Abstract
Two experiments were conducted to (1) determine how gaze serializes objects in working memory, and (2) study the retrieval process contributing to working memory failure. Both studies used a gaze-contingent working memory paradigm to obtain forgetting functions for objects presented in complex scenes. Exp 1 had observers (n = 6) freely view 9-item scenes in order to remember each object's identity and location. After gaze left a pre-determined target, observers could fixate 1–7 intervening non-target objects before the scene was replaced with a spatial probe at the target location. The task was to select the target from among 4 alternatives. Consistent with studies using sequential presentation, a steep recency benefit was found over the 1–2 intervening object range (87% and 76%, respectively) that declined into a flat level of pre-recency accuracy (65%) over the remainder of the forgetting function. Exp 2 asked whether the Exp 1 working memory constraints reflected encoding or retrieval limitations. The procedure was identical to Exp 1 except that the test was 9AFC rather than 4AFC (n = 10). Importantly, the encoding conditions were the same between the two experiments. Results showed a dramatic decline in recency relative to Exp 1, with accuracy after 1 and 2 intervening objects dropping by 17% and 14% respectively. We interpret these data in terms of “fuzzy object file theory” in which fixations on intervening objects interfere with target memory by forming forward temporal associations between properties of the target and the intervening objects. Because of these associations, the retrieval of target properties following presentation of the spatial probe is accompanied by the retrieval of unwanted intervening object properties. This “fuzz” or non-specific retrieval of object properties accounts for the reduced memory accuracy in Exp 2. Given a fuzzy retrieval set, the 9AFC task used in Exp 2 makes it more difficult to exclude response alternatives.