Disfluencies reflect a...hm... competition between response options: evidence from a drift diffusion analysis

Abstract

Disfluency can occur when a speaker faces difficulty in language production, but also as a strategy to stall for time and create an illusion of continuity in speech. To better understand the origin of disfluency, the current study used a drift diffusion model (DDM) in a context of semantic competition. We developed a speeded version of the picture–word interference paradigm, in which participants name pictures while ignoring distractor words. We applied the DDM approach to data from related vs. unrelated distractors but also from fluent vs. disfluent answers (e.g. “apple” vs. “hm...a-apple”). All results were mapped onto the drift rate parameter. Unrelated distractor words resulted in a higher drift rate compared to semantic distractors, and fluent answers resulted in a higher drift rate compared to disfluent answers. Results suggest that picture-word interference taps into the process of spreading activation through the lexical-semantic system. Similarly, disfluency reflects competition between response options, and is not a strategy from the speaker to stall for time to accurately name the picture.

Semantic interference affects speech production by increasing disfluencies, not errors

“…Therefore, we demonstrated the effects of semantic interference on disfluency production. These findings are in line with a previous study that examined disfluency production in a PWI, using time pressure as well [15]. However, this previous study was based on single picture naming rather than connected-speech production, and the proportion of disfluent answers was much lower than what we found in the current study (around 7% of disfluent trials).…”

“…In our previous experiments [ 6 , 8 ], the time it took the point marker to complete one network was 42 s, which was identified as the natural speech rate in Dutch. However, in the current study, we reduced the time to complete a network to 35 s in order to increase the effect of semantic interference [ 15 ].…”

“…We expected more errors (i.e. distractor naming) in the semantically related condition than in the unrelated condition, as observed in single picture naming [ 15 ]. However, a substantial number of errors and disfluency was expected in both conditions, compared with previous network tasks that manipulated lexical access.…”