Quantitative Analysis of Disfluency in Children with ASD

Background:  Difficulties with pragmatic language, including conversational reciprocity, are common in children with autism spectrum disorders (ASD). Compared to non-affected children, children with ASD have difficulty initiating conversation (Tager-Flusberg 1996), responding to the initiations of others (Capps et al. 1998), taking turns (Botting & Conti-Ramsden 2003), and staying on topic (Losh & Capps 2003, Paul et al. 2009). Another element of conversational reciprocity is the use of disfluencies, which act as interpersonal signals between speaker and listener. On hearing a disfluency, a listener may infer that the speaker is experiencing difficulty planning or delivering speech (Clark & Fox Tree 2002) or that the speaker wishes to correct earlier speech (Clark 1994). We hypothesized that this element of conversational reciprocity would be another area of difficulty for individuals with ASD. Indeed, several studies find that children and adults with ASD produce different patterns of disfluency than nonaffected individuals (Lake et al. 2011, Shriberg et al. 2001, Suh et al. 2014, Thurber & Tager-Flusberg 1993), though these studies report conflicting results concerning the exact nature of these differences.

Objectives: We conducted a large-scale exploratory study to better quantify disfluency use in children with ASD.

Methods: 110 children ages 4;0-9;0 participated, in three groups: ASD (n = 50), specific language impairment (SLI; n = 17) and typical development (TD; n = 44). All children were high functioning monolingual English speakers. Diagnoses were verified by best-estimate clinical consensus. A clinician administered the Autism Diagnostic Observation Schedule (ADOS; module 2 or 3) to each child. These sessions were recorded and transcribed, and all disfluencies (n = 12,888) were annotated and tabulated. Disfluencies were either categorized as "fillers" ("uh", "um", "mm", etc.) or "content mazes", including repetitions (e.g., "...and these---and these too"), revisions (e.g., "He's waterskiing with a balloon---with a parachute."), and false starts (e.g., "He got like---it was this finger.").

Results: Disfluencies were analyzed using mixed effects logistic regression; covariates included chronological age, full-scale IQ, mean length of utterance in morphemes (MLU), maze position (i.e., utterance-initial vs. non-initial), and ADOS activity. The three groups exhibited comparable overall rates of disfluency. However, there were significant group differences in the relative frequency of fillers vs. content mazes (p < .001). Post-hoc tests revealed that children with ASD produced a higher rate of content mazes (compared to fillers) than children with SLI (p = .030) or typical development (p < .001). There was also a significant interaction between group and maze position: typically-developing children tended to produce content mazes initially and fillers non-initially, but children with ASD showed the opposite pattern. Disfluency type was not associated with age, IQ, or MLU.

Conclusions: Children with ASD (but not children with SLI) produce different patterns of disfluencies than TD children, including significantly higher rates of content mazes. If confirmed, this result may provide clinicians with a novel feature distinguishing ASD, and may ultimately be a promising target for intervention in verbal, high-functioning children with ASD.