks (e.g., Abbeduto & Hesketh, 1997). This raises the possibility that the ability to reason about mental states in the abstract is not involved in “on-line” communication (Klin, Jones, Schultz, Volkmar, & Cohen, 2002).
In the case of auditory memory, it may be that individual differences in this domain are more important in acquiring than in using the tools of language (e.g., vocabulary) in discourse. It is also possible that auditory memory is less important for speaking than for listening or at earlier stages of development than represented in the present sample.
The null findings for maladaptive behavior might reflect the fact that such behavior is context dependent; for example, attentional problems might contribute more to communication in a context in which distractions are more numerous than in the present study, which required that the participant talk to a highly skilled partner in a quiet room with few visual distractions. It will be important, therefore, to extend the present findings either by examining referential discourse in natural environments or by systematically manipulating the conditions thought to provoke maladaptive behaviors to the extent that it is possible and ethical to do so.
We must acknowledge three limitations of the study. First, the sample size was small and, therefore, the findings must be replicated with larger samples. This will be especially important so that possible differences between males and females with fragile X syndrome can be evaluated (Murphy & Abbeduto, 2003). Second, although we examined language use in relation to several important domains of behavioral functioning, we sampled only a few of the domains that define the phenotypes of Down and fragile X syndromes. Third, we included only individuals who did not also have an autism diagnosis. However, 25% of individuals with fragile X syndrome may have an autism diagnosis (Philofsky, Hepburn, Hayes, Hagerman & Rogers, 2004; Rogers, Wehner, & Hagerman, 2001) and, thus, the extent and nature of their referential problems remains to be determined.
Two strengths of our approach should also be acknowledged. First, we have shown the utility of conceptualizing language use as multidimensional, with different dimensions entailing different skills and different challenges across different populations. Second, we have shown the utility of including more than a single syndrome group so that we can begin to distinguish those language problems that are syndrome-specific and those that reflect mental retardation more generally.
The research reported in this article was supported by Grants R01 HD24356 and P30 HD03352 from the National Institutes of Health and by the Graduate School Research Committee of the University of Wisconsin-Madison. The authors thank the families who participated for their time, patience, and enthusiastic support. Thanks also to Randi Hagerman and Marcia Braden for their advice on recruitment and testing accommodations; Susan Kirkpatrick for her assistance with interpreting genetic reports; Robert Nellis, Jill Roxberg, and Patti Johnstone for conducting the audiological evaluations; and Pamela Lewis for assistance with the autism diagnoses. Brief summaries of preliminary analyses of some of the data presented here have appeared in chapters by Abbeduto and Murphy (2004) and Murphy and Abbeduto (2003). Requests for reprints should be sent to Leonard Abbeduto, University Wisconsin-Madison, Waisman Center on Mental Retardation and Human Development, 1500 Highland Ave. Madison, WI 53705. email@example.com