Dissertations, Theses, and Capstone Projects

Date of Degree

9-2026

Document Type

Doctoral Dissertation

Degree Name

Doctor of Philosophy

Program

Psychology

Advisor

Kerstin Unger

Committee Members

Kerstin Unger

Veronica J. Hinton

Justin Storbeck

Yvette Caro

Sally Izquierdo

Subject Categories

Cognitive Psychology | Cognitive Science

Keywords

Working Memory, Output Gating, gating framework, retro-cue, Selective, Global

Abstract

Working memory (WM) keeps information accessible and available for further processing. However, not all the information we maintain in WM is behaviorally relevant at any given moment. To act effectively, we must dynamically control which memory representations are accessed to support ongoing goal-directed behavior. While the mechanisms underlying the selection of single items from WM access are comparatively well-studied, less is known about the principles governing multi-item access. Existing models of WM access imply that selecting multiple items should take longer and be more error-prone than single-item access. Yet, paradoxically, prior studies have shown that retrieving all items in WM (global access) can be faster and more accurate than selectively retrieving subgroups of items (selective access), even though global access involves a larger set of relevant items. In the present work, we explored this puzzle, focusing on the mechanisms underlying WM multi-item access. In Experiment 1, we tested whether the performance advantage of global over selective access can be attributed to the interplay of inhibitory and facilitatory mechanisms. Specifically, we asked whether the need to inhibit irrelevant items reduces retrieval efficiency in selective WM access. To this end, we manipulated the number of irrelevant items across three conditions in a spatial WM task: participants retrieved one item (Selective-1), two items (Selective-2), or all three items (Global) from a previously encoded memory set. Thus, the number of irrelevant items—and, by extension, inhibitory demands—decreased from Selective-1 (two irrelevant items) to Selective-2 (one irrelevant item), to Global (no irrelevant item). Contrary to the notion that greater inhibitory demands account for the cost of selective WM access, we found that retrieval times were faster in the Selective-1 condition compared to the Selective-2 condition. Nonetheless, global access remained faster than both selective conditions. Our findings rule out inhibition as the primary source of the performance difference between selective global WM access and instead support the view that selective and global access rely on distinct mechanisms. Selective access is thought to rely on bindings between the memory items and individual context representations (e.g., an item’s spatial location) that can act as item-specific retrieval cues. In contrast, global access may depend on a shared context representation, such as the overall trial context. Whereas item-specific contexts must be (re-)activated to initiate selective retrieval, the global context is already active at the time of retrieval, potentially allowing faster and less error-prone access. In Experiments 2a and 2b, we further explored why selecting multiple items from WM (Selective-2) is less efficient than selecting a single item (Selective-1). We hypothesized that the reduced efficiency of multi-item access could result from competition at two possible levels: the item level and the context level. To dissociate these factors, we independently manipulated the number of cued contexts (spatial locations) and the number of relevant items in a spatial WM task. We found that responses times increased with the number of cued locations but not with the number of selected items. These results indicate that the primary bottleneck in selective WM access lies in activation multiple context representations, rather than in selecting multiple items per se. When two items were associated with a single context (i.e., appeared in the same location), dual‑item retrieval incurred little extra cost, mirroring the efficiency of global access. Our findings have important implications for theories of WM selection by identifying context-level competition as a key constraint on multi-item access. Nonetheless, they leave open the question of whether the observed bottleneck reflects limitations in stimulus-driven (bottom-up) processing or in top-down control mechanisms. Future studies may use neuroimaging and/or electrophysiological techniques to offer a more complete picture of how stimulus-driven and top-down processes impose limits on multi-item selection—and why these limits are more pronounced at the context level. For instance, new insights could be gained by comparing the neural decoding of context-specific and item-specific information for single- vs. multi-context cue conditions, and by examining whether such information emerges first in higher-order control regions before propagating to lower-order sensory regions—consistent with a top-down biasing account.

Download

Share

COinS