Abstract
Visual short-term memory (VSTM) has been claimed to maintain 3–5 feature-bound object representations. A few studies investigating memory for feature binding report significantly smaller capacity estimates (Saiki, 2003; Wheeler & Treisman, 2002), which have been interpreted as the effects of interference in memory retrieval, based on the finding that retrieval cues improve performance. However, change detection tasks may not properly evaluate feature-bound representations in VSTM, because observers can detect a change with other types of memory representations. To overcome this problem, we devised modified tasks, type identification and relevant-feature switch detection, to evaluate memory for feature binding more directly. With objects defined by shape, color, and spatiotemporal location, and changes being feature switches across objects as in previous studies on binding memory, the type identification task asked observers to identify the switch type among 4 alternatives (shape, color, shape-and-color, and no-switch). The relevant-feature switch detection required observers to monitor either color or shape alone, and to ignore the other. Both tasks need discrimination of different types of change, particularly between color- and shape-switch, not only detecting the presence of any change. Using a paradigm called multiple object permanence tracking (MOPT), where observers monitored a switch in feature combination between objects during an occlusion period, we manipulated memory retrieval by providing a cue right after a switch occurs. We found that retrieval cues revealed no facilitation with both type identification and relevant-feature detection tasks. In contrast, MOPT with the simple change-detection task showed significant facilitation, replicating previous findings. Additionally, object motion eliminated the cuing effect even with the simple change detection task. These results suggest that binding deficit in VSTM reflects memory maintenance, not retrieval, and that previous findings of retrieval effects are likely produced by memory unrelated to feature binding, such as saliency.
Support from 21st Century COE (D-10 to Kyoto Univ.), and PRESTO from JST.