Feature bindings endure without attention: Evidence from an explicit recall task

JAMES R. BROCKMOLE 0 1 2 0 D.A.G. was supported by NSF-IGERT Grant DGE-0114378, and J.R.B. was supported by NSF Grant BCS-0094433 awarded to John M. Henderson. We thank John Henderson, David Irwin, Erik Altmann, Geof- frey Woodman, and two anonymous reviewers for helpful comments on this research ; Sonia Dutt and Paul Turnpaugh for help with data collec- cerning this article should be addressed to D. A. Gajewski, Department of Psychology, Michigan State University, 211 Psychology Building , East Lansing, MI 48824-1116 ( 1 University of Edinburgh , Edinburgh, Scotland 2 DANIEL A. GAJEWSKI Michigan State University , East Lansing, Michigan Are integrated objects the unit of capacity of visual working memory, or is continued attention needed to maintain bindings between independently stored features? In a delayed recall task, participants reported the color and shape of a probed item from a memory array. During the delay, attention was manipulated with an exogenous cue. Recall was elevated at validly cued positions, indicating that the cue affected item memory. On invalid trials, participants most frequently recalled either both features (perfect object memory) or neither of the two features (no object memory); the frequency with which only one feature was recalled was significantly lower than predicted by feature independence as determined in a single-feature recall task. These data do not support the view that features are remembered independently when attention is withdrawn. Instead, integrated objects are stored in visual working memory without need for continued attention. - Whereas attention is generally thought to play a role in the binding of features into integrated object percepts (see, e.g., Treisman, 1986, 1999; Treisman & Gelade, 1980; Wolfe & Cave, 1999), the role of attention in the maintenance of feature bindings in visual working memory (VWM) is currently being debated. Determining the fragility of the feature bindings is critical because it differentiates among competing views on the representational format of VWM. One view holds that VWM stores integrated object representations (Luck & Vogel, 1997; Vogel, Woodman, & Luck, 2001; see also Irwin & Andrews, 1996). Implicit in this view is the idea that an object is represented in memory as a singular structure that either remains coherent when attention is withdrawn or is entirely lost. A second view holds that features are stored independently and attention is needed to maintain proper associations between them (Wheeler & Treisman, 2002). According to this view, removing attention should sever the appropriate bindings, leaving potentially good memory for the features that were present but poor memory for their correct combinations. Support for the object-unit hypothesis was provided in a series of influential studies employing a change detection paradigm (Luck & Vogel, 1997; Vogel et al., 2001). Briefly presented memory displays containing objects comprising two or more simple features (e.g., colored bars in various orientations) were followed by a blank interval and a test display. In separate blocks, participants were to detect changes in one or either of the two feature dimensions. Accuracy depended on the number of objects present in the display, as opposed to the number of features that needed to be remembered. Critically, feature capacity increased even when objects were defined by multiple features from the same feature dimension (e.g., bicolored squares). If equivalent performance for single-feature and multifeature objects were a result of separate stores for each of the feature dimensions, this increase in the number of features from the same dimension should have hindered performance. Thus, these results suggested that integrated objects are the unit of capacity of VWM. The independent-stores hypothesis is based on two criticisms of the Luck and Vogel (1997; Vogel et al., 2001) experiments. First, Wheeler and Treisman (2002) were unable to replicate the finding that feature capacity increases when the objects are comprised of features from the same dimension (see also Olson & Jiang, 2002; Xu, 2002). Second, because the changes in the experiments of Luck and colleagues involved only a single feature of a single item, performance could have been based on memory for the features that occurred in the display without knowledge of which features occurred together. To pursue these issues, Wheeler and Treisman employed a binding condition in which a pair of features were swapped between two items in the initial display. A memory display containing a yellow square and a red triangle, for example, would be tested with a display containing a red square and a yellow triangle. A decline in performance in the binding condition was observed relative to performance in conditions analogous to those used by Luck and colleagues. As a result, Wheeler and Treisman concluded that features from different dimensions are stored in separate caches in VWM and that these dimensions have independent capacities. Interestingly, this binding-specific decrement was alleviated when memory was tested using a single probe item rather than the whole display. This difference was explained by the suggestions that the preservation of bindings in VWM requires focused attention, and that the onset of the multi-item test display disrupted attention to the feature bindings to a greater degree than did the onset of the single-item test probe. In the present study, we pitted the object-unit and independent-stores hypotheses against each other by directly investigating the role of attention in the maintenance of feature bindings in VWM. The fundamental question at stake is whether attention, in addition to creating bound percepts of multifeature objects, is also required to preserve those feature bindings in VWM. One potential approach to testing whether or not attention is required to maintain feature bindings in memory would be to manipulate attention during a memory retention interval in a change detection task. In such a situation, the independent-stores hypothesis would predict that a disruption of attention would also disrupt memory for specific feature bindings. The object-unit hypothesis, on the other hand, would predict that variations of attention would have little impact on memory for feature conjunctions. Through this approach, however, support for the object-unit hypothesis would entail the acceptance of a null hypothesis. Therefore, we developed a methodological approach that independently generated specific predictions for memory performance based on the object-unit and independent-stores hypotheses when attention is disrupted. We then contrasted these predicted outcomes against actual performance. Our approach to the problem had two primary components. First, we employed an exogenous cue to manipulate attention during the memory delay. Nonpredictive cues presented after the offset of a memory display have been shown t (...truncated)