Disfluency: The Cost of Attending to Listener Feedback?

Hannele Nicholson

The current paper analyses eye-gaze and disfluencies, instances where communication goes awry, in 96 interactive Map Task dialogues, to determine whether there is a cognitive cost associated with interaction with a listener. Previous disfluency studies have argued that disfluencies are strategic signals to a listener in times of difficulty (Clark & Wasow, 1998). Other research suggests that disfluencies are a possible indication of cognitive overload (Bard et al., 2000). Bard et al. (2003) demonstrate that interaction with a follower could induce more difficulties for a speaker.

All 96 trials were coded for Move type (Carletta et al., 1997). A Move is a sub-unit of the conversation in which the Instruction Giver instructs the Instruction Follower along the route or asks a clarifying question. For each Move time span, I analysed:

A. whether the Follower's position was Correct or Wrong

B. whether the Giver's eye-gaze attended to the Follower's location

C. whether the time span contained a disfluency

An ANOVA of Time-pressure by Gaze revealed an association between disfluent Moves and Giver's gaze at the Follower's square when it was on a correct landmark (F1(1,23) = 39.97, p < .001). IGs are more disfluent when they attend to the correct Feedback square (Mean = 0.222) than when they do not (Mean = 0.076). Disfluency rate is greater in longer Moves (F1(1,23) = 14.447, p < .001) but the gaze effect is still significant amongst short Moves (F1(1,23) = 7.20, p < .02). This finding is analysed in terms of whether or not in shows an associated cost for attending to a listener's feedback.