Abstract: PURPOSE: The world around us contains multiple objects, and each object consists of multiple visual features. Successful feature binding integrates discrete features into a holistic object representation, which means the establishment of visual object integrity. Attention has an impact on feature binding and object integrity, at both perception level and working memory level. At the memory level, most previous studies have used a dual-task paradigm, manipulating attentional resources rather than directly manipulating visual attentional conditions to observe the maintenance and extraction of binding features in working memory. Here, we attempt to explore the maintenance of feature binding and object integrity by manipulating dynamic attention in working memory directly.
METHODS: A joint continuous reporting paradigm was adopted. A 4-segments display will be shortly presented, and after a while spatial cues will appear to manipulate dynamic attention in working memory. We designed three attentional conditions (hold, shift, and split) in two experiments (16 participants in experiment 1 and 18 in experiment 2). In the hold and shift attention blocks, the participants should pay attention to the location of the upcoming cues. If a second spatial cue appears, they should shift their attention to the new cue location. Hold and shift trials were intermixed in a block to ensure that participants had to attend to the first cue and could not simply wait for the second cue. For split attention blocks, two cues appeared simultaneously to make sure that the participants pay attention to the two cue locations, and after a while a third cue will appear at one of two cued locations. For all the conditions, participants were required to make a joint continuous report to indicate the color and orientation of the target (T) indicated by the last cue. Of the three nontarget items, the critical nontarget (N1) appeared adjacent to the target, which was either initially cued in shift trials or simultaneously cued in split trials; N2 appeared at the other adjacent location and N3 appeared at the diagonal location to target. The stimuli and procedure in experiment 2 were same as experiment 1, except that the interval time between the two spatial cues changed from 1000ms to 300ms in the shift attention trials. The participants' reports on the color and orientation of target were collected on each trial to examine whether the errors were correlated or independent when recalling multiple features of the same object under different spatial attention conditions. The different types of errors include (a) correlated swap (e.g., N1_CN1_O), or (b) independent T errors which was only reporting one feature of target, including unbound guesses (e.g., T_CU_O) and illusory conjunction (e.g., T_CN1_O).
RESULTS: The majority of responses were reporting both the color and orientation of the target object (correlated target responses), and the error patterns vary with attention conditions. In experiment 1, compared to hold attention condition, attention shift resulted in both greater feature errors (i.e., less probability of target responses) and less precision of color feature. Splitting attention also decreased the precision of color feature, but did not impact the probability of target responses. We analyzed the error types and found that, compared to hold attention, shifts of attention were more likely to maintain object integrity or feature bindings (more correlated N1_CN1_O swaps and less independent T* errors). In all attention conditions, unbound guesses occurred significantly less than illusory conjunctions, which suggests that participants were more likely to report a feature of another object than to guess another feature. In experiment 1, in shift trials, the long processing of N1 (1000 ms) would cause loss of memory about the features of the real target. To avoid it, in experiment 2 we changed the SOA of two spatial cues to 300 ms in the shift trials. The decrease of SOA did increase the accuracy of the reported target under attention shifting condition. Compared to hold attention condition, shifting and splitting attention reduced the probability of target responses and the accuracies of both features in the joint-feature report. We analyzed the error type and found that the types of errors made in the different dynamic attention conditions were similar, and the object integrity can be maintained to a same extent. In all attention conditions, unbound guesses occurred significantly less than illusory conjunctions.
CONCLUSIONS: Dynamic attention in working memory affects feature binding and the maintenance of object integrity. Attention splitting will degrade the integrity of objects which are maintained in working memory. Compared with attention splitting, attention shifting maintained the integrity of the object in working memory to a higher degree, especially when the shifting occurs relatively late.

Key words: feature binding, dynamic attention, attention shifting, attention splitting, working memory