Implicit learning for probable changes in a visual change detection task

Abstract

Previous research demonstrates that implicitly learned probability information can guide visual attention. We examined whether the probability of an object changing can be implicitly learned and then used to improve change detection performance. In a series of six experiments, participants completed 120–130 training change detection trials. In four of the experiments the object that changed color was the same shape (trained shape) on every trial. Participants were not explicitly aware of this change probability manipulation and change detection performance was not improved for the trained shape versus untrained shapes. In two of the experiments, the object that changed color was always in the same general location (trained location). Although participants were not explicitly aware of the change probability, implicit knowledge of it did improve change detection performance in the trained location. These results indicate that improved change detection performance through implicitly learned change probability occurs for location but not shape.

Introduction

Change blindness, the failure to detect a change in the visual world from one moment to the next, demonstrates that the amount of visual information that can be attended and then retained in visual short-term memory (VTSM) is limited (see Simons & Rensink, 2005 for review). In a change detection task, it may be useful to attend to information that has a high probability of changing from one moment to the next. Beck, Angelone, and Levin (2004) demonstrated that probable changes (e.g., a lamp changing from being turned on to being turned off) are detected more frequently than improbable changes (e.g., a green lamp changing into a blue lamp; see also Beck, Peterson, & Angelone, 2007). However, participants were not explicitly aware that probable changes are preferentially detected (Beck et al., 2004), suggesting that information about the probability of objects changing affects change detection performance implicitly. The current experiments address the question of whether the change probability of a novel shape can be learned implicitly during a change detection task and then used to improve change detection performance.

Participants learn and use many kinds of regularities to improve performance in many tasks, including tasks involving strategy selection (Lemaire & Reder, 1999), auditory information (Aslin et al., 1998, Creel et al., 2004, Newport and Aslin, 2004, Saffran et al., 1996), the spatial layout of visual objects (Chun and Jiang, 1998, Fiser and Aslin, 2001), and the temporal relationships among visual objects (Fiser and Aslin, 2002, Kirkham et al., 2002, Olson and Chun, 2001). For example, participants modified their strategies for solving math problems according to base rates of success although they were unaware of these strategy modifications (Lemaire & Reder, 1999). In addition, the phenomenon of contextual cueing occurs when participants performing a visual search task are able to find the target faster when it is in distracter arrays that are repeated (Chun & Jiang, 1998). This type of learning is referred to as implicit learning because it occurs without the intent to learn and participants are not explicitly aware of the learned information.

Implicit learning can guide visual attention and thereby influence what information is stored in VSTM. For example, when a cue predicts the location of the target, participants are faster to respond to the target (Lambert, Naikar, McLachlan, & Aitken, 1999). Furthermore, when a target is more likely to appear on one side of the screen than the other, eye movements go to the predicted side of the screen more readily than the unpredicted side of the screen (Walthew & Gilchrist, 2006). Attentional cueing by probability information has been shown to lead to decreased processing of distractors (Awh, Matsukura, & Serences, 2003). Therefore, the target to which attention is drawn is more likely to be stored in VSTM and the distractors are less likely to interfere with this representation. This suggests that if probability information could be learned and used to guide attention in a change detection task, change detection would improve for the probability consistent target.

Research has suggested that probability information can be learned and used implicitly in a change detection task (Olson, Jiang, & Moore, 2005). In this task, the spatial configuration of several boxes predicted the location of a box that was deleted in the post-change array. Participants learned the spatial configuration of the arrays and were able to use that information to give priority to the spatial location that was predicted to change. This study demonstrated the use of a learned association between a spatial configuration and a particular location to guide attention and VSTM. The current studies assess the degree to which an association between the shape of objects and the likelihood they will change can lead participants to attend and create VSTM representations sufficient to detect changes.

Research on implicit learning suggests that participants should be able to learn an association between the shape of an item and the probability of that shape changing color and use this information to improve color change detection. Implicitly learning that a particular shape is more likely to change color than other shapes would require the ability to implicitly learn temporal information because change detection involves monitoring objects over time. It would also require the ability to implicitly learn information about the shape of an attended object. Finally, it would require that implicit learning occur automatically for attended information. Previous research suggests that all of these conditions are met and therefore, participants should implicitly learn the probability that a particular shape will change, leading to improved change detection performance.

Improved change detection performance by implicit knowledge of change probability information would require knowledge about the stability of an object across time. Implicit learning of temporal relationships has been demonstrated in participants’ ability to implicitly learn a predictable relationship among the order in which different shapes are grouped across time (Fiser and Aslin, 2002, Turk-Browne et al., 2005). Participants watched a sequence of shapes traveling across the computer screen with a repeated pattern of co-occurring shape triplets. Although participants were not told that a pattern would occur, after watching the sequence, they rated the repeated triplets as more familiar than novel triplets, despite not being aware that patterns had repeated. Furthermore, in a shape identification task, participants responded faster to items in a repeated triplet than to items in a novel triplet (Turk-Browne et al., 2005). Given that implicit learning for the temporal relationships among novel shapes does occur, implicit learning could also occur for novel shapes in a change detection task.

Information about the shape of an object can be learned and implicitly used to enhance visual search performance. Beck et al., 2007, Chun and Jiang, 1998 paired a set of distractors with a particular target. The target and distractor sets were repeated across several trials, but the location of the target and distractors was randomly determined on each repetition. Search was faster when the target appeared with the same distractor set as on previous trials even though all the items were in new positions. Therefore, the shapes of the distractors were used to improve search time for the target. Evidence for implicit learning of temporal and shape information among arrays of novel objects suggests that implicit learning could also occur for novel objects that are likely to change over time.

Successful change detection and improved performance on a task through implicit knowledge both require attention. Implicit knowledge is gained automatically (Fiser & Aslin, 2002), but improves performance on a task only when the relevant information is attended (Jiang and Chun, 2001, Jiang and Leung, 2005, Jimenez and Mendez, 1999, Turk-Browne et al., 2005). When asked to find a white target in a display containing white and black items, contextual cueing was only found for repeated configurations in the attended color (the white items; Jiang & Leung, 2005). However, when the repeated configurations in the unattended color were presented in the attended color, search was facilitated immediately. These results indicate that implicit learning occurred for the unattended color, but was not used until the configuration was presented in the attended color. Therefore, in order to use implicit knowledge to improve performance, the information relevant to the implicit knowledge must be attended.

Accurate change detection requires attention to the pre- and post-change information (Hollingworth and Henderson, 2002, Levin and Simons, 1997, Rensink et al., 1997, Simons and Levin, 1998a, Simons and Levin, 1998b). According to feature integration theory when an object is attended the features of the object are bound together into an object file in memory (Kahneman et al., 1992, Treisman and Gelade, 1980). Therefore, if a color change to an object is detected, the object was attended and the features of the object (shape and color) were bound together. However, contrary to feature integration theory, it has been proposed that color and shape are stored independently and that attention can be selectively directed to one but not the other (Garner, 1974). If attention to an object’s color leads to a bound representation of all of the objects features, then detection of a color change implies that the object’s shape was also represented in VSTM. It follows that detecting a color change should lead to implicit learning of an association between the shape of an object and its probability of changing color.

In the current studies, we examined whether change probability information could be learned during a change detection task and then lead to improved change detection performance. In Experiments 1–3 and 5, participants completed a set of training trials in which the same shape (trained shape) changed color on every trial. Training trials were followed by a set of test trials in which the trained shaped changed on half of the trials and untrained shapes changed on the other half of the trials. Performance on the test trials was examined to see whether changes to the trained shape were detected at a higher rate than changes to the untrained shapes. If participants learned during the training phase that a given shape would always be the shape that changed color, then the load on VSTM could be lessened by only attending to that shape. Therefore, on the test trials, performance for the trained shape should be higher than performance for the untrained shapes.