The Attentional Blink
back The Attentional Blink: Introduction forward

The Attentional Blink

 

          The term “Attentional Blink” refers to a research paradigm in which subjects are found to be momentarily attentionally blind to a stimulus if presented in close temporal proximity to an initial stimulus. Broadbent and Broadbent (1987)  were among the first to discover this phenomena. Their findings marked the start of an extensive research program into the nature of the attentional deficit. The most widely used method explored in this area of research is the rapid serial visual presentation (RSVP).

The RSVP procedure has become the cornerstone of all attentional blink research. Whereas visual search tasks require the detection of an item in an array of items presented simultaneously at different locations (see Treisman, 19801988 ). The RSVP procedure typically requires the detection of a target item(s) among a stream of stimuli presented in the same location in rapid succession (see Broadbent (1957), Intraub, (1985), Kanwisher (1987), Lawrence (1971), Raymond, Shapiro and Arnell (1992, 1995 1998). Broadbent (1957)  modelled the attentional system suggesting that a filter work’s to suppress the attentional mechanism. The model works like this. In an RSVP task, the attentional system grabs the initial image (target one) and the filter then blocks additional items from reaching recognition centres of the brain. This is done to prevent interference with the initial item, implying that additional items are only processed to a level, which allows for their rejection. Broadbent (1957)  suggested that without the filter working like a gate, closing after the initial stimuli, the attentional system would become blocked with irrelevant data. Therefore once the initial data are being processed, the attentional system effectively goes off line, preventing additional data from reaching consciousness. Treisman (1960) however proposed that the filter suggested by Broadbent  (1957) was not an all or nothing gate, but tended to allow salient information through to be processed to deeper level, and through to consciousness.

Treisman proposed that the filter worked as an attenuator. During a dual-listening task Cherry (1953) found that subjects became aware of specific information in the unattended ear if the data were salient to the subject. For example, if subjects were attending to a piece of prose in the left ear, and the subjects’ names were presented to the right, unattended, ear often the subjects would become aware of their name being presented. This phenomenon was compared to a situation in a noisy cocktail party in which a listener is attending to one conversation but becomes aware of another conversation elsewhere in the room, when their name is spoken aloud. Treisman (1960) suggested that particular words have low thresholds so, as the word enters the attentional system, it is compared with known highly salient words and allowed entry through the filter. This finding has a number of implications. Broadbent  (1957) put their filter mechanism  very early in the processing of sensory information. This model accounts for the idea that information is only processed to a level to which the data can be rejected. Treisman’s (1960) finding suggests that the filter is set very much later in the attentional system, suggesting that information is processed to a relativity deep level. The attenuation model suggests that within the attentional system there must be a catalogue of words with which comparisons are made. For example, as the subjects’ name enters the attentional system, it would be compared with known items. If found to be salient, it passes through the attenuation filter in order to be processed to the level of consciousness. Treisman (1960) does not go as far to suggest that the information is processed to the level of semantics, but implies that a simple like-for-like comparison is made. This model, however, does appear to agree with Broadbent’s (1957),  model in that irrelevant data is rejected before reaching a level of meaning.

 

Shapiro et al (1997) took Treisman’s (1960) model and mapped it on to the RSVP paradigm. In a short series of studies the experimenters examined the idea that the visual system may exhibit the same cocktail party effect observed by Treisman. Using a traditional RSVP design in attentional blink studies, subjects were asked to attend to a number of specific items in a stream of stimuli. In this particular study, the task was to attend to an RSVP stream made up of nouns. The experimental design called for two target items T1 and T2. The subjects were asked to attend to the entire stream and recall whether they has seen both target items. Traditionally, if the subject recalls target one, they have an increased chance of not recalling the second target item if it followed the first in close temporal proximity (within 270 milliseconds). The results indicated that when the subject’s own name was the second target, there was a significant attenuation of the attentional blink. This finding implies that the mechanism underlying the attentional blink cannot be an all or nothing process, which is explicit within Broadbent’s (1957)  gating model. The implication, therefore, is that the underlying gating system proposed by Broadbent (1957). must be more complex. We are still left with the question of whether Treisman’s (1960) attenuation model can account for the loss of visual processing described above. According to Maki et al. (1997), a late selection view of the attentional blink, such as the one suggested by Treisman, would predict that word meaning is available at an early stage of visual processing. Late selection theories suggest that all stimuli are processed to a point of identification before selection (Maki et al . 1997; Shapiro et al. 1994; Duncan and Humphrey, 1989) Maki et al. (1997) designed a series of studies to examine the possibility of word priming during the attentional blink. He based his study on an experiment undertaken by Broadbent and Broadbent (1987)  in which they failed to find any evidence of a priming effect. As with the Broadbent et al, (1987) study two targets were used in a series of RSVP trials, in which the targets were either associated or unrelated words. The subjects were instructed to attend, remember and recall the two differentially coloured target items. The results indicated that when the two targets were associated, recall was significantly better than when they were unrelated. The target priming effect was observed in all serial positions in the RSVP steam. Even though a strong attentional blink was observed, a robust priming effect was also detected. This indicated that identification of the first target seems to aid the identification of the second, related target. Even when the second target was not differentially marked from the rest of the stream, it still primed by the first target. This result, therefore, suggests that Treisman’s (1960) attenuation model cannot explain the attentional blink because, according to her model, irrelevant (non-target) items are rejected before reaching a level of meaning. Maki et al.’s (1997) findings indicate that all items presented during the attentional blink appear to be processed to the level of semantics, but only target one reliably reaches consciousness.

 

Maki et al (1997) suggested that the reason for Broadbent and Broadbent’s (1987) failure to find a priming effect was due to fundamental design errors in the research protocol. Maki  et al. (1997) suggest that they used related and unrelated words that were not linguistically correct for the period. They composed their related words from a list that had been developed some thirty years earlier. Maki et al drew their list from associates developed in the 1990 which were, therefore, more socially meaningful. 

     In the next experiment, Maki et al. (1997) investigated the possibility that the priming effect was dependent on attending to both target items (T1 and T2). Taking a control condition from Raymond et at. (1992), subjects were instructed to ignore T1 and only attend to T2. The results indicated a small but reliable priming effect, which was of around the same magnitude as was seen in experiment one..

 

The results from these studies allow a number of theoretical questions to be answered. The first question to be answered is “does word meaning survive the attentional blink”. The answer to this question appears to be yes, according to Maki et al. (1997). Further results suggest that priming by distractors may be short lived, but when the priming word is a target, the time period can be extended.

 

These results have a number of very important implications, the first of which is that the deficit in attentional processing during RSVP trials, described as the attentional blink, is no longer a correct statement. The term Attentional blink has become an oxymoron. It now seems clear that the attentional mechanism(s) do not shut down upon the receipt of the first target. Raymond et al. (1992) proposed a gating mechanism, in light of the results reported by Maki et al (1997). This mechanism now appears to be implausible. None of the existing models of the attentional blink seem to allow for semantic processing of non-attended items in the RSVP stream. Shapiro and Raymond’s (1995) retrieval competition model suggests, that post-trial retrieval processes influence the reporting of RSVP events. This suggests an element of post-event processing. However, this model does not address the matter of early semantic processing. Chun and Potter (1995) devised an ‘online/late’ model of the attentional blink. The model suggests that processing is online and real-time; in other words the rejection of non-target items is done without the attentional system shutting down and is done as the non-targets reach the semantic level of processing. This model is a ‘late selection’ one because conceptual and associative information results from processing an item before selection. Chun and Potter (1995)  argue that each item is processed and represented in a conceptual short-term memory (CSTM; Potter, 1993). This early stage of processing contains information about the physical characteristics of the item, for example colour, as well as which category the item belongs in. It also contains associative information received from long-term memory. This process, however, is thought to be very short lived, and would not survive interference from subsequent items in an RSVP stream. The model therefore advocates a second stage of processing. Under normal circumstances in an RSVP trial, the subject attends to the first target according to Chun and Potter (1995 ). A “transient attentional response” is triggered; these in turn triggers the second level of processing, which processes the target more completely.   This has the effect of making the representation more durable, thus longer lasting, and less susceptible to interference from subsequent items. All of this processing takes time, consequently processing of the second target is delayed. The representation of T2 in the CSTM is subject to interference from further distractors; this interference has been hypothesised as causing the attentional blink, according to Chun and Potter (1995) . This model is the most plausible, as it accounts for a number of previously unanswered phenomena, such as associative priming. It also provides an explanation of the attentional blink through delayed processing of T2. However, this model does not account for all of the mechanisms that make up the attentional blink phenomenon. It is possible that a hybrid model needs to be created. This model may need to have a mix of late and early processing, as well as allowing for the possibility of some parallel and serial processing and a mix of attentional and mnemonic, and even on and off line, controls. It may not be possible, however, to have a completely accurate model of the attentional blink until more is known about the physiology. This biological approach is still in its infancy and relies specifically on imaging techniques such as functional magnet resonance imaging (fMRI) or electroencephalography (EEG).  

 

The aim of Vogel et al. (1998) study was to use event related potential (ERP) techniques with the attentional blink paradigm in order to examine at which stage of processing the attentional deficit begins to take effect. It was assumed that the attentional blink reflects the suppression of sensory processing.  A number of attentional researchers have turned to event related potentials (ERP) because they provide continuous measurements of processing between a stimulus and a response. Vogel et al (1998)  describe ERP’s as being “scalp recordings that reflect synchronous neuronal activity associated with sensory, motor or cognitive events” (1998: p.1657) Therefore, it is often possible to pinpoint the exact time at which attention begins to influence processing. ERP’s occur as small fluctuations in the EEG recording in response to a given stimulus. Figure one demonstrates that an ERP waveform is composed of both positive and negative voltage fluctuations. Each peak and trough represents a component of the temporal progression of mental processing; each labelled according to the polarity and temporal position. 

           

A number of studies have found that the first two components (N1 & P1) are typically elicited by visual stimuli and reflect sensory processing. Luck et al (1994) found that these components are meditated by spatial attention, and that stimuli presented in a ‘to be ignored’ location elicited smaller P1 and N1 waveforms than stimuli presented in a ‘to be attended to’ location. Vogel et al. (1998) hypothesised that if the same underlying mechanisms are reflective of the attentional blink, then smaller peaks would be elicited to stimuli presented during the blink than stimuli presented outside the attentional deficit.  However, if the attentional blink reflects a late stage of processing then no attenuation in the P1 and N1 components will be observed.

 

Vogel et al. (1998) found that ERP recordings in RSVP tasks can lead to some technical difficulties, which potentially place tight controls on the experimental design.  They found that each item presented during a RSVP trial produces an ERP response that lasts several hundred milliseconds after stimulus onset, which intrudes on the next stimulus in the stream. The overlap makes reading and making sense of the ERP recording extremely difficult. Vogel et al. (1998) eliminated the overlap problem by using an ‘irrelevant-probe technique’, which was developed previously in ERP studies in spatial attention. On a number of trials a white square is flashed behind the probe (second target). The white flash elicits a larger P1 and N1 component than other RSVP items. It is therefore possible using subtraction techniques to isolate the experimental trials.

Experiment One used right-handed, neurologically normal subjects. This first study used a standard RSVP procedure and design. The stimuli were letters displayed in blue against a uniform grey background. T1 was either an odd or even digit also displayed in blue. On half the trials a white flash equal in size and duration to a target item accompanies T2.

The results from this study showed the same pattern as before; subjects are attentionally blind to stimuli that follow T1 in close temporal proximity. However despite the attentional deficit there was no suppression of the P1 and N1 components for T2. This seems to suggest that the attentional blink reflects a relatively late suppression of attention. These results appear to highlight a degree of difference between temporal and spatial attentional studies. As mentioned before, typically ignored spatial items elicit smaller P1 and N1 components. In this experiment no such suppression was experienced. Vogel et al (1998). believed that the absence of the expected suppression might have reflected a lack of power. In re-examining the experimental design, Vogel et al. (1998) believed they had a robust experimental design. They had used twenty subjects, more than twice the size of traditional attentional blink studies. Also, the behavioural accuracy was very similar to previous spatial experiments that had used ERP recordings. The results from experiment one strongly suggest that the attentional blink reflects a post-perceptual suppression of visual processing. This result seemingly sweeps aside the theories that suggest that the attentional blink reflects an early suppression of visual processing. Vogel et al. (1998) felt, however, that the results might reflect a suppression of a relatively late sub-stage of perception. In order to test this idea, Vogel et al. (1998) decided to look at the possibility of measuring wording meaning and semantic context. N400 is a large negative component that peaks approximately 400ms after onset of a mismatch between a word and a previously established semantic context (Vogel et al, 1998; Besson et al., 1992; Kutas and Hillyard, 1980). Vogel et al. (1998) gives the example that a large N400 would be given by the sentence “the women drove to work in her new shiny nose” but a normal N400 would result from last word in the sentence “the women drove to work in her new shiny car”. It has also been established that sequentially presented mismatched words also elicit a large N400, for example ‘pickle and rope’, but not for ‘shoe and foot’. Therefore, if there were no increase in N400 during the period of the attentional blink it would be appear to be strong evidence to suggest that words presented during the blink are processed to the level of semantics. Vogel et al. (1998) suggest that word meaning must be accessed before it can be compared with its semantic context. Therefore, the presence of a large N400 for a mismatched presentation indicates that the word has been identified to the level of meaning. Vogel et al. state that the presence of a normal N400 during the attentional blink appears to indicate that words presented during this period are processed to a very deep semantic level. However, if semantic processing is suppressed for words presented during the attentional blink, the N400 should also be suppressed, relative to words presented outside this period.

 

Vogel et al. (1998) designed the next experiment with T1 as a digit (presented in blue) and T2 as a word of three to seven characters long (presented in red). X’s that created a seven-character target flanked words less than seven characters long. Prior to each trial a context word was displayed which was semantically related or unrelated to the word presented in location T2. The N400 component was measured from a difference wave in related T2 trials and was than subtracted from unrelated T2 trials. Therefore any difference in the N400 component must be related to the semantic relationship between the context word and T2.  

 

The results from this study found no suppression of the N400 component. This strongly suggested that items presented during the attentional blink are identified to the point of meaning. This evidence, therefore, suggests that information is lost after the stimulus is fully identified. Vogel et al. (1998) suggested that information might be lost at the level of working memory.

It is possible however that the difference wave used to measure the N400 component did not reflect this element, but instead reflected a difference at another point. Vogel et al. suggest that because there was a significant difference between the related and unrelated trials this indicates that the word displayed at position T2 was identified to the level of meaning. The experimenters of the study, however, maintain that the difference wave does reflect a difference in the N400 component.   

Vogel et al. (1998) believed that it is possible that the attentional blink produces a modest amount of perceptual degradation that may produce a significant decrease in the N400 amplitude. In the next study, the experimenter’s hypothesised that adding visual noise to the target would lead to a decline in its correct identification and would also lead to a decline in the amplitude of the N400 component. For this experiment only one target was used, in order to assess the perceptual degradation without any additional effects of attentional suppression. Visual noise was added by way of introducing randomly spaced dots in front of the target word with one of three levels of magnitude; dim, medium and bright. The results from this study indicated that, as the noise level increased, both behavioural accuracy and the N400 component were affected. As the noise level increased, the correct identification of the target decreased alongside a decline in the N400 component.

 

The results appear to indicate that the N400 component is a true indicator of perceptual quality during the attentional blink. This, therefore, implies that the evidence gained strongly suggests that the items presented during that attentional blink are fully identified to a level of semantics, and that the attentional deficit is influenced by processes that occur after all items are fully identified. This result, then, begs the question of the stage at which the attentional deficit occurs, if the deficit does not occur at a perceptual level. In the next experiment, Vogel et al. (1998) examined the P3 component, which is thought to reflect working memory processes.  The P3 component is a large, positive wave which typically peaks at 400-600 ms post-stimulus (Vogel et al., 1998, p.1666). The amplitude of this component can be modulated by the frequency of the target category. Vogel et al. give the example that in a study where observers are required to discriminate between two classes of names displayed in a randomised sequence, the least presented category would elicit a much larger P3 component. It has been hypothesised that the P3 component reflects post-perceptual processing, and more specifically updating of information in working memory. Two possibilities emerge with this evidence. The suppression of the P3 component during the attentional blink would suggest that the deficit occurs at, or just before, this stage of working memory. However, if the P3 component is not suppressed, it would indicate that the deficit occurs after the information has reached working memory.  The experimental procedure followed a traditional approach, similar to experiment one, with a number of exceptions. Subjects were required to make two alternative forced choice responses. On 15% of the trials, T2 was an ‘E’ and on all other trials the computer randomly selected other letters. Subjects were asked to make a decision about T1, whether it was an odd or even number. Subjects were also required to make a decision as to whether T2 was an “E” or not. Even though the letter “E” appeared more often than any other letter it was defined as the infrequent category to the subjects and, according to previous research, it is the “probability of the task-defined category rather than the actual stimulus that determines P3 amplitude” (Vogel et al., 1998, p.1666). The results from this study show that the P3 component was completely suppressed during the attentional blink. This finding, therefore, suggests that the attentional blink reflects an operation that occurs either just before or during working memory.

The above finding is extremely important. Firstly, it appears to support a number of models that propose a post-perceptual mode of suppression (Shapiro et al., 1994, Chun et al 1995). It is also consistent with the findings of Maki et al. (1997) that words presented during the attentional blink could prime subsequent stimuli. It would also seem consistent with the finding that the attentional blink is attenuated if the presented words are meaningful to the subject (Cherry, 1953 Shapiro, Caldwell and Sorensen 1997).

 

Using these results, Vogel et al. (1998) proposed a new model of the attentional blink. They suggest that all items in the RSVP stream are fully identified and then stored in a conceptual short-term memory (CSTM). They are, however, not in a form that would enable them to be reported, and the information tends to decay very rapidly and is also prone to being replaced by subsequent information. Vogel et al.’s (1998) model, then, suggests that the attentional system attempts to consolidate the information in visual working memory (VWM). The route from CSTM to VWM is dependent on the degree of match between the representation in CSTM and the target template. Whilst the attentional system is occupied with this series of events, it has no resources available to process T2 beyond CSTM, resulting in intrusion errors and T2 not being able to be reported at all. (See appendix One taken from Vogel et al 1998 page 1670).

 

The diagram in appendix one illustrates two cognitive processes, namely, spatial selection and a more general mechanism. In the task subjects would be required to report the middle letter from each presentation. Vogel et al. (1998) suggest that the initial steps of sensory coding occur in parallel and appear not to be subjected to capacity limitations. It is further proposed that spatial attention is used to select a subset of information from the initial sensory feature and is passed on to higher perceptual processes, identified as being the pathway from the initial sensory processing to the stimulus identification stage. This pathway is hypothesised as being sluggish and has a tendency to be subjected to levels of interference. However, it is suggested that this pathway is fast enough to identify simple stimuli such as words and letters presented in an RSVP stream. The next stage appears to be too slow and, although all representations are received by the CSTM, the pathway from the CSTM to the VWM is limited, which means that several representations are lost in the process. It is additionally hypothesised that attention determines the transfer process, which only transfers relevant (and presumable salient) information to the VWM. Even once the information is in VWM, because of its severe capacity limitation, the information rapidly decays. The designers believe the model accounts for a perceptual-level attentional mechanism of the ‘early selection’ advocated by Cave and Wolfe (1990) and Treisman (1996), and for a post-perceptual attentional mechanisms promoted by Bundesen (1990), Duncan, (1980), and Duncan and Humphreys (1989). Vogel et al. (1998) believe that the two discreet attentional mechanisms are brought together in the one model. They advocate that having two attentional systems is the only way to explain the vast array of conflicting evidence already discussed in this paper. The evidence from this study strongly suggests that the attentional system is able to process stimuli to a point of lexical access and meaning extraction without storing it in a form that can be accessed for overt report. The evidence also seems to contradict the findings from a number of previous studies. For example, Broadbent and Broadbent suggested that non-target items are processed to a lesser degree than targets items. However, the results from a number of studies including, Allport (1977), Maki (1997) and Vogel et al. (1998) suggest that all items in an RSVP stream, including non-target words, are processed to a level that allows for priming and are most probably processed to a level of meaning, even items presented during the period of the attentional blink. According to Chun and Potter (1993) this stage contains information about the psychical characteristics of the item i.e. colour, category etc. and a comparison of information is drawn from the long-term memory and used in the selection process.

           

Raymond et al. (1995) hypothesised that the mechanisms involved in target identification are temporarily shutdown. There is now enough evidence to suggest that this is not the case and at no point during attentional processes do any of the stages appear to shut down. There may, however, be sub-units of visual processing that operate more slowly but the evidence strongly suggests that all processes are continuous, even during the attentional blink. Raymond et al. (1995) also suggested that once an item reaches the ‘sensory store’ an element of competition and confusion reign. Raymond et al. (1995) may, themselves, have been referring to this stage. However, Vogel et al. (1998) hypothesised that this pathway, although sluggish, is fast enough to process information present in an RSVP stream.


Broadbent (1957) and Raymond et al. (1995) suggested an early gating mechanism, which prohibits information once the initial information has been required. This inhibition of additional information acquisition leaves the attentional system blind, much like an eye blink.  The new evidence seems to point away from this idea, suggesting, instead, that all information is processed to the level of semantics, and information is only lost at or just before visual working memory. Deutsch and Deutsch (1963) might argue that selection occurs within the CSTM in order to pick out only the most ‘important’ information to send to the VWM.  Selection would not be based on any one aspect of the stimulus but on how meaningful the information is to the subject. If Broadbent’s (1957) filter does exist, the CSTM would be a good candidate.  This idea would also account for Treisman’s (1960) attenuation model of attention. Words that have high salience for the subject, such as their name, would be compared with meaning laden  items in long term memory and would be processed to the further stage of VWM.