Selection of objects and tasks in working memory

Abstract

When people hold several objects (such as digits or words) in working memory and select one for processing, switching to a new object takes longer than selecting the same object as that on the preceding processing step. Similarly, selecting a new task incurs task- switching costs. This work investigates the selection of objects and of tasks in working memory using a combination of object-switching and task-switching paradigms. Participants used spatial cues to select one digit held in working memory and colour cues to select one task (addition or subtraction) to apply to it. Across four experiments the mapping between objects and their cues and the mapping between tasks and their cues were varied orthogonally. When mappings varied from trial to trial for both objects and tasks, switch costs for objects and tasks were additive, as predicted by sequential selection or resource sharing. When at least one mapping was constant across trials, allowing learning of long-term associations, switch costs were underadditive, as predicted by partially parallel selection. The number of objects in working memory affected object-switch costs but not task-switch costs, counter to the notion of a general resource of executive attention.

Keywords:

• Working memory
• Executive control
• Object switching
• Task switching
• Interference
• Retrieval
• Memory load

Acknowledgments

This work was supported by Grant OB 121/3–3 from Deutsche Forschungsgemeinschaft (DFG).

Notes

¹ In the condition with set size two, we always used the same two digits. Thus, one could argue that although all four digit–cue associations are well learned, the subset of two digits that are used in the set size two conditions are learned better and, thus, can be retrieved faster. To address this possibility, we ran the analysis of set-size effects in Experiments 3 and 4 again, using only those RTs that required access to one of the two digits that were used in both the small and the large set-size conditions. For this comparison, cue–digit associations in LTM must be equally strong in the set size two and set size four conditions. The interaction between object switching and object set size was still significant in Experiment 3, F(1, 19) = 11.5, MSE = 10,691, p = .003, and in Experiment 4, F(1, 19) = 23.0, MSE = 3,193, p < .001. Therefore, the set-size effect cannot be due to differential practice of the cue–digit associations.