What is being masked in Object Substitution Masking?

Journal of Experimental Psychology: Human Perception and Performance 2006, Vol. 32, No. 6, 1422–1435 Copyright 2006 by the American Psychological Association 0096-1523/06/$12.00 DOI: 10.1037/0096-1523.32.6.1422 What Is Being Masked in Object Substitution Masking? Angus Gellatly and Michael Pilling Geoff Cole The Open University University of Durham Paul Skarratt University of Hull Object substitution masking (OSM) is said to occur when a perceptual object is hypothesized that is mismatched by subsequent sensory evidence, leading to a new hypothesized object being substituted for the first. For example, when a brief target is accompanied by a longer lasting display of nonoverlapping mask elements, reporting of target features may be impaired. J. T. Enns and V. Di Lollo (2000) considered it an outstanding question whether OSM masks some or all aspects of a target. The authors report three experiments demonstrating that OSM can selectively affect target features. Participants may be able to detect a target while being unable to report other aspects of it or to report the color but not the orientation of a target (or vice versa). We discuss these findings in relation to two other visual phenomena. Keywords: object substitution masking, visual attention Visual masking is said to occur when two stimulus displays are presented in close spatial and temporal contiguity and the visibility of one of them (the target) is reduced by the presence of the other (the mask). The target is flashed only briefly, but the mask display may be presented for a shorter or longer period depending on the requirements of the particular study. Over the years, several supposedly distinct forms of masking have been proposed. In a recent influential article, Enns and Di Lollo (1997) reported what they claimed to be a new form of visual masking, which they termed object substitution masking (OSM). They (Di Lollo, Enns, & Rensink, 2000; Enns, 2004) have contrasted OSM, supposedly involving substitution of one perceptual object by another, with what Enns (2004) has called object formation masking (OFM). The latter refers to masking that supposedly results from interference with the perceptual formation process involved in segmenting the target from the camouflage of background and other nearby objects. The term OFM subsumes much of what has previously been referred to as integration, interruption, or metacontrast masking, although Di Lollo et al. (2000) and Enns (2004) suggested that demonstrations of these purported categories of masking may often have included components of both OFM and OSM. It is not our intention in this article to review the huge literature on visual masking or to debate fine details of taxonomy and nomenclature in relation to it. We adopt the terminological dichotomy of OFM and OSM on pragmatic grounds because it serves our present purpose. In recent articles dealing with OSM (e.g., Di Lollo et al., 2000; Enns, 2004; Enns & Di Lollo, 1997; Kahan and Mathis, 2002) a usage has developed in which new labels are sometimes used in place of terms with a longer history in the literature on masking. Again, we do not wish to argue the superiority of one nomenclature over the other. We aim to use both sets of terms interchangeably and in such a manner as to be understood equally by those accustomed to the more traditional terminology and those familiar with usage in recent articles dealing with OSM. According to Di Lollo et al. (2000), OFM is sensitive to factors such as contour abutment and overlap and relative luminances of target and mask displays. It also depends critically on the exact timing of target and mask onsets. When studied as integration or metacontrast masking, OFM typically peaks at a target–mask stimulus onset asynchrony (SOA) of 50 ms or less and is largely absent at SOAs of 100 ms or more (see Enns, 2004). OFM is also little affected by manipulations of spatial attention toward or away from the target. OSM, by contrast, is highly sensitive to attentional manipulations but not to the local spatiotemporal contour interactions thought to give rise to OFM. A standard demonstration of OSM uses what Kahan and Mathis (2002) have called the briefly masked control method, comparing two conditions in which target and mask onset simultaneously (common onset). In the briefly masked control (or “no masking” control) condition, they also offset simultaneously. In the mask condition, the (temporally trailing) mask remains present after target offset. In the earlier literature, these were frequently referred to as simultaneous-offset and delayed-offset conditions. Reporting of some target feature is markedly reduced in the second condition relative to the first. Angus Gellatly and Michael Pilling, Department of Psychology, The Open University, Milton Keynes, United Kingdom; Geoff Cole, Department of Psychology, University of Durham, Durham, United Kingdom; Paul Skarratt, Department of Psychology, University of Hull, Hull, United Kingdom. Michael Pilling is now at the MRC Institute for Hearing Research, University of Nottingham, Nottingham, United Kingdom. This research was supported by Economic and Social Research Council Grant R000223824. Correspondence concerning this article should be addressed to Angus Gellatly, Department of Psychology, The Open University, Walton Hall, Milton Keynes MK7 6AA, United Kingdom. E-mail: a.gellatly@open.ac.uk 1422 OBJECT SUBSTITUTION MASKING Because spatial and temporal contour relationships at onset are identical, the degree of OFM is usually thought to be equal in both conditions, and the reduction in target visibility is, therefore, taken as a measure of OSM. (An alternative interpretation would be that the greater time-integrated energy of the trailing mask simply produces a greater degree of OFM than does the simultaneousoffset mask. In our introduction and discussion of Experiment 2, we present evidence and arguments against this interpretation.) The theory of OSM (Di Lollo et al., 2000) assumes that perception arises from continuous and recurrent communication between neurons at lower and higher levels within the visual system. Newly appearing objects stimulate lower level cells with spatially local receptive fields and geometrically simple stimulus requirements. In a feed-forward sweep, output from these cells activates higher level neurons, which have larger receptive fields and are tuned to more complex stimulus properties. Competing pattern hypotheses are generated at this higher level. Resolution of competition between these hypotheses, as well as binding of patterns to precise spatial locations, is thought to require feedback sweeps. Activations at higher and lower levels are compared for consistency, and there may be some number of iterations of forward and backward sweeps before a stable percept emerges. If the visual scene remains constant over the iterations required to achieve dynamic stability, the new object will be consciously perceived. However, if a mismatch is detected between activation at the different levels, the iterative process will begin again only on the basis of current sensory input. OSM is said to occur as a result of such a mismatch. Onset of target and mask sets up lower level activation leading to the hypothesis of target plus mask. If both offset simultaneously before the arrival of the feedback sweep, this hypothesis can still be matched to persisting but fading activity at the lower level. However, if the mask display continues after offset of the target, the hypothesis will mismatch strong sensory evidence that there is now only a mask present. Further iterations result in only the mask being consciously perceived. Perception of the target plus mask will have been substituted by perception of the mask alone. Spatial attention is thought to modulate the masking effect because if attention is already focused at the appropriate location, conscious perception of the target will be achieved with fewer iterations than if it is focused elsewhere or is diffuse. In the same vein, studies by Neill, Hutchinson, and Graves (2002) and Tata and Giaschi (2004) have shown that the extent of OSM is modulated by the power of the mask to capture attention away from the target. Supposedly, OSM takes place after object formation, at the level of object representation (Enns, 2004; Lleras & Moore, 2003; Moore & Lleras, 2005). But what does this mean? In one example of OSM, a diamond target missing either a left or right corner is briefly presented (usually 33 ms or less), and the observer has to report the side of the missing corner. Under a variety of conditions, accuracy of report can be reduced by a mask of just four dots surrounding but not abutting the target. For example, in Experiment 1 reported by Enns and Di Lollo (1997), there were three horizontally aligned locations at which the briefly presented target might appear. After a variable SOA, a four-dot mask appeared for 30 ms at the same location as the target or at a different location. Accuracy of reporting the missing corner of the target was greatly reduced in the former condition relative to the latter. The same outcome was obtained by Kahan and Mathis (2002, Experiment 1), 1423 when the target diamond appeared unpredictably in one of four quadrant positions and a simultaneous-onset four-dot mask with delayed-offset appeared in the same or in a different quadrant. Similarly, in their Experiment 3, Enns and Di Lollo found that accuracy of report was reduced for displays containing two distractor diamonds in addition to the masked target as compared with displays containing only a masked target. We are interested in the following question: When, in studies such as those just cited, observers are unable to report the side of the missing corner of the diamond, what is it that is being masked? For example, are observers unaware that a target was presented? Given that detection performance is generally a more sensitive measure than discrimination performance, it seems unlikely that this need be the case. However, the proposal that OSM occurs at the level of object tokens has about it the suggestion that all representation of the target object is erased from perception. Although we are unaware of published data that address the issue, statements by several authors come very close to such a claim. For example, in relation to four-dot masking of a diamond, Kahan and Mathis (2002) stated that “the phenomenological experience of this effect is that a mask surface replaces the target” (p. 1249). Neill et al. (2002) reported for four-dot masking of a letter that “not only does the space inside the dots appear blank, but there is a strong subjective impression of the contours of a square connecting the 关masking兴 dots” (p. 683). For masking of variously oriented C shapes, Di Lollo et al. (2000) stated, Although the four-dot mask was insufficient as a source of contour interference, it was entirely adequate for defining a trailing configuration (a square surface) that replaced the target as an object of perception . . . .At longer durations of the trailing mask . . . the four dots appeared to be clearly visible, but the target location appeared empty. (p. 492) While for similar displays, Tata (2002) reported that by contrast with previous metacontrast masking effects in which “visibility of the target is reduced, but its presence is nevertheless detected,” in his studies masking “was phenomenologically complete . . . the observers reported seeing a blank space among the distractors where the target should have been” (p. 1036). Clearly, Di Lollo et al. and Tata were open to the possibility that OSM may eliminate all trace of the target representation. In contrast to these strong if introspectively based claims, Enns and Di Lollo (2000, p. 351) took a more cautious position and considered it an outstanding question whether OSM (which in the context they termed common-onset masking) interferes with the perception of some or all aspects of a target. They wrote, For example, many iterative cycles might be required to perceive specific attributes of the target such as its detailed shape or colour. Simpler attributes such as mere presence or absence might require fewer cycles, in which case masking for these attributes would be reduced or eliminated. (p. 351) The first aim of the present article is to provide data to address this outstanding question identified by Enns and Di Lollo. Before we proceed further, it is important to emphasize the logic of inquiry in relation to this matter. Suppose the phenomenological claims reported above were supported by behavioral evidence that observers could not discriminate between presence and absence of the target. What would this demonstrate? Consider that Di Lollo et GELLATLY, PILLING, COLE, AND SKARRATT 1424 al. (2000) used displays with up to 15 distractors and thus had 16 potential targets and target locations. It is possible that in conditions of such high perceptual load, target detection might indeed be at the level of chance. This would not prove that OSM necessarily involves elimination of all trace of the target. It would simply show that there happen to be conditions in which detection performance is reduced to chance and that also produce OSM. However, proving the obverse case, that OSM for some feature of a target can occur without all trace of the target having been erased from conscious perception, requires only a single counterdemonstration. If a substantial degree of OSM can be obtained for a particular target feature in conditions that produce a much smaller decrement in detection performance and if the level of detection is higher than the level of discrimination, then there must be some trials on which the target is detectable while the feature is not discriminable. In Experiment 1 we had the aim of investigating this issue. Our results may be summarized as the following: Experiment 1 showed that strong four-dot masking can occur for a discrimination task under conditions in which reporting of presence or absence is much less affected. This raised the further question of whether, in OSM, properties such as color and orientation can be independently masked. The results of Experiment 2 indicate that masking of different properties is at least partially independent, and those of Experiment 3 reveal that it can be fully independent. Theoretical implications of this independent processing are discussed. Experiment 1: Detection and Discrimination Under OSM This study was closely modeled on Enns and Di Lollo’s (1997) Experiment 3. Participants reported either which corner of a masked diamond was missing (discrimination) or whether a target had been present or absent at the masked location (detection). Method Participants. Ten postgraduate students and employees at the Open University served as paid participants. All had normal or corrected-tonormal vision. Materials and procedure. Stimuli were presented on a PC monitor. Following a warning tone, a trial began with presentation of two blue bars, 2° above and below the center of the screen, between which participants were instructed to fixate throughout the trial. After 400 ms, three diamonds were presented for 17 or 33 ms, each with either the left or right corner deleted at random. One diamond, the target, was surrounded by four simultaneously appearing dots (except on half of the detection trials, on which the two distractor diamonds appeared with the dots surrounding an empty location). Mask dots either offset simultaneously with target and distractors (unmasked or simultaneous-offset condition) or remained for 500 ms (trailing mask or delayed-offset condition). For discrimination, the missing corner of the target was to be indicated with the left or right slash key. For detection, the same keys were used to report target presence or absence, with this response mapping and the order of the two tasks counterbalanced across participants. Target and mask (or mask alone) appeared randomly and equally often in each of the three locations. Each task comprised demonstration trials (with extended frame durations), followed by 32 practice trials and four blocks of 48 experimental trials. The whole session lasted approximately 40 min. Stimuli were presented on a PC controlled by custom software and were viewed from 70 cms. Target and distractors were white diamonds (all monitor color guns at 63) on black (all monitor color guns at 0); they were 0.9° on a side with a corner missing (triangular section 0.25°on its equal sides). Masking dots were squares 0.4° on a side, forming a virtual square of 2°. Target and distractors appeared at the center and 3° to left and right of center. Results Accuracy for both tasks is shown in Table 1. False positive errors occurred on less than 1% of target-absent detection trials. An initial analysis of variance (ANOVA) on the accuracy data showed target duration did not produce a significant main effect or interactions with other factors (F ⬍ 1.5 in all cases), therefore the data were subsequently collapsed across durations. A two-way repeated-measures ANOVA showed significant main effects of task, F(1, 9) ⫽ 63.3, p ⬍ .001, and masking, F(1, 9) ⫽ 23.5, p ⬍ .001, and a significant interaction between these two factors, F(1, 9) ⫽ 14.5, p ⬍ .005, reflecting the larger effect of masking on discrimination performance than on detection performance. Post hoc one-tailed tests showed, however, that masking had a significant effect on both discrimination, t(9) ⫽ 5.04, p ⬍ .005, and detection, t(9) ⫽ 2.12, p ⬍ .05. Discussion Targets that masking has reduced in visibility or even rendered phenomenally absent may serve as effectively as unmasked targets to elicit implicit measures of detection such as response time to display onset (Fehrer & Biederman, 1962; Fehrer & Raab, 1962) or two-alternative forced choice discrimination (Schiller & Smith, 1966). Present or absent responses are thought to more closely reflect phenomenal experience, although see Jacoby (1998) for a discussion of the relative contributions of conscious and unconscious processes to performance on explicit performance tests. Nevertheless, the present results indicate that four-dot masking can reduce explicit detection performance even for fairly low-load visual displays. The effect is quite small in the present experiment, but it is significant. On this basis, it is possible that in different circumstances, especially with higher load displays such as those used by Di Lollo et al. (2000), the effect might be much larger, Table 1 Percentage Mean Correct Responses and Standard Deviations on Present/Absent Detection Task and Left/Right Discrimination Task in Experiment 1 Unmasked Present–absent detection Left–right discrimination Masked Masking effect % Correct SD % Correct SD % Correct SD 93.8 81.2 2.3 7.2 89.8 65.1 6.8 12.0 4.0 16.1 5.9 10.1 OBJECT SUBSTITUTION MASKING with detection performance possibly even down at chance level (as implied in the quotations cited earlier). Be that as it may, however, in Experiment 1 OSM had a much greater effect on discrimination than on detection, indicating that on some proportion of trials the target must have been detectable while the missing corner was not discriminable. The answer to the question posed by Enns and Di Lollo (2000) is that OSM can interfere to a greater extent with one aspect of a target than with another. Contrary to one reading of the introspective reports cited earlier, OSM need not be an all-or-none affair; it does not have to entail complete substitution of the conscious representation of the target (Di Lollo et al., 2000). So exactly what aspects of the target do get masked in OSM? A striking feature of OSM is that aspects of a target can be obscured by masking elements or even a single element (Lleras & Moore, 2003) very different from it in shape and location. Typically, however, OSM experiments use target and mask stimuli that differ in various uncontrolled ways and are poorly matched psychophysically. The diamond targets and masking dots used in Experiment 1 and in several previous studies of OSM (e.g., Enns & Di Lollo, 1997; Kahan & Mathis, 2002) were characteristic in this respect. The dots were actually small squares; target and mask objects differed greatly in size and by 45° in their major orientation (although the truncated diamonds contain a vertical edge also). That such differently shaped and spatially distant mask elements could impair perception of the target was, of course, precisely what made the OSM demonstration so impressive. At the same time, however, it left open the question of whether the intensity of OSM may be determined by the extent of the physical differences between target and mask objects. Although there have been careful parametric studies in relation to OSM of target–mask SOA, target–mask separation, and number of distractors (e.g., Di Lollo et al., 2000; Enns, 2004), we are unaware of any studies of OSM that have systematically varied feature differences between target and mask elements (though see Moore & Lleras, 2005). In our next two experiments we used target and mask stimuli that differed in a controlled manner on two physical dimensions. As in Experiment 1, a target item was surrounded by four masking items, but all these were identically shaped rectangles. Target and mask bars could differ in color, orientation, both, or neither. In Experiments 2 and 3, we addressed the question of whether discrimination of target color and target orientation can be independently subject to OSM. One way of putting this is to ask whether what gets substituted is the representation of an integrated object or of a bundle of stimulus features that remain somewhat unbound and independent (Wolfe & Cave, 1999). If OSM occurs at the level of object representations, it might be expected that features would already be bound together and so not subject to independent OSM. We defer full consideration of such theoretical issues until the data have been reported. Experiment 2: OSM for Reporting Color or Orientation of Target The task for Experiments 2 and 3 is illustrated in Figure 1. Participants fixated between two blue bars presented for 500 ms and reported either the color or orientation (Experiment 2), or both (Experiment 3), of a target bar. Target and distractors occurred centrally and to the left and right. The 17-ms target display 1425 Delayed offset Till Response(s) 500 ms Simultaneous offset 17 ms 500 ms Till Response(s) 17 ms 500 ms Figure 1. Presentation sequence for the control mask (simultaneous offset) and trailing mask (delayed offset) conditions of Experiments 2 and 3. contained either the target and two distractor bars or just two distractors. Target location was identified by four horizontal masking bars surrounding one of the three potential targets’ locations. There were three masking conditions. The mask display could onset and offset simultaneously with the target, the simultaneousoffset, or (briefly masked) control condition. Or the mask display could onset simultaneously with the target but remain present for 500 ms, the delayed-offset, or trailing mask condition. Or, finally, the mask could onset at a 100 ms SOA following target onset and remain present for 500 ms, the delayed-onset condition. Participants pressed the left slash key (for Orange or Horizontal or the right slash key (for Red or Vertical) to indicate either the color or the orientation of the target, or the space bar if they either did not know the color or orientation of the target or thought that no target had been presented. They were instructed not to guess the color or orientation of the target if they were uncertain but to press the space bar in such cases. Responses of this kind are referred to as omission responses as opposed to correct responses (reporting the correct color or orientation) and error responses (reporting the incorrect color or orientation). The wording of these instructions and the inclusion of targetabsent trials require explanation. Two-alternative forced choice decisions are commonly used in psychophysical work because they have the advantage that, in principle, data interpretation need not take response bias into account. The use of target-absent trials, as well as the instruction to report color or orientation only when fairly confident of them, potentially introduces issues of response bias to the interpretation of our data. However, we deliberately chose this method rather than a two-alternative forced choice decision precisely to avoid the danger of a particular response bias. Our participants were required to report either the color or the orientation of a target bar fleetingly presented at the center of four clearly visible horizontal bars that themselves had one of the two 1426 GELLATLY, PILLING, COLE, AND SKARRATT values of color and orientation that the target could take. During a run of repetitive experimental trials, participants reporting nearthreshold experiences might sometimes unconsciously respond with the color (orientation) of the mask bars rather than with what they thought might be the color (orientation) of the target. Alternatively, participants might very consciously adopt a strategy of responding with the color (orientation) of the mask when they were uncertain about the color (orientation) of the target precisely because they hypothesized that target color (orientation) was more discernible when different from that of the mask. In other words, a possible strategy would be, “when you don’t see the color (orientation) of the target, respond with the value of the mask.” Inclusion of target-absent trials and instructions not to guess when uncertain was intended to discourage use of this strategy.1 This seems to have worked, so much so that there were even signs of an opposite strategy having been adopted by some participants in Experiment 3. It also allowed us to assess whether participants distinguished between absence and presence of a masked target, as they had in Experiment 1. If the false positive rate was low, then target-absent trials were not being mistaken for target-present trials. Note, though, that because of the nature of our instructions, false negative responses— omission responses on target present trials— could not be interpreted as failures to detect the target. The participant may have detected that “something was there” but not known its color or orientation. Indeed, the results of Experiment 1 and the very low false positive rates we report for Experiments 2 and 3 strongly suggest that this was usually the case when an omission response was made, suggesting that in these experiments OSM occurred at the feature level. Moreover, we believe that the robust pattern of data across Experiments 2 and 3 justifies our decision on this matter of method. Mask objects in Experiments 2 and 3 were narrowly separated from the target, and some target contour was paralleled by mask contour. In order to draw any conclusions about OSM, one must demonstrate that the results obtained do not simply reflect OFM. One way to do this is to use the briefly masked control method used in Experiment 1, which involves comparing a simultaneousoffset mask with a delayed-offset (trailing) mask. Because both conditions involve simultaneous onset of target and mask the extent of OFM is usually thought to be equal in both cases, so—as described earlier—any difference in performance on the two conditions is then attributed to OSM alone. Alternatively, OSM is sometimes thought to be demonstrated by comparing a simultaneous-onset and delayed-offset (trailing) mask with the same duration mask presented at an SOA of (in our case) 100 ms, a delayed-onset mask. As just described, the former condition is thought potentially to give rise to both OFM, because of simultaneous onset of mask and target, and OSM, because of the delayed offset of the mask elements (though see also below). In the delayed-onset condition, by contrast, this interpretation predicts that there should be little if any OFM because with a 100 ms SOA between target onset and mask onset, OFM will be greatly attenuated if not absent (Spencer & Shuntich, 1970), leaving only OSM caused by the trailing mask elements. In Experiment 2 we used both control methods to ensure that the effects we reported were indeed examples of OSM, not OFM. Note that we did not attempt to establish whether OFM actually occurred in any of the conditions of Experiment 2. One way to have done this would have been to vary the separation between target and mask elements. If separation had no effect on the extent of masking, it could have been concluded that only OSM and not OFM was affecting performance (Enns & Di Lollo, 1997, Experiment 2). However, a problem arises with this method of distinguishing between OSM and OFM if target–mask separation does prove to have an effect on performance because there is no means of measuring the relative contribution of each type of masking. Accordingly, instead of adopting this procedure, we followed the logic of common-onset masking (Di Lollo et al., 2000; Enns & Di Lollo, 2000). As just described, the usual argument here is that if OFM has occurred, its strength should be equal for simultaneousoffset and delayed-offset masks because both involve an identical simultaneous onset. As noted earlier, however, it could be argued that the greater time-integrated energy of the trailing mask simply causes more OFM than does the simultaneous-offset mask. Our use of a delayed-onset mask with the same energy as the trailing mask should allow us to distinguish between these competing interpretations. Because OFM is known to decrease with target– mask SOA and to be almost absent at SOAs at or beyond 100 ms (Spencer & Shuntich, 1970), then on either account the delayedonset mask should cause less OFM than the simultaneous-onset trailing mask. If the two types of mask cause similar degrees of masking, this indicates that both are causing OSM rather than OFM. To summarize, the logic of the experiments was as follows. Target and mask could have the same orientation and color, could differ on color or orientation, or could differ on both features. If OSM occurs after object features have been bound into an integrated representation, then the extent of masking in any condition should be equal for reporting of either feature. This follows because if the representation of the target (plus mask) has been substituted by a representation of the mask alone, then access to a record of either target feature should be equally impossible. By contrast, if OSM occurs prior to the binding of features into an integrated representation, then it presumably occurs at the level of individual features. Therefore, when target and mask differ on a single feature, reporting of that feature should evidence less masking than reporting of the feature that is the same for target and mask elements. This is because the signal-to-noise ratio is greater for the former than for the latter. By the same reasoning, because reporting of a feature should reflect only the signal-to-noise ratio for that feature, accuracy of reporting should be unaffected by 1 One method of avoiding the problem might have been to borrow techniques used by Mounts and Melara (1999) and to use two pairs of similar colors such as red or orange and green or blue. A red or orange target, for example, could have been surrounded by either red and orange mask elements (similar condition) or by green and blue mask elements (dissimilar condition). In this case, participants would not have been able to utilize a strategy of simply responding with either the mask color or its opposite. However, pilot testing and the results of Experiments 2 and 3 show how difficult it is to hold performance on the present task below ceiling even when the color difference between target and mask elements in the dissimilar color condition is very small (i.e., red vs. orange). This rules out use of the method just described. Following the same logic, Mounts and Melara also used pairs of orientations that were close to vertical or close to horizontal. But this technique was also inapplicable to the present studies because of ceiling and floor effects and the impossibility of matching for contour proximity across similar and dissimilar conditions. OBJECT SUBSTITUTION MASKING whether there is a match or mismatch between target and mask on the other feature. The extent of masking should vary across conditions. Reporting of color should be affected only by match or mismatch on color, and reporting of orientation should be affected only by match or mismatch on orientation. 1427 Participants. Twenty University of Keele undergraduates with normal or corrected-to-normal vision served as participants in the experiment in partial fulfillment of a course requirement, half reporting target color and half target orientation. Equipment and stimuli. Stimuli were presented as for Experiment 1 and viewed from 70 cm. Following a warning tone, two horizontal blue fixation lines appeared. After 500 ms they were joined by the target display, containing two or three bars at the three potential target locations, at the center of the screen and 3° to left and right. Target, distractor, and mask bars were 1° ⫻ 0.2° and colored red (45, 0, 0) or orange (45, 28, 0). They were photometrically matched for on-screen luminance. Simultaneously with target onset (control and trailing mask conditions) or 100 ms later (delayed mask condition), one location was surrounded by four red or orange horizontal masking bars, centered 0.5° above or below and 0.8° left or right of the location center. After 17 ms, the target (if any) and distractor bars offset. The mask bars offset either simultaneously (control) or after being present for 500 ms (trailing mask and delayed mask). The target location contained a target bar on two thirds of trials. Unmasked locations always contained horizontal distractors the same color as the mask bars. When a target was present at the masked location, it was equally often the same color and orientation as the mask (SC–SO), different color and same orientation (DC–SO), same color and different orientation (SC–DO), or different on both features (DC–DO). Procedure and design. We explained the task by using demonstration trials with prolonged frame durations. Participants were informed that response speed was unimportant and that their sole aim was to be accurate in color or orientation decisions and response key selection. They were also told (a) that they should not be reluctant to press the space bar to indicate either “target absent” or that they were uncertain as to the color or orientation of the target, and (b) that on one third of trials no target would be presented. Central fixation throughout a trial was emphasized. Participants completed a practice block of 40 trials making color or orientation judgments followed by 18 experimental blocks of 60 randomly ordered trials of the same judgment. There were 60 target-present and 30 targetabsent trials per combination of target type (SC/SO, DC/SO, SC/DO, DC/DO) and mask condition (control/trailing/delayed). “don’t know”) and that the percentage of omissions was effectively the inverse of the percentage of correct responses. The latter figures are shown in Panel A of Figure 2 for reporting color and Panel B for reporting orientation. For the simultaneous-offset control condition, reporting of both dimensions was highly accurate for all target types. For the delayed-offset (trailing) mask and delayed-onset mask conditions, data patterns and absolute accuracy levels were highly similar. In both, reporting of target color was greatly improved by a target–mask difference in color and to a lesser extent by a difference in orientation. Similarly, for both conditions, reporting of target orientation was greatly improved by a target–mask difference in orientation and to some extent by a difference in color. To compare the delayed-offset (trailing) mask and delayedonset mask conditions, we conducted a 2 ⫻ 2 ⫻ 4 mixed ANOVA with related variables of mask type (delayed-offset/delayed-onset) and target type (SC/SO, DC/SO, SC/DO, DC/DO) and an unrelated variable of reported dimension (color/orientation). Mask type had a nonsignificant effect and did not enter into any significant interactions, F ⬍ 1 in all cases. We do not comment on other aspects of this analysis because they recurred in the analyses that follow and are considered in the next paragraph. Also, because these two sets of data were indistinguishable, only one of them was used in further analysis. We calculated masking scores by subtracting for each participant his or her scores in the delayed-offset (trailing) mask condition from the corresponding score in the simultaneous-offset control condition. These data were entered into separate 2 ⫻ 2 within-participant ANOVAs for the report color and report orientation groups, with factors of target color (same/different) and orientation (same/different). For the report color group, there were significant main effects of target color, F(1, 9) ⫽ 36.72, p ⬍ .001, and orientation, F(1, 9) ⫽ 23.67, p ⬍ .001. Target color had a slightly larger effect when target and mask orientation were the same rather than different, but this interaction effect was nonsignificant, F(1, 9) ⫽ 3.36, p ⬍ .1. For reporting orientation, there were main effects of target color, F(1, 9) ⫽ 26.81, p ⬍ .001, and orientation, F(1, 9) ⫽ 23.09, p ⬍ .001, and a significant interaction, F(1, 9) ⫽ 10.44, p ⬍ .01, because target orientation had a larger effect when target and mask color were the same rather than different. Results Discussion The false positive rate was very low, with participants correctly pressing the space bar on 99% of target-absent trials. Responses on target-present trials were either correct, omissions, or errors. Errors— giving the wrong color or orientation of a presented target— occurred on only 1.1% of target-present trials, indicating that participants were able to follow the instruction not to guess when uncertain as to target color or orientation. A 2 ⫻ 3 ⫻ 4 ANOVA on the error data with factors of reported dimension (color/orientation), mask condition (control/trailing/delayed-onset) and target type (SC/SO, DC/SO, SC/DO, DC/DO) gave no significant effects (F ⬍ 2, p ⬎ .15 in all cases). Because error rates did not differ across condition, guessing corrections were not applied, and we reported the percentage of correct responses (out of 60) for each condition. The low error rate also means that the vast majority of noncorrect responses were, therefore, omissions (“no target” or Experiment 2 was designed to provide two checks on whether the pattern of data obtained with the delayed-offset (trailing) mask reflected OSM, OFM, or a combination of the two. The first involved comparing the simultaneous-offset and delayed-offset conditions. Figure 2 shows that these produced very different patterns of data, suggesting that the pattern for the delayed-offset condition was due to OSM rather than to OFM. However, the strength of this conclusion could be open to question. There is a potential problem because performance with the simultaneousoffset mask was at or close to ceiling for all conditions. It is possible that if performance with this control mask had been sufficiently below ceiling, then the same pattern of results might have been obtained as with the delayed-offset (trailing) mask, which would have suggested that in both cases OFM was at work. This would be consistent with the interpretation that says that the Method GELLATLY, PILLING, COLE, AND SKARRATT 1428 Report Colour Simultaneous Onset Control Delayed Offset (Trailing) Mask Delayed Onset Mask Percentage correct score 100 A 80 60 40 20 0 Target Colour: Same Different Same Different Target Orient.: Same Same Different Different Report Orientation Percentage correct score 100 Simultaneous Onset Control Delayed Offset (Trailing) Mask Delayed Onset Mask B 80 60 40 20 0 Target Colour: Same Different Same Different Target Orient.: Same Same Different Different Figure 2. Percentage of correct responses for report color (Panel A) and report orientation (Panel B) by target type and masking condition in Experiment 2. Error bars denote the standard error of the mean. OBJECT SUBSTITUTION MASKING more intense masking with the trailing mask was due to its greater energy causing a higher degree of OFM. However, evidence against this comes from the second check built into the experiment. Results for the delayed-offset (trailing) mask and the delayedonset (also trailing) mask are indistinguishable. If the former had its effect largely by means of OFM, then much smaller masking effects should have been caused by the latter. That this was not the case indicates that the identical effects seen in both these conditions were the result of OSM, not OFM. In the two trailing mask conditions (delayed-offset and delayedonset), a target or mask difference on either color or orientation reduced masking more for that feature than for the other feature. This indicates a degree of dimensional independence in OSM, which in turn implies that what was being masked was not an integrated representation of the target object, an object token. Conversely, the fact that for reporting of both color and orientation, masking was also significantly reduced by a target or mask difference on the other dimension suggests that independence of dimensional processing may have been less than complete. The data from Experiment 2 are, then, somewhat ambiguous with respect to the predictions outlined earlier. Once again, however, account has to be taken of the possible role of ceiling effects. Performance in reporting of either dimension was close to ceiling for all target types in the simultaneous-offset control condition (and for some target types—particularly DC–DO—in the delayedoffset and delayed-onset conditions). By obscuring possible differences between target–mask conditions, these high levels of performance may have distorted the calculated masking scores. Our main aim in Experiment 3 was to test the same predictions as described for Experiment 2 under conditions in which ceiling level performance could be avoided. 1429 Equipment and stimuli. Equipment and stimuli were as in Experiment 2, except the bar stimuli were one third of their previous length and distances between locations were halved. Procedure and design. The procedure and design were as in Experiment 2, except for the changes already indicated, and there were 48 rather than 60 target-present trials per condition. As previously, one third of trials contained no target. Half of participants reported color before orientation, half the reverse. Results To this end, we made target signals for color and orientation smaller in Experiment 3 by reducing the size of the stimulus elements. Also, because the delayed-offset mask and the delayedonset mask had produced indistinguishable results in Experiment 2, only the former condition was included in Experiment 3. Finally, instead of reporting either color or orientation of the target, participants in Experiment 3 were required to report both features with equal priority. We made this change because the ambiguous evidence of dimensional selectivity of OSM observed in Experiment 2 might have resulted from participants selectively attending to their to-be-reported dimension. Such a top-down attentional set might have biased participants against forming an integrated representation of the target object, resulting in data that reflected only a contingent and partial dimensional independence of masking. Requiring participants to report both dimensions was intended to ensure that they attended equally to both features and that, to the extent this might be subject to top-down control, they were biased to form an integrated representation of the target from which both features could be read off. The false positive rate was again very low, with participants correctly pressing the space bar on 98.5% of target-absent trials. For target-present trials, responses were either correct, omissions, or errors. As in Experiment 2, the large majority of noncorrect responses were omissions. However, errors (reporting the wrong color or orientation of a presented target) averaged 3% for the simultaneous-offset control mask and 10% for the delayed-offset (trailing) mask, being higher for color reports of same-color targets than of different-color targets and higher for orientation reports of same-orientation targets than of different-orientation targets. Because errors clearly were not evenly distributed across conditions, guessing corrections were applied to individual participant data for each combination of mask type and target type. For reporting color, errors on SC/SO were subtracted from number correct on DC/SO and vice versa, and similarly for SC/DO and DC/DO. For reporting orientation, errors on SC/SO were subtracted from number correct on SC/DO and vice versa, and similarly for DC/SO and DC/DO. The guessing-corrected scores are shown in Figure 3. The correction procedure reduced effect sizes but did not alter the data pattern. Accuracy for reporting color in the simultaneous-offset control condition remained close to ceiling for all target types, even with smaller stimuli and a requirement to report both target dimensions. However, accuracy for reporting orientation in the control condition was lower in all conditions. Data of the control conditions were entered into a two-way related ANOVA with factors of reported dimension (color/orientation) and target type (SC–SO, DC–SO, SC–DO, DC–DO). There was a main effect of reported dimension, F(1, 9) ⫽ 15.58, p ⬍ .01, and also of target type, F(3, 27) ⫽ 5.34, p ⬍ .01 but, it is important to note, no interaction, F(3, 27) ⫽ 1.80. For the delayed-offset (trailing) mask condition, performance for reporting both features was well below ceiling and, as can be seen from Figure 3, target type appeared to have differential effects on reporting of color and orientation. Masking scores were calculated as in Experiment 2, by subtracting guessing-corrected delayed-offset scores from guessing-corrected simultaneous-offset scores (See Figure 4). For reporting of each feature, a two-way ANOVA was computed with factors of target color (same/different) and target orientation (same/different). For reporting color, there was a main effect of target color, F(1, 9) ⫽ 6.13, p ⬍ .05, but no effect of target orientation and no interaction, F ⬍ 1 in both cases. For reporting orientation, there was a main effect of target orientation, F(1, 9) ⫽ 14.13, p ⬍ .005, but no effect of target color and no interaction (F ⬍ 1.3 in both cases). Method Discussion Participants. There were 10 new participants from the Open University, as described for Experiment 1. Comparing Figures 2 and 3 demonstrates that there are clear similarities in the pattern of results obtained in Experiments 2 and Experiment 3: OSM for Reporting Color and Orientation of Target 1430 GELLATLY, PILLING, COLE, AND SKARRATT Figure 3. Percentage guessing-corrected scores for reporting target color and orientation by target type and mask condition in Experiment 3. Error bars denote the standard error of the mean. 3. In both, a color difference was more important for reporting color and an orientation difference more important for reporting orientation. But whereas the results of Experiment 2 are ambiguous as to whether masking is wholly dimension specific, those of Experiment 3 are rather more clear-cut. With performance in the delayed-offset (trailing) mask condition well below ceiling, the ability to report target color was affected only by a target-mask color difference and not by an orientation difference. For reporting target orientation, accuracy is affected mainly by a target–mask orientation difference, although there is also some sign that a color difference helped. However, the same slight benefit of a color difference to reporting orientation is also seen for the simultaneous-offset control condition. Presumably, this arose because color was the more salient stimulus dimension for these stimuli, as shown by performance still being very close to ceiling for reporting of color in the control condition. However, once masking scores are calculated (see Figure 4), the dimensional specificity of OSM becomes apparent. As borne out by statistical analysis, and unlike in Experiment 2, OSM for reporting of each feature was affected only by a target–mask difference on that feature and not by a difference on the other feature. It might be objected that because reporting of color in the simultaneous-offset control condition remained close to ceiling for all target–mask combinations, therefore strong conclusions should no more be drawn from the results of Experiment 3 than from those of Experiment 2. This objection supposes that ceiling effects are obscuring differences between target–mask combinations in the report color control condition that, were they evident, would alter calculated masking scores in such a way as to remove evidence for dimensional independence of OSM. This would be the case only if they were obscuring a benefit to reporting color or a difference in orientation. A number of counterarguments can be mounted against such a possibility. First, if ceiling effects were distorting the data of the report color control condition when there were clearly no such ceiling effects for report orientation, it might well be expected that there would be an interaction between feature reported and target–mask combination. Yet the analysis of control data showed no interaction. Performance in the control conditions was slightly higher for different color targets than for same color targets, regardless of which feature was being reported. This is wholly consistent with color being the more salient feature, as evidenced by greater accuracy of report for color. Secondly, as already noted, there were no ceiling effects for reporting orientation and yet the evidence for feature independent OSM is as strong in this case as it is for reporting color. This is all the more striking in the light of the greater salience of target color. Given the imbalance in saliency, it might well have been expected that the more salient color difference signal would affect orientation judgments even if the reverse were not the case (Nothdurft, 2000). That such an asymmetry of effect was not observed suggests the independent processing of different feature dimensions was robust. Of course, it is possible that with a sufficiently large imbalance in feature saliency a difference on a more salient feature would affect reporting of a less salient feature. However, in Experiment 3 the significant difference in saliency did not lead to such an asymmetry of effect on reporting target features. Along the same lines, the fact that participants had to attend to and report both features of the target might have been expected to bias them OBJECT SUBSTITUTION MASKING 1431 Simultaneous minus Delayed score Report Colour Report Orientation 60 40 20 0 Target Colour: Same Different Same Different Target Orient.: Same Same Different Different Figure 4. Size of masking effect by target type for reporting target color and target orientation in Experiment 3. Error bars denote the standard error of the mean. against independent processing of features. That it did not do so once again suggests a robust independence of feature processing. General Discussion Experiment 1 demonstrated that OSM of a target feature can occur even though the presence of the target has been detected. (i.e., if a representation of the target had been established, but masking prevented perception of its features). Experiments 2 and 3 examined whether the degree of OSM is a function of target– mask feature similarity and whether target features become available for report conjointly or independently. The data show that target–mask similarity on a particular dimension affected reporting of that dimension only, indicating that target features are independently processed. In other words, consistent with Experiment 1, OSM was occurring at the level of features rather than of integrated objects. In the remainder of this section we first address some possible objections to the conclusions we have drawn. We then relate the present findings about OSM to two other visual phenomena and consider various theoretical implications of all three phenomena. OSM That the extent of visual masking can be modulated by target– mask similarity on such features as color, shape, and spatial frequency has been shown several times previously (e.g., Breitmeyer, 1984; Growney & Weisstein, 1972; Uttal, 1970; Yellott & Wandell, 1976). The present results confirm these earlier findings but, when viewed in the context of a distinction between OFM and OSM, also extend them in two important ways. First, whereas the earlier studies were not designed to indicate whether target–mask similarity effects are associated with OFM, OSM, or both, the present experiments have shown that they can definitely be associated with at least OSM. Our conclusion that the extent of OSM is a function of target–mask similarity (or dissimilarity) on feature dimensions might appear to be challenged by two aspects of our own data. Although detection performance in Experiment 1 was less affected by four-dot masking than was discrimination, there was such an effect on discrimination. Because target and mask elements differed in many respects, this could be taken to show that OSM occurs even when target and mask seem not to share features. In fact, however, target and mask elements in Experiment 1 did share at least some features. They were the same color, which Experiments 2 and 3, as well as a recent study by Moore and Lleras (2005), showed to be an important dimension of similarity. They also both contained right angles, though whether this would be important we do not know. Indeed, our starting point for these studies was precisely that little was known as to what, if any, dimensions of similarity would be important for OSM. We have shown that color and orientation are important dimensions but there could well be others such as degree of overlap of spatial frequency content, closure, or goodness of figure. A second aspect of our data that might also be thought to challenge our interpretation comes from Experiment 3 in which, for both reporting color and reporting orientation, OSM was observed even when target and mask differed on both dimensions. It could be argued that given these feature differences, OSM should not have occurred. However, it should be noted that in Experiment 2 there was 1432 GELLATLY, PILLING, COLE, AND SKARRATT no masking of targets that differed on both dimensions. Indeed, it was in order to avoid such ceiling effects that the color and orientation feature differences in Experiment 3 were deliberately reduced in comparison with those of Experiment 2. Our overall results indicate that the extent of OSM is a function of the degree of target–mask similarity along various feature dimensions, of which color and orientation are two. Varying target–mask separation (or overlap, or signal-to-noise ratio) on one of these feature dimensions affects reporting of that dimension only and not reporting of the other. But for the stimuli of Experiment 3, the feature values deliberately varied by such small amounts that even targets differing on both dimensions were subject to some degree of OSM. That feature-specific OSM has been demonstrated with the present displays does not preclude the possibility that, with attention spread across larger and busier displays including more eccentric targets, OSM may be obtainable with target–mask combinations differing as much as possible on as many features as possible. In addition to feature-level OSM, there may also be object-level OSM (Lleras & Moore, 2003; Moore & Lleras, 2005; Treisman & Kanwisher, 1998), a topic to which we return later in this article. The second way in which our results go beyond previous studies of target–mask similarity is in demonstrating that under conditions of OSM, the effect of similarity is a function of what target feature is to be reported. That is, target–mask similarity cannot be specified simply in terms of the physical features of the stimulus configuration but must also be defined with respect to what the observer is reporting.2 We go on now to consider two other visual phenomena that have been shown to exhibit this kind of taskspecific tuning. Feature Specificity of Pop-Out and Sparse Representation Mounts and Melara (1999) asked participants to report the color or orientation of a target that preattentively popped out of a 48-item array from which it differed in either color or orientation. All items in the array were subject to interruption masking. Participants were better able to report the color than the orientation of a color pop-out target and better able to report the orientation than the color of an orientation pop-out target. Furthermore, and consistent with the results of Experiment 3, this was the case even when both features had to be reported on every trial. Mounts and Melara attributed the effect to the fact that in their experiments the target was more discriminable from distractors on the pop-out dimension than on the non-pop-out dimension. Unlike in the present study, the design of Mounts and Melara’s experiments did not distinguish between OFM and OSM nor allow an evaluation of whether featural processing was wholly or only partially independent. Mounts and Melara interpreted their findings as evidence against object-based theories of attention (e.g., Driver & Baylis, 1989; Duncan, 1984; Duncan & Humphreys, 1992; Egly, Driver, & Rafal, 1994), according to which objects are selected in an all-or-none fashion, so knowing an object’s color should entail knowing equally its orientation and vice versa. Mounts and Melara (1999) concluded that, whether linkages among features are set up in terms of a common spatial location or of a common object token, attentional selection is coordinated at the level of features. They saw their results as consistent with more recent object-based models of attention that allow for partially independent processing of object features (Duncan, 1996; Logan, 1996, 2004). The present data, especially from Experiment 3, push this line of reasoning still further by evidencing wholly independent processing of different features of the same object. But if object features are processed completely independently, then in what sense can selection in these conditions be said to be object based? One approach to the issue is that of sparse representation, a view advocated by various authors over the years (e.g., Hochberg, 1984; MacKay, 1973; O’Regan, 1992; Rensink, 2000). According to this view, there is no rich and detailed internal representation of the visual world. Representation of a particular object or area will be only as detailed as it need be for the task at hand. We initially posed the question “Is what gets substituted in OSM the representation of an integrated object or of a bundle of stimulus features that remain somewhat unbound and independent?” On the sparse representation view, the question itself is faulty because it takes for granted that, given adequate presentation conditions, visual representations inevitably get rich and detailed independently of the task at hand.3 The alternative is that we never see whole objects but only those aspects of an object relevant to what we need to do. A frequently cited analogy is with the perception of an object held in the hand (MacKay, 1973; O’Regan, 1992). The experience is of a complete object although it is one based on a fragmentary representation, with sensory input restricted to just those object parts actually in contact with the skin (plus haptic information). The experience of visual scenes and complete visual objects is similarly derived from partial input and representation. According to Rensink (2000) and others, a seen object seems to appear before us as real and fully complete only because any particular property can be made explicit as required simply by interrogating the external world. In Rensink’s coherence theory (CT), preattention yields proto-objects, which can be surprisingly detailed—for instance, including three-dimensional organization— but are coherent only over small spatial and temporal extents, having constantly to be regenerated. Focused attention is a process by which one (or a few) of these proto-objects acquires a high degree of coherence such that it can retain its identity across brief interruptions, and its various features and their interrelationships can be experienced as required. CT has much in common with Treisman’s (1988) feature integration theory (FIT). For example, in CT attention selects a protoobject, which is transformed into a coherence field, or visual object (Rensink, 2000). In FIT, attention selects “loosely organized feature bundles (Wolfe & Cave, 1999, p. 17) the features of which are then glued together to make an integrated visual object. So far, so similar. How the two approaches seem to differ is that in CT the features of the attended object are represented only as required, whereas in FIT all the features of the attended object are automatically represented, by which it is meant that they are bound together. How many objects may be so richly represented at one time depends in FIT on the overall perceptual and cognitive load (Lavie, 2005; Treisman, 1995). Rensink (2000) was somewhat ambivalent as to whether there is rich representation of an attended object, first saying we may “instead of simultaneously representing 2 We thank Jim Enns for pointing out this way of thinking about our results. 3 We thank Jim Enns for pointing this out to us. OBJECT SUBSTITUTION MASKING in detail all of the objects in our surroundings, represent only those objects – and only those particular properties of those objects – needed for the task at hand” (p. 1475, italics added) but then later on more ambiguously suggesting, “An interesting possibility in this regard is the binding problem may be illusory – it may be that the properties of only one object at a time are ever bound together” (p. 1484). A very strong version of the sparse representation view has been defended by Enns and Austen (2003, in press), who have argued that not all features of even a single attended object are necessarily represented perceptually. Consistent with this view, Droll, Hayhoe, Triesch, and Sullivan (2005) reported evidence that object features may be represented only while they remain relevant to action selection, and Rafal, Danziger, Giordano, Machado, & Ward (2005) argued that explicit object representation is at the level needed to select task relevant actions. Our results for Experiment 3, in which both features had to be reported, together with those of Mounts and Melara (1999), show that representation of task-relevant features is constrained independently by the quality of the data (signal:noise ratio) on each feature. Nevertheless, however sparsely or not features may be represented, they require to be linked to an object representation of some kind. We turn now to consideration of token and type representations and their possible involvement in OSM. Repetition Blindness and Type–Token binding Another visual phenomenon shown to exhibit task-specific tuning is repetition blindness (RB). Kanwisher, Driver, and Machado (1995) found that spatial RB is modulated by selective attention to stimulus dimensions. Their participants were briefly presented with two simultaneous characters (C1 and C2), each followed by an interruption mask; each character could be one of three letters and one of three colors. Participants reported the identity or color of first the left (C1) then the right (C2) character, responding “no character” when they thought a location had been empty. Reporting of the relevant feature (identity or color) was markedly less accurate when its value was repeated across C1 and C2 than when it was not repeated but was unaffected by repetition of the nonrelevant feature. For example, if both characters were red, reporting of their colors was less accurate than if they were different colors, regardless of the identities of the two letters, whereas if both letters were Es, reporting of their identities was less accurate than if they were different, regardless of the colors of the two letters. For reporting either feature there was RB for the relevant but not for the irrelevant stimulus dimension, and this was shown to be unrelated to unwillingness to repeat a response. As with the present results, variation on an irrelevant dimension did not affect accuracy of reporting the relevant dimension. Kanwisher (1987; Kanwisher et al., 1995) has argued that RB arises because although repeated characters (or colors) are usually appropriately recognized (matched to stored types), they are less likely than unrepeated characters to be individuated as distinct perceptual tokens. This failure of type–token binding supposedly applies equally to temporal or spatial repetition of items. Consideration of the type–token distinction raises the question of how it may apply to the phenomenon of OSM. The present masking displays were similar in many ways to the displays used by Kanwisher et al. (1995) to demonstrate spatial RB. Consider those conditions in which target and mask elements onset simulta- 1433 neously (i.e., all conditions of all three experiments other than the delayed mask condition of Experiment 2). The location of the mask indicated which of the three potential targets had to be reported on, so it was necessary to attend first to the mask elements and then to the target itself, which was similar to Kanwisher’s participants having to attend first to C1 then to C2. Just as C2 might repeat one, both, or neither of the features of C1, so targets in our Experiments 2 and 3 repeated one, both, or neither of the features of the mask. And in both cases, repetition on a dimension reduced accuracy of report for that dimension—more RB in the Kanwisher et al. study, more OSM in our experiments—whereas repetition on the other dimension had no effect. Perhaps the two performance deficits reflect a common processing failure. In the RB literature, in which there has been an emphasis on similarity between C1 and C2, this has been taken to be a failure to individuate separate occurrences of a type. In the OSM literature, in which— especially in relation to four-dot masking—the emphasis has been on dissimilarity between mask elements and the target, the failure has been taken to be in maintaining a representation (of some sort) of the target. In OSM terminology, cells early in the system initially code for features, with cells later in the system synthesizing candidate object descriptions. However, drawing parallels between OSM and RB reminds us that there is a distinction between type and token representations (see also Lleras & Moore, 2003; Moore & Lleras, 2005). Just as the present experiments included no-target trials, the experiments of Kanwisher et al. (1995) included trials in which either C1 or C2 was omitted. As in our experiments, participants were well able to distinguish target absence from a target glimpsed indistinctly. Kanwisher et al. stressed that their results “reflect a failure in binding the appropriate identity (type) to distinct object representations (tokens) when a type is repeated, rather than a complete failure to set up distinct tokens” (pp. 329 –330). In the terms of OSM theory, it may be that early visual cells not only process an object’s features but also cause an object token representation to be set up on the master map of locations (Treisman, 1988). Later cells would then synthesize a candidate type representation, which reentrant processing would link (or not) to the appropriate object token on the master map. Our findings with reporting of both features, along with those of Mounts and Melara (1999), imply that under impoverished conditions different features need not be equally well bound into the type representation nor, consequently, into the token representation. (Indeed, for visual objects as structurally and semantically pared down as brief colored bars, perhaps the type representation is no more than a loose conjunction of more or less well-represented features.) The essential idea behind the concept of OSM is that it is a form of masking that takes place at the level of object representation (Di Lollo et al., 2000; Enns, 2004; Enns & Di Lollo, 1997). The present demonstration of feature-specific OSM indicates that what is masked need not be an integrated object representation, but we have already conjectured that, dependent on experimental conditions, there could be different forms of OSM involving either feature-specific or object-level representations. Lleras and Moore (2003; Moore & Lleras, 2005) have presented evidence for OSM involving object-level representations, favoring the term object token in their first study and object file in the second. Moore and Lleras showed that OSM was reduced when mask and target were different colors and when, through the use of apparent motion, the GELLATLY, PILLING, COLE, AND SKARRATT 1434 mask appeared to slide across the target, so only incidentally surrounding it briefly. Their interpretation was that “susceptibility to OSM is determined by the extent to which separate object representations can be established for the target and mask prior to mask offset; when separate object representations can be established, target information associated with that representation can be protected from OSM” (p. 1178). Although this is certainly an interesting argument, we note that the reductions in OSM could also be due to target and mask having differed on the features of color and motion or stationarity, respectively. Although the idea of separate object-level and feature-level OSM effects is an intriguing possibility, empirically distinguishing between levels is likely to pose considerable challenges (Scholl, 2001). Although it is important to try to theoretically relate different visual phenomena to each other, it would be unwise at this stage to force the comparison between OSM and RB too far. It may be that they are similar yet distinct phenomena. Only empirical investigation will reveal whether, for example, one can be obtained under conditions in which the other cannot or whether it may be possible to obtain additive effects of the two. These are issues we are currently pursuing. Summary The present findings suggest that the theory of OSM needs to be expanded to take account of the distinction between type and token representations. OSM can occur independently for separate target features (Experiments 2 & 3) and has similarities to RB in that both phenomena can be thought of as failures of type–token binding. However, there may be examples of OSM involving masking of object tokens or object files, as well as examples involving failure to bind feature and/or type representations appropriately to a token representation. Possibly, the answer to our original questions may be that OSM can operate at more than one level. Certainly, it can operate at the level of independent features. References Breitmeyer, B. G. (1984). Visual masking: An integrative approach. New York: Oxford University Press. Di Lollo, V., Enns, J. T., & Rensink, R. A. (2000). Competition for consciousness among visual events: The psychophysics of reentrant visual processes. Journal of Experimental Psychology: General, 129, 481–507. Driver, J., & Baylis, G. C. (1989). Movement and visual attention: The spotlight metaphor breaks down. Journal of Experimental Psychology: Human Perception and Performance, 15, 448 – 456. Droll, J. A., Hayhoe, M. M., Triesch, J., & Sullivan, B. T. (2005). Task demands control acquisition and storage of visual information. Journal of Experimental Psychology: Human Perception and Performance, 31, 1416 –1438. Duncan, J. (1984). Selective attention and organization of visual information. Journal of Experimental Psychology: General, 113, 501–517. Duncan, J. (1996). Cooperating brain systems in selective perception and action. In T. Inui & J. L. McClelland (Eds.), Attention and performance XVI: Information integration in perception and communication (pp. 549 –578). Cambridge, MA: MIT Press. Duncan, J., & Humphreys, G. W. (1992). Beyond the search surface: Visual search and attentional engagement. Journal of Experimental Psychology: Human Perception and Performance, 18, 578 –588. Egly, R., Driver, J., & Rafal, R. D. (1994). Shifting visual attention between objects and locations: Evidence from normal and parietal lesion subjects. Journal of Experimental Psychology: General, 123, 161–177. Enns, J. T. (2004). Object substitution and its relation to other forms of visual masking. Vision Research, 44, 1321–1331. Enns, J. T., & Austen, E. (2003). Change detection in an attended face depends on the expectations of the observer. Journal of Vision, 3, 64 –74. Enns, J. T., & Austen, E. (in press) Mental schemata and the limits of perception. In Peterson, M. A., Gillam, B., & Sedgwick, H. A. (Eds.), In the mind’s eye: Julian Hochberg on the perception of pictures, film, and the world. New York: Oxford University Press. Enns, J. T., & Di Lollo, V. (1997). Object substitution: A new form of masking in unattended visual locations. Psychological Science, 8, 135– 139. Enns, J. T., & Di Lollo, V. (2000). What’s new in visual masking? Trends in Cognitive Science, 4, 345–352. Fehrer, E., & Biederman, I. (1962). A comparison of reaction time and verbal report in the detection of masked stimuli. Journal of Experimental Psychology, 64, 126 –130. Fehrer, E., & Raab, D. (1962). Reaction time to stimuli masked by metacontrast. Journal of Experimental Psychology, 63, 143–147. Growney, R., & Weisstein, N. (1972). Spatial characteristic of metacontrast. Journal of the Optical Society of America, 62, 690 – 696. Hochberg, J. (1984). Form perception: Experience and explanations. In P. C. Dodwell & T. Caelli (Eds.) Figural synthesis (pp. 1–30). Hillsdale, N. J.: Erlbaum. Jacoby, L. L. (1998). Invariance in automatic influences on memory: Toward a user’s guide for the process dissociation procedure. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 3–26. Kahan, T. A., & Mathis, K. M. (2002). Gestalt grouping and common onset masking. Perception & Psychophysics, 64, 1248 –1259. Kanwisher, N. (1987). Repetition blindness: Type recognition without token individuation. Cognition, 27, 117–143. Kanwisher, N., Driver, J., & Machado, L. (1995). Spatial repetition blindness is modulated by selective attention to colour or shape. Cognitive Psychology, 29, 303–337. Lavie, N. (2005). Distracted and confused?: Selective attention under load. Trends in Cognitive Science, 9, 75– 82. Lleras, A., & Moore, C. M. (2003). When the target becomes a mask: Using apparent motion to isolate the object component of objectsubstitution masking. Journal of Experimental Psychology: Human Perception and Performance, 29, 106 –120. Logan, G. D. (1996). The CODE theory of visual attention: An integration of space- based and object-based attention. Psychological Review, 103, 603– 649. Logan, G. D. (2004). Cumulative progress in formal theories of attention. Annual Review of Psychology, 55, 207–234. MacKay, D. M. (1973). Visual stability and voluntary eye movements. In R. Jung (Ed.), Handbook of sensory physiology (Vol. VII/3A, pp. 307– 331). Berlin: Springer. Moore, C. M., & Lleras, A. (2005). On the role of object representations in object substitution masking. Journal of Experimental Psychology: Human Perception and Performance, 31, 1171–1180. Mounts, J. R. W., & Melara, R. D. (1999). Attentional selection of objects or features: Evidence from a modified search task. Perception & Psychophysics, 61, 322–341. Neill, W. T., Hutchinson, K. A., & Graves, D. F. (2002). Masking by object substitution: Dissociation of masking and cuing effects. Journal of Experimental Psychology: Human Perception and Performance, 28, 682– 694. Nothdurft, H.-C. (2000). Salience from feature contrast: Additivity across dimensions. Vision Research, 40, 1183–1202. O’Regan, J. K. (1992). Solving the “real” mysteries of visual perception: The world as an outside memory. Canadian Journal of Psychology, 46, 461– 488. OBJECT SUBSTITUTION MASKING Rafal, R., Danziger, S., Giordano, G. Machado, L., & Ward, R. (2002). Visual detection is gated by attending for action: Evidence from hemispatial neglect. Proceedings of the National Academy of Sciences, 99, 16371–16375. Rensink, R. A. (2000). Seeing, sensing and scrutinizing. Vision Research, 40, 1469 –1487. Schiller, P. H., & Smith, M. C. (1966). Detection in metacontrast. Journal of Experimental Psychology, 71, 32–39. Scholl, B. J. (2001). Objects and attention: The state of the art. Cognition, 80, 1– 46. Spencer, T. J., & Shuntich, R. (1970). Evidence for an interruption theory of backward masking. Journal of Experimental Psychology, 85, 198 – 203. Tata, M. S. (2002). Attend to it now or lose it forever: Selective attention, metacontrast masking and object substitution. Perception & Psychophysics, 64, 1028 –1038. Tata, M. S., & Giaschi, D. E. (2004). Warning: Attending to a mask may be hazardous to your perception. Psychonomic Bulletin & Review, 11, 262–268. 1435 Treisman, A. M. (1988). Features and objects: The fourteenth Bartlett Memorial lecture. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 40(A), 201–237. Treisman, A. M. (1995). Modularity and attention: Is the binding problem real? Visual Cognition, 2, 303–311. Treisman, A. M., & Kanwisher, N. G. (1998). Perceiving visually presented objects: Recognition, awareness, and modularity. Current Opinion in Neurobiology, 8, 218 –226. Uttal, W. R. (1970). On the physiological basis of masking with dotted noise. Perception & Psychophysics, 7, 321–327. Wolfe, J. M., & Cave, K. R. (1999). The psychophysical evidence for a binding problem in human vision. Neuron, 24, 11–17. Yellott, J. L., & Wandell, B. A. (1976). Colour properties of the contrast flash effect: Monoptic vs dichoptic comparisons. Vision Research, 16, 1275–1280. Received August 17, 2005 Revision received April 19, 2006 Accepted April 19, 2006 䡲