How does conceptual similarity between the targets modulate the attentional blink?

Task-relevant-, and task-irrelevant target-target

similarity in Attentional Blink

Author: Husain al Sabari (10456473)

Supervision: Daniel Lindh, Ilja Sligte

University of Amsterdam

28-06-2019

Keywords: Attentional blink, Repetition blindness, Target-target similarity, Task-relevance

Abstract

In a phenomenon known as the attentional blink (AB), report of a second target (T2) in a stream of stimuli is constrained when it is presented within close temporal proximity to the first target(T1). Studies on the effect of target-target similarity on T2 reportability show inconclusive results. We believe that task-relevant (TR) repetition leads to a decline in T2 processing, while task-irrelevant (TI) repetition leads to an improvement of T2 processing. We measured T2 accuracy in response to TR repetition, and TI repetition. Although an attentional blink effect was elicited by the task, we could not reproduce earlier findings on target-target similarity in relation to T2 accuracy. We believe the absence of substantial results is due to insufficient statistical power, and suggest a replication of this study with methodological adjustments.

Introduction

Every day we are overloaded with visual stimuli. Through attention we are able to select a subset of the relevant information for further processing. For decades, the mechanisms underlying attention and visual processing have been studied. In their seminal paper, Raymond, Shapiro, and Arnell (1992) showed that when two targets (T1 and T2) are presented in close temporal proximity (200-500 ms) in a rapid serial visual presentation (RSVP), participants often show a reduced ability to report T2. This phenomenon is known as the attentional blink (AB) and indicated that there are temporal constraints on how fast we can process consecutive targets. A popular model explaining this phenomenon is the two-stage information-processing model (Chun & Potter, 1995; Dux & Marois, 2009). This model assumes that visual information is processed in two stages. In the first stage, visual information is rapidly detected, and features are analyzed, but this information is volatile and is susceptible to decay and being overwritten. It is in the second stage that visual information can be transferred to working memory for conscious report. But this second stage is capacity limited, thus a bottleneck withholds the second target to be processed for conscious report. Aside of time, there appear to be other factors affecting T2 reportability in classic attentional blink paradigms. For instance, categorical similarity between a target and a distractor produces a greater AB (Chun & Potter, 1995; Maki et. al., (2003); Shapiro et. al., (1994)). But also target-target similarity has been found to constrain conscious access (Kanwisher, 1987; Fagot & Pashler, 1995; Harris & Morris, 2004; Sy & Siebrecht, 2009; Lindh et. al., 2019) . Kanwisher (1987) discovered a reduction in the ability to report a second target stimulus in an RVSP when the target was visually similar to, or the same as the first target stimulus. This phenomenon is known as repetition blindness (RB) (Bavelier & Potter, 1992; Kanwisher, 1987). Other studies on target-target similarity within an AB paradigm referred to a multi-channel account of interference hypothesis (Awh et al., 2004; Serences et. al., 2009). This hypothesis proposes that two targets (T1 and T2) can only be reported if they are visually different enough to be processed in multiple feature channels. Sy & Siebrecht (2009) built on previous results and found that visual similarity was detrimental to T2 accuracy. This was only the

case when similarity was relevant to the task. In another study Lindh et. al., (2019)

defined

similarity based on mid-level visual features derived from a convolutional neural network,

and created four conditions for an attentional blink task.

Here, they found that images that were shared mid-level features led to a higher probability of reporting T2. Even though RB indicates that T2 processing is disrupted if it’s similar to T1, Lindh et. al., (2019) showed that visual similarity could also be beneficial for T2 processing, as is the case when the two targets show low- and mid-level similarity. These previous studies indicate that there is both a time and representational constraints on conscious access. Where RB-studies such as the one conducted by Sy & Siebrecht (2009) define similarity in a task relevant domain that is processed in working memory, similarity in the study conducted by Lindh et. al., (2019) is based on visual features, which are not stored in working memory and are irrelevant. This hypothesis leads us to two main questions: What is the effect of task-relevant repetition on T2 processing? And what is the effect of task-irtask-relevant repetition on T2 processing?

The aim of this study is to investigate the influence of conceptual similarity on T2 accuracy with less complex stimuli. In this study T1 and T2 will contain digit-, and letter targets that are conceptually similar (e.g. T1 = 2222, T2 = TWEE) or dissimilar (T1 = 4444, T2 = ACHT) in a rapid serial visual presentation (RSVP). Conceptual similarity is defined here as containing the same semantic meaning. A study conducted by Snowden (2002) showed that color attracts attention in a RVSP and attenuates AB. To investigate the role of task irrelevant low-level statistics, a second dimension will be added to the experiment. Thus, T1 and T2 will have similar or dissimilar colors (e.g. T1 and T2 both red vs T1 in green and T2 in red). Furthermore, to investigate the temporal perimeters of conceptual similarity in an AB paradigm, various T2 lags will be investigated (lag 1, lag 2, and lag 7). Here we tested our hypothesis that relevant repetition will lead to a larger AB, while task-irrelevant repetition will facilitate T2 processing and lead to a more probable recovery of T2 (see figure 1). The results of this study could give us insights on the parameters governing conscious visual attention and working memory, especially regarding the mechanisms underlying

priming and

interference and their role in attentional blink.

Figure1. Expected effect of types of similarity between T1 and T2 on T2 reportability. + +: increased T2 accuracy, + -: regular T2 accuracy, - -: decreased T2 accuracy.

Methods

Participants

We collected data from 18 native Dutch speaking participants recruited through a participant pool at the University of Amsterdam. Each participant completed two sessions of 1 hour each. Participants were awarded either through research credits or 10 euros per hour. No participants showed a lower than chance performance on reporting T1 identity.

Design

Participants were shown RSVP streams of non-word distractors, containing number targets that are either conceptually similar (e.g. T1 = TWEE (two in Dutch), T2 = 2222) or dissimilar (e.g. T1 = ACHT (eight in Dutch), T2 = 4444) on a computer screen with a 60hz refresh rate implemented using psychopy (Peirce et. al., 2011) and python 2.7 (Peirce 2007). The non-word distractors were

presented in black on a gray background and in order to manipulate a T-I dimension, T1 and T2 were of the same or of different colour (in balanced trial proportions). These two dimensions result in 4 conditions (See figure 1). The interstimulus interval (ISI) varied between lag 1, lag 2, and lag 7. Combinations of each dimension were randomised and evenly distributed among all trials. The stimulus-onset asynchrony was 110 milliseconds (ms), and the stimulus exposure time was 100 ms. After each trial, two dice were presented, and the participant was asked to identify the value that corresponds to each target. This is to force participants to represent each target value, rather than a low-level property (such as the shape of the digit). This design results in a 2 (T-R similar/dissimilar) x 2 (T-I similar/dissimilar) x 2 (lag 2 / lag 7) design.

Figure 2. Each combination of the task-relevant-, and the task-irrelevant dimensions. task-relevant different + task-irrelevant different (TRD-TID), task-relevant same + task-irrelevant different (TRS-TID), task-relevant different + task-irrelevant same (TRS-TIS), task-relevant same + task-irrelevant same (TRS-TIS)

Analysis

Trials with incorrect T1 assessment were filtered out from further analysis. T2 accuracy of the remaining trials were used for further analysis. The effect of T-R relevant) and T-I (task-irrelevant) similarity on T2|T1 accuracy was tested using a 2x2x3 ANOVA consisting of the factors: (1) T-R similarity/dissimilarity, (2) T-I similarity/dissimilarity, and (3) lag 2/lag 7.

Results

To test whether the task design elicited an attentional blink, the overall mean of T2 accuracy was calculated. A repeated measures ANOVA showed a significant effect of lag on T2 accuracy (F=13.3), df=2, P<0.001). There was a significant difference in the mean of T2 accuracy between lag 1 (0.904 ± 0.058) and lag 7 (0.948 ± 0.037, Bonferroni post-hoc, P=0.002), and lag 2(0.89 ± 0.049) and lag 7(0.948 ± 0.037, Bonferroni post-hoc, P<0.001) (see figure 2). Furthermore, A repeated measures ANOVA was used to measure the effect of task relevant repetition, and the effect of task-irrelevant repetition on T2 accuracy. The data showed no significant decrease in T2 accuracy in the Task relevant similarity condition (TRS-TID) (P>0.05). There was also no significant increase in T2 accuracy in the task-irrelevant conditions (TRD-TIS) (P>0.05). The repeated measures ANOVA also showed no significant difference between the mixed trials (P>0.05).

Figure 3. Combined mean of T2 accuracy for each interstimulus interval. A significant difference is present between lag 1 and lag 7, and lag 2 and lag 7.

Figure 4. The relative T2 accuracy is shown for the four conditions. Each colour corresponds to one of the four conditions. TRD: T1 was different in the task-relevant domain. TID: T1 was different in the task-irrelevant domain. TRS: T1 was the same in the task-relevant domain. TIS: T1 was the same in the task-irrelevant domain.

Discussion

With this study we tested our hypothesis of opposite effects conveyed either by task-relevant or task-irrelevant repetition. We manipulated task-relevancy by either having T1 and T2 being the same or different value used for report. Task-irrelevancy was manipulated by having T1 and T2 being the same or different color. We show no effect of task-relevant repetition on T2 reportability, nor did we find an effect of task-irrelevant repetition on T2 reportability, although an attentional blink effect was elicited by the task.

The absence of any repetition blindness (RB) effect is incongruous with our expectations that were based on robust previous findings in several studies (Kanwisher, 1987; Sy & Siebrecht, 2009). Similarly, we did not replicate earlier findings of task-irrelevant similarity leading to more effective processing of T2 (Lindh et. al., 2019). Lindh et. al. (2019) used a convolutional neural network to select images based mid-level features to test in an attentional blink (AB) paradigm. They showed that when the targets shared mid-level features, participants had a much higher probability to recover T2. Sy & Siebrecht (2009) showed that visual similarity between T1 and T2 was detrimental for T2 recovery, when similarity was task-relevant. Our hypothesis explaining the difference between these findings is that in the study of Lindh et. al. (2019) similarity was based on visual features, which are not stored in working memory (i.e. what we call task-irrelevant), while the study conducted by Sy & Siebrecht (2009), repetition blindness, and other studies investigating target-target similarity define similarity in a task-relevant domain.

The absence of substantive results should be seen in light of methodological factors. We believe that the lack of sufficient statistical power is a major cause for this. A greater number of participants should increase statistical power. Another factor that might have contributed to the absence of substantive findings, is the stimulus exposure time. Considering the fact that the average T2 accuracy in lag 2 this study is 89% (±4.9%), the task might have been too easy due to the stimulus exposure time being too long. Other studies on target-target similarity report T2 accuracies between 65% and 85% (Kanwisher (1987); Sy & Siebrecht (2009); Lindh et. al., (2019)). Shortening the

stimulus exposure time makes the task harder and magnifies the attentional blink effect, possibly

Lag 1 Lag 2 Lag 7

TRD-TID 0.918 0.896 0.938 TRD-TIS 0.913 0.907 0.945 TRS-TID 0.894 0.873 0.957 TRS-TIS 0.891 0.89 0.951 0.75 0.8 0.85 0.9 0.95 1 Re lat iv e cor re ct T 2

leading to replication of the previous results (Kanwisher (1987); Nieuwenstein (2009). We

recommend using a stimulus exposure time of 58ms instead of 100 (Kanwisher (1987). This should increase the difficulty level of the task, increasing the probability of an AB per trial, thus creating a bigger contrast between conditions. We suggest a replication of this study with the suggested methodological adjustments, as a better understanding of the processes governing attentional blink and other attentional limitations might give us a better understanding of the mechanisms underlying our attention and memory system.

References:

Awh, E., Serences, J., Laurey, P., Dhaliwal, H., van der Jagt, T., & Dassonville, P. (2004). Evidence against a central bottleneck during the attentional blink: Multiple channels for configural and featural

processing. Cognitive Psychology, 48(1), 95–126. https://doi.org/10.1016/S0010-0285(03)00116-6

Bavelier, D., & Potter, M. C. (1992). Visual and phonological codes in repetition blindness. Human Perception and Performance, 18(1), 134–147.

Chun, M. M., & Potter, M. C. (1995). A two-stage model for multiple target detection in rapid serial visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 21(1), 109 127.

Dux, P. E., & Marois, R. (2009). The attentional blink: A review of data and theory. Attention,

Perception, & Psychophysics, 71(8), 1683-1700.

Fagot, C., & Pashler, H. (1995). Repetition blindness: Perception or memory failure?. Journal of

Experimental Psychology: Human Perception and Performance, 21(2), 275.

Harris, C. L., & Morris, A. L. (2004). Repetition blindness occurs in nonwords. Journal of Experimental

Psychology: Human Perception and Performance, 30(2), 305.

Kanwisher, N. G. (1987). Repetition blindness: Type recognition without token individuation. Cognition, 27, 117–143.

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet Classification with Deep Convolutional Neural Networks. Advances In Neural Information Processing Systems, 1–9.

https://doi.org/http://dx.doi.org/10.1016/j.protcy.2014.09.007

Lindh, D., Sligte, I. G., Assecondi, S., Shapiro, K. L., & Charest, I. (2019). Conscious perception of natural images is constrained by category-related visual features. https://doi.org/10.1101/509927

Maki, W. S., Bussard, G., Lopez, K., & Digby, B. (2003). Sources of interference in the attentional blink: Target distractor similarity revisited. Perception and Psychophysics, 65(2), 188 201.

Peirce, J. W. (2007). PsychoPy—psychophysics software in Python. Journal of neuroscience methods,

Peirce, J., Gray, J., Halchenko, Y., Britton, D., Rokem, A., & Strangman, G. (2011). PsychoPy–A Psychology Software in Python.

Raymond, J. D., Shapiro, K. L., & Arnell, K. M. (1992). Temporary suppression of visual procesing in a RSVP task: an attention blink? Journal of Experimental Psychology, 18(3), 849–860.

Serences, J., Scolari, M., & Awh, E. (2009). Online response-selection and the attentional blink: Multiple-processing channels. Visual Cognition, 17(4), 531–554.

https://doi.org/10.1080/13506280802102541

Shapiro, K. L., Raymond, J. E., & Arnell, K. M. (1994). Attention to visual pattern information produces the attentional blink in rapid serial visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 20(2), 357 371.

Sy, J. L., & Giesbrecht, B. (2009). Target-target similarity on the attentional blink: Task-relevance matters! Visual Cognition, 17(3), 1–10. https://doi.org/10.1080/13506280802349746