When “Easy” Conversations Seem Harder: Filler Words and Social Context in Adults with ASD

Background:  Social communication is a core challenge in autism spectrum disorder (ASD), but has proven difficult to reliably measure, especially in natural contexts. In this study, we use techniques from Natural Language Processing and theories of conversational dynamics to assess speech during different types of informal conversations. Speech disfluencies occur when speakers pause, revise, or add filler words to otherwise fluent speech, and may indicate problems with planning or may serve as a pragmatic tool to communicate information with interlocutors (Arnold, Fagnano, & Tanenhaus, 2003; Bell et al., 2003). Prior research shows that individuals with ASD generally exhibit higher disfluency rates than matched controls, but it is not yet known how different kinds of conversational settings affect fluency in this population.

Objectives:   Compare disfluency rates, measured by filler words, in adults with ASD and typical adult interlocutors during conversations that are socially “easy” (i.e., with an interested interlocutor) and socially “hard” (i.e., with a bored interlocutor).

Methods:   Twenty-nine adults with ASD (mean age=27y; mean IQ=104) engaged in 3-minute conversations with two different typical adult confederates as part of the Contextual Assessment of Social Skills (Ratto et al., 2011). The first confederate acted interested in the participant (Interested condition), while the second acted bored (Bored condition). Audio recordings of both conversations were orthographically transcribed. Percentage of filler words (e.g., uh, um, er, eh) relative to total words produced was calculated using Linguistic Inquiry and Word Count software (Tausczik & Pennebaker, 2010).

Results: A repeated measures ANOVA with condition (Interested, Bored) and speaker (Participant, Confederate) as within-subjects factors revealed a significant interactive effect on filler rates, F(1,28)=8.60, p=.007; Figure 1). Whereas adults with ASD produced a significantly higher percentage of filler words in the Interested condition (M=7.34%) than the Bored condition (M=6.34%, t(1,28)=-2.42, p=.02), typical confederates showed the reverse pattern, producing significantly fewer filler words in the Interested condition (M=5.25%) than the Bored condition (M=6.90%, t(1,28)=2.26, p=.03). Comparing filler word rates across diagnostic groups revealed no difference in the Bored condition, t(1,28)=-.59, p=.56). Significant group differences emerged in the Interested condition t(1,28)=2.61, p=.01; Figure 1).

Conclusions:   Subtle language differences can influence whether social communication is successful, and examining conversational dynamics could shed light on meaningful heterogeneity in this domain. In this study, we found that rates of filler words were affected by social context in adults with ASD, with higher rates occurring during interactions that may otherwise be seen as socially “easy” (i.e., with an interested interlocutor) and lower rates when interactions are socially “hard” (i.e., with a bored interlocutor). Using more filler words during conversations with an interested partner could indicate increased responsiveness to social expectations (positive or negative), and will be clarified as we conduct the next iteration of this study on typical adults conversing with typical confederates (anticipated by May, 2017). Our findings have implications for understanding the effect of context on social communication in ASD, and may be valuable for clinicians interested in improving conversational competence in adults.