Abstract
People can collaborate during a search task (Zelinsky et al.; VSS 05), but how does this collaboration translate into a division of search labor? Previous work demonstrated a spatial division of labor, but did this finding hinge on the availability of information in the task? To discover whether other coordination strategies might spontaneously emerge, we had groups of 2, 3, or 4 people engage in 3 collaborative search tasks: (1) searching for an oval dot among 494 black circular dots, (2) a color version of Experiment 1 having red, blue, green, and black dots partially segregated into non-geometric regions, and (3) a multiple-target task in which subjects searched for 1, 4, or 8 possible targets in 14-item displays (photo-realistic objects). We quantified division-of-labor by correlating the targets' properties (e.g., location, color, identity) with the responses of individual searchers from a collaborating group. Consistent with previous work, subjects in Experiment 1 divided the search labor spatially, either splitting the display in two (2-person condition) or by quadrant (4-person condition). Experiment 2 produced very different results. Despite an identical configuration of dots, subjects now divided the search labor by feature, with each member of a 4-person group searching a different distractor color. Experiment 3 yielded evidence for yet another form of division-of-labor. Subjects divided the task by target rather than by space or feature, with searchers in the 4-person group each taking responsibility for a different potential target from the 4-object target preview. We conclude that subjects do indeed use spatial division-of-labor collaborative strategies as originally reported, but only when the task does not allow for more meaningful divisions of search labor. When the distractor set is multi-colored, or when the task requires searching for multiple targets, simple spatial division-of-labor strategies are replaced by collaborative strategies based on feature and target information.
This work was supported by National Science Foundation grant 0527585