A pervasive theoretical assumption is that cognition strictly precedes action, like the old saw, “think before you act”. Information accumulates until enough evidence for one decision over others accrues. Thereafter, action simply executes the decision. This assumption, whether implicit or explicit, is most evident in how cognitive performance is studied (Rosenbaum, 2005). Response outcomes, typically as response latencies or accuracies, are the standard measures of cognitive performance. Measuring only the end of a response suggests what intervenes between the onset of a stimulus and the completion of a response is irrelevant. However, recent work guided by dynamic systems theory suggests the relation between cognition and action may actually be one of reciprocal interdependence, where cognition guides action, but action also feeds back to reinform cognition (Stephen, Dixon, and Isenhower, 2007; Van Orden, Holden,and Turvey, 2003, 2005).
In the current studies, participants perform a yes/no semantic categorization task to exemplars (tulip; a type of flower?), nonexemplars (stove; a type of flower?), ambiguous homophones (rows; a type of flower?), and spelling controls (robs; a type of flower?). Trials are completed by navigating a mouse cursor on computer screen to either a ‘yes' or ‘no' response box. Participants' response trajectories are tracked on-line as they unfold. The current results demonstrate that overt changes from one response to another are anticipated by increased instability in motor performance, just prior to the change.
Surprisingly, ambiguous homophones result in the same number of decision changes as all other words. Ambiguity does not manifest itself as a competition between multiple response options, as is commonly assumed (Gottlob, Goldinger, Stone, and Van Orden, 1997; Plaut, McClellend, Seidenberg, and Patterson, 1996; Rubenstein, Garfield, and Millikan, 1970; Siakaluk, Pexman, Sears, and Owen, 2007; Spivey, 2006). Instead, the homophone ambiguity effect in semantic categorization appears appears to be a tendency to only making errors when participants are certain those errors are correct. When this confound is controlled for, the widely replicated finding that homophones have longer response latencies than spelling controls (Van Orden, 1987) disappears.