Given that backward masking techniques play such a prominent role in our discussions of the difference between conscious and unconscious processing, then it is important that we are clear that we fully understand just what is going on. So far we have been reasonably rigorous in setting out the conditions under which masking will and will not be effective. For instance, traditionally it was accepted that iconic memory is abolished by backward pattern masking (Coltheart, 1972; Gegenfurtner & Sperling, 1993). This fits with the view that, unless items from the target display are read out from iconic memory before the masker arrives, they will be lost. That is, the items will not be perceived – they will not enter consciousness. Given that such important claims have been made, then it is fairly critical that the supporting evidence is unequivocal.
For instance, the traditional view of the effective- ness of backward masking led to the assumption that the experimenter had a rather strict control of the contents of consciousness – roughly, only four items will be read out from the icon and it is only these that are consciously perceived. However, we can contrast the idea of masking by replacement with that of the notion of interruption by curtailment. To reiterate: by the replacement view, information about the current target is completely over-written by information about the next stimulus (i.e., the mask). According to the curtailment view, however, it is assumed that the processing of the first stimulus simply stops when the next arrives, and it is not necessary to suppose that masking eradicates all trace of the target. Curtailment
allows for the possibility that some record of the target may still be available even though processing has moved on. Accordingly, the idea is that iconic repres- entation may survive following the presentation of a backward visual-noise mask. Given this possibility it then becomes no easy matter to try to control the duration of iconic storage or indeed be confident that masking blocks or erases events from consciousness. Indeed recently, evidence that the icon may survive backward visual-noise masking has been provided by Smithson and Mollon (2006).
What they did was run a variant of the standard Sperling paradigm (see Figure 3.5), but importantly, they did not present the cue until after the presenta- tion of the mask (see Figure 4.13, and Smith & Fabri, 1975, for similar manipulation). Clearly if the mask eradicates iconic representation, then participants should be unable to report cued items correctly, because the cue follows the mask and the mask erases the icon. However, what Smithson and Mollon (2006) showed was the following:
• There was no partial report advantage when the mask was presented at the offset of the target display, so an immediately following mask did eradicate the representation of the target display.
Figure 4.13 Examples of displays used by Smithson and Mollon (2006)
The target display was presented for <10 ms and followed either immediately or after 100 ms by a ‘checkerboard pattern mask’. The mask was presented for <20 ms and then after a variable delay the cue (the bar markers) were presented. Time moves across the page left to right from the start of the trial.
Source: Smithson, H., & Mollon, J. (2006). Do masks terminate
the icon? The Quarterly Journal of Experimental Psychology, 59, 150 –160 (fig. 2, p. 153). Reproduced with permission from Taylor & Francis Ltd.
• There was, however, a robust partial-report advan- tage when the mask followed the offset of the target display by 100 ms, thereby replicating the basic Sperling (1960) effects.
• Most critically, the partial-report advantage occurred even when a delay was introduced between the offset of the mask and the onset of the cue. The effect was large when the cue immediately followed the mask but this declined gradually as the delay between the mask and the cue increased to 700 ms. Indeed the data indicated that the partial-report advantage survived for up to 400 ms after the onset of the mask.
Such a pattern of results clearly stands in contrast to the claim that the mask eradicates the icon in all cases. If the cue follows the mask to up to about 400 ms then the partial-report superiority survives.
Smithson and Mollon (2006) went further and argued that the data bolstered the discrete moment hypothesis (see Chapter 3). What their data suggest is that as visual events unfold over time, there is an internal sequence of snapshots that is generated in much the same way that movies are made up from a sequence of frames. Each visual event is captured in a frame on this kind of internal roll of film. So the sequence of visual events is represented in the visual system in some form of memory trace that preserves the sequential order of the input. By this view the icon survives the mask because it is coded as one discrete perceptual event and because this event is different from the other discrete perceptual event that captures the mask.
In conclusion, what we have here is further evidence in favour of the discrete moment hypothesis together with some rather alarming evidence that shows how backward masking may not act to erase or block items from entering consciousness in the way that has been traditionally accepted. Backward masking may not act according to interruption by replacement in all cases. Clearly such evidence is important to bear in mind if we are attempting to achieve experimental control on the contents of consciousness. We will return to such issues when we address effects of so-called subliminal face processing (see Chapter 16).
objective threshold The point at which perceptual discrimination is at chance.
subjective threshold The point at which a person feels as though they are performing at chance even though, in actual fact, they are performing better than chance.
Evidence has emerged that subliminal stimuli may indeed be processed to a semantic level in the absence of explicit awareness. However, this evidence is only present in very transitory effects in tightly controlled experiments on word recognition. The effects, such as they are, seem only to reveal something about the very earliest stages of word recognition. This is a far cry from the more grandiose claims about how semantic information more generally affects the processes of sensory encoding. So although some aspects of advert- ising, and other attempts at mind control, are based on assumptions about the power of subliminal process- ing, the actual empirical data are far from compelling. Despite the rather mixed evidence that has been reviewed in terms of semantic activation without con- scious identification, the discussion has allowed us to introduce many basic concepts and methodological tools that have much more general application. Signal detection theory crops up time and again in the liter- ature. The distinction between effects of perceptual sensitivity and bias will be revisited as we examine more thoroughly claims that have been made in the literature about the nature of human perception. What should be appreciated, however, is that in many, many cases when cognitive psychologists discuss the differ- ence between conscious and unconscious processing, somewhere lurking in the background is an experiment involving masking. From what we now know about the limitations of this technique we therefore need to be very, very cautious.
CHAPTER SUMMARY
l The chapter begins by drawing out what a sequential account of processing looks like – namely, the stages
of processing are set out in a chain such that this stage can only begin once information from the immedi- ately preceding stage arrives. Turvey’s concurrent and contingent model of processing is described and an important distinction between early peripheral mechanisms responsible for stimulus encoding and later central interpretative mechanisms is set out. The model is based on arguments concerning feedforward pro- cesses in which information flows from the stimulus to the central mechanisms in a bottom-up, stimulus- driven way. In contrast there are feedback processes in which information may flow in the opposite direction.
l An explanation of masking by object substitution was provided by Di Lollo et al. (2000) in terms of
re-entrant processes. In their model of visual processing, a distinction is made between an input layer, a working space layer and a pattern layer. The input layer represents retinal information, which is then passed to the working space layer. The pattern layer makes shape explicit and then passes this information back to the working space layer. Masking by object substitution occurs as a result of over-writing any representa- tion of the target + mask in the working space layer with a representation of the mask alone.
l These principles of masking can be used to address some fairly fundamental questions in cognitive psycho-
logy, including the nature of consciousness. On the basis of the re-entrant model, stimulus identity at the pattern layer might influence the working space layer, and as a result of this feedback, it is possible that semantic activation could occur without the stimulus actually being consciously identified at all. Allport
lexical decision task An experimental paradigm in which participants must judge whether, on a given trial, a letter string is a word or a non-word. Typically performance is carried out under reaction time conditions.
semantic relatedness Typically, the degree to which items are related by virtue of the fact that they share the same category.
Concluding comments
Although many of the basic ideas about establishing a boundary between conscious and non-conscious processing have been traced back to the first half of the twentieth century, it is clear that much remains unresolved and there is still much work to do. We can, perhaps, be confident in asserting that, at the very least, we have a much firmer understanding of the nature of perceptual thresholds than was the case 100 years ago. In addition, much progress has been made methodologically. We are now much clearer empiric- ally about how best to establish a perceptual threshold and what establishing such thresholds means in terms of information processing. Nevertheless, the degree to which subliminal stimuli control our behaviour remains moot.
(1977) presented data to suggest that while participants could not accurately report a target item (a word) that had been masked, the errors that participants made suggested that the semantic meaning of the target had been activated. For example, participants might not have reported seeing the word ‘beach’ but instead responded ‘pebble’. However, it is important in such studies to control for the likelihood that these semant- ically related errors occurred by chance (Williams & Parkin, 1980).
l Perceptual thresholds are critical to the understanding of conscious awareness. The issues were discussed in
the following example. The assumption is that a number of internal feature detectors become active when certain properties are present in the environment. The feature detector might operate according to a simple all-or-nothing rule, in which the detector is on once evidence has accrued past a particular threshold, and is off when evidence is below that threshold. Alternatively, the feature detector might operate with respect to a continuous system, in which the amount of detector firing is directly related to the amount of evidence for that feature. A hybrid model can also be thought of in which both threshold and continuous systems are combined. If stimulus presentation is above perceptual threshold it is said to be supraliminal, and if it is below perceptual threshold it is said to be subliminal. For cognitive psychology, absolute thresholds might be difficult to obtain as a result of stimulus and participant factors. Therefore, experimenters aim to create psychophysical functions that map out the likelihood of stimulus detection over a range of different factors.
l One piece of evidence for semantic activation without conscious awareness is perceptual defence (Postman
et al., 1953). Here, different perceptual thresholds were found for neutral and taboo words, with taboo words requiring greater levels of luminance relative to neutral words. The conclusion was that the meaning of the word was affecting its sensory encoding. While the direction of this effect suggested that threatening stimuli tend to be resisted, contradictory evidence has since been found (Barber & de la Mahotiere, 1982). In addition, it wasn’t clear whether perceptual defence was indeed a perceptual effect or whether it existed at the level of more central decision-making processes (Dixon, 1971).
l Signal detection theory (SDT; see for example Wickens, 2002) has been used to try to adjudicate between
the perceptual and response-based accounts of the effects. Two parameters known as d′ (d prime) and β (beta) can be derived and respectively these have been taken to reflect the operation of the perceptual analysers and mechanisms responsible for making a decision and making a response. In an early study by Broadbent and Gregory (1967a), the evidence was mixed. Low frequency bad words were hard to detect – with the implication that the perceptual analysis for these stimuli is affected by their meaning – whereas participants tended to guess neutral words over good and taboo words.
l Marcel (1983a) compared presence/absence, graphic and semantic judgements for words over a range of
detection thresholds. As the SOA between target and mask increased, the accuracy of judgements also increased, but at different rates. Interestingly, there were SOAs in which participants could perform rela- tively well on the semantic judgements but not very well on the presence/absence judgements. Therefore, under certain conditions of stimulus presentation, semantic activation seemed to be occurring in the absence of conscious identification.
l However, Cheesman and Merikle (1984) pointed out that what is critical in these kinds of studies is the
whether the measure of conscious perceptual experience is valid. Using a Stroop paradigm in which colour words are presented in different kinds of ink colour, they showed that when participants were performing at chance accuracy, neither facilitation with congruent stimuli nor inhibition with incongruent stimuli was observed. In this respect, the distinction between objective and subjective threshold is paramount and the task for future research is to show semantic activation in the absence of conscious awareness at the level of objective threshold.
l One critical assumption of the iconic memory literature is that backward masking erases the icon. That is,
when the stimulus of interest is immediately followed by the presentation of a masker, the target stimulus fails to enter a more durable form of memory and is lost forever. Smithson and Mollon (2006) provided evidence to suggest that this may not be the case. In a variant of the Sperling paradigm in which the cue for partial report was presented after the mask, a partial-report advantage was shown when a delay was introduced between mask offset and cue onset. The data provide evidence for the discrete moment hypo- thesis as well as suggesting that backward masking does not necessarily erase the icon.
ANSWERS TO PINPOINT QUESTIONS
4.1 The concurrent and contingent model is referred to as a sequential account because the model operates according to strictly feedforward prin- ciples. Information produced by the peripheral nets is fed forward to the central nets. First the peripheral nets are engaged by the stimulus, then the central nets are.
4.2 Masking by object substitution works by over- writing any representation of the target in the working space layer with a representation of the mask.
4.3 The errors in the Allport study provided evid- ence for semantic awareness in the absence of conscious identification since, although parti- cipants could not report a particular target, their errors tended to be semantically related to the target.
4.4 The browning wheel on a toaster is a continuous system.
4.5 Perceptual defence is revealed by taboo words requiring a higher perceptual threshold than neutral words.
4.6 d′ is a measure of sensitivity and β is a measure
of response bias.
4.7 Good and bad words are relatively rare. Neutral words are much more common so if you have to guess you are more likely to guess right if you guess with a neutral word.
4.8 Marcel (1983a) found that at certain SOAs, participants could not accurately report the presence or absence of a target, but could report which of two words was more like the (unseen) target in meaning. Therefore semantic informa- tion was independent of conscious identification. 4.9 A traditional Stroop stimulus must be a colour word printed in a certain colour, with ink colour and the colour name being different from one another. Participants are particularly impaired in naming the ink colour when this differs from the colour name.
4.10 The data from Hirshman and Durante (1992) suggest that even though participants were unable to identify the primes correctly, there was semantic activation that influenced the subsequent processing of the target.