Advances in TVA-based visual attention research, Part I: Moving towards new stimuli, tasks, and data

The Attentional Blink (AB) refers to a deficit reporting a second target (T2) presented 200-500 ms after a preceding target (T1) imbedded in a stream of distractors. Several theories about the origin of the AB have been proposed. Some theories claim that the AB is the result of a temporarily closing of an attentional gate, others that it is caused by a reduction in the capacity to process visual information. A third group of theories argue that it is an impairment in the ability to filter out the surrounding distractors which leads to the T2 deficit. In this talk, I will present the results of three experiments in which we systematically vary the exposure duration and composition of the T2 display allowing us to decompose the T2 deficit in terms of well-established parameters based on a Theory of Visual Attention (TVA). As the different AB theories make specific predictions in regard to which parameters should be affected during the AB, we are able to test their plausibility. All three experiments consistently show a lower processing speed of T2 during the AB, supporting theories of reduced capacity. No evidence supporting gating or filtering theories are found.

Dual-task costs – the decrement in performance in one or both tasks under a dual-task condition – are commonly explained through the central bottleneck model, and are often studied with the Psychological Refractory Period (PRP) paradigm. PRP proposes that whilst perception and motor response can operate in parallel, the response selection stage is subject to a central bottleneck. However, whilst the central bottleneck model can explain what occurs on a processing level, there is still the question as to how information is processed on a more basic perceptual level. We therefore integrated the “Theory of Visual Attention” (TVA; a mathematically formulated model which allows visual attentional parameters to be independently quantified) into the PRP paradigm. Specifically, we want to investigate whether response selection influences visual attention processing. The results showed that both the processing speed (parameter C) and the storage capacity of the visual short-term memory (parameter K) varied as a function of the stimulus onset asynchrony. This indicated that these attentional processes share a common capacity limitation with the response selection. However, the visual threshold (parameter t0) remained unaffected, suggesting that the earliest stages of visual processing are performed independently of the central bottleneck.

As a formal theory, Bundesen’s theory of visual attention (TVA) allows the precise estimation of several theoretically meaningful parameters involved in visual selection and recognition. As of yet, TVA has almost exclusively been used in restricted empirical scenarios such as whole and partial report and with strictly controlled stimulus material. We present a series of experiments in which we test whether the advantages of TVA can be exploited in more realistic scenarios with varying degree of stimulus control, for instance with photographs of different scenes, in a driving simulator with computer games, or with brief experimental sessions conducted on a mobile device . Besides answering the question whether TVA is helpful to study attention in the wild, our focus is on possible trade-offs between factors such as motivating tasks, length of experiment, and number of participants.

Theoretical accounts of attention that have been dominant over the last close-to 50 years have trouble explaining recent results from studies on visual foraging tasks, where observers have to select many targets of different types among distractors within a single trial. For example, the results suggest that theories with simple fixed capacity limits may be too restrictive. Participant’s strategies may reflect a more complex interaction between capacity and task demands. Secondly, typical response time patterns from visual search tasks, that have been used to inspire prominent theories of visual attention, are only seen for the last target within a foraging trial. These selections are also much slower than the majority of other selections within trials. Thirdly targets that are distinguishable from distractors by a single color should pop out in a feature map, according to standard theories, but clearly they do not. Fourth, selection times are often comparable for feature and conjunction foraging which contradicts standard theories of attention, while patterns of the order of target selection differs strongly between those conditions. I will discuss what sort of additions and modifications should to made to theoretical accounts of attention, and whether current theories should be modified or whether we should even „start from scratch“ by building a new theory of visual attention to account for the results. I will also speculate whether the flexibility of the theory of visual attention (TVA) may make it uniquely suited to account for performance patterns in foraging studies.

Visual foraging, the search for many targets of the same types in an area, requires observers to maintain multiple target templates in active states or to multiplex their activation. There is much debate concerning the underlying mechanisms, but so far, no formal, quantitative models have been put forward. In the present work, we extend Bundesen’s Theory of Visual Attention to visual foraging tasks and model template switching. Using the model, we simulate foraging data to compare simulations with experimental recordings. In addition to manual selections recorded with a stylus on large tablet PC, we also analyze gaze data gathered with a new setup that allows for manual foraging and concurrent eye tracking. Our results show that TVA, with a few plausible additions, can account for typical patterns in target choice. Moreover, the manual responses occur in close lockstep with the preceding eye movements.

We present recent theoretical advancements in TVA that provide an account for response time (RT) distributions in attention-based tasks. Although the scope of TVA is more general, TVA-based research so far has focused on accuracy-based tasks using highly discriminable stimuli to investigate effects of attention on selection of categories (pigeonholing) and selection of elements (filtering). The new RT model—we called it Poisson Random Walk—is a parallel processing model that can explain RT distributions in speeded RT tasks by describing response selection in mutually confusable stimuli. It is compatible with many other theories serving as a front-end to explain the processing rates underlying the decisions. Using TVA as a front-end, we have tested the new model with data from single stimulus recognition tasks with two or more perceptual categories. Here, we present new applications to speeded responses to targets presented among distractors. The first application is visual search. The Poisson Random Walk can account well for simple (feature) search distributions by incorporating the filtering mechanism of TVA. Conjunction search times, on the other hand, require an introduction of another set of weights that describe possible serial processing. With these new weights, we also include a central idea of Guided Search into the TVA-based model. The second application is a speeded target recognition task with multiple distractors. This task is a variant of the standard speeded 2-alternative forced choice task combined with partial report. We close our presentation with an outlook on future directions.