Continuity in perception: Contrasting serial dependence, aftereffects and learning of ignored information.

Abstract

When we are viewing the environment around us, we are exposed to an overwhelming continuous stream of information: not everything can be processed at once, and the perceptual system must pick and choose. We must direct our attention and select stimuli of interest while ignoring or discarding irrelevant information in order to solve daily tasks and problems. An effective tool to simplify such a daunting task is for the perceptual system to chunk information together. Various pieces of spatial and temporal information can be used to bind several features which correlate in space and time into singular objects. However, because the visual world is ambiguous and prone to noise, correctly combining and selecting similar yet distinct features and objects is still an ambitious undertaking. Prior assumptions and expectations of our visual environment provide additional aid. They do this, for example, by assuming that objects and their defining features are stable entities which are maintained over time and taking advantage of such stability and regularities over time. Additionally, maintaining representational maps containing predictions of relevant or irrelevant locations or features associated with objects further eases the task of sorting the vast amounts of continuous visual information. A vast number of history-driven biases in attention and behavior have been demonstrated experimentally to improve performance when locating and discriminating between visual stimuli. Encoding of sequentially presented stimuli is attentionally facilitated when stimuli share a feature or location. Similarly, if such characteristics occur within the same object, attentional effects are spread within the entire object. The continuous flow of information can be subject to transient changes, such as blinking and changes in lighting, requiring a mechanism which employs this general predictability in order to maintain perceptual stability despite such transient changes. Serial dependence, a general attraction to previously viewed stimuli information, has been proposed as such a perceptual smoothing mechanism (J. Fischer & Whitney, 2014). It is proposed that perception is smoothed, and perceptual decisions of present stimuli are influenced by past stimuli within a spatial and temporal continuity field. Further, attention has been deemed a crucial factor: perceptual decisions about attended stimuli are attracted towards previously attended stimuli. In the papers of this thesis, we further explore the role of attentional selection in serial dependence by employing and adapting spatial, feature-based, and object-based attentional and serial dependence paradigms. In the first paper, we investigated representation traces left over after attentional filtering of irrelevant stimuli when spatially suppressed. Observers performed two tasks sequentially: first, they performed a discrimination task while ignoring a distractor Gabor, reproducing the orientation of a Gabor afterwards. Notably, we did not observe a difference between suppressed or non-suppressed distractors, which equally interfered with perceptual decisions of the reported Gabor. This showed that perceptual decisions are biased away from the features of recent distractors. Such effects could be caused by either attentional filtering or active removal of irrelevant traces of information. We speculate that distractors were processed to the extent that they left a trace or interfered with other ongoing traces of prior stimuli. In the second paper, we studied the role of feature-based attention and task demands related to secondary contextual features. Participants performed either a detection task, responding to a cued color, or a discrimination task, actively responding differently to two colors, while a sequence of red or green Gabors were presented; the orientation of the last Gabor was reproduced. We observed that during the detection task, attraction only occurred towards the target color, whereas during the discrimination task perceptual decisions were attracted to previous stimuli regardless of color. The results provide further evidence that serial dependence is modulated by task demands and task contextual information. They further suggest that serial dependence can operate on features and does not need object representations, i.e., attraction can occur between features of different objects. We propose an account of parsimony, wherein the representation required by the task determines what information is propagated from one instance to another. Finally, in the third paper, we investigated the tuning of serial dependence to object representations, employing object-based attention by presenting a Gabor at one of the ends of two triangles. When the previous Gabor was presented within the same object, we observed a widening of the range of relative orientation differences in which attraction took place. We suggest an increase in stimulus sensitivity caused by attentional employment based on objects. Overall, our results show that task demands, attentional facilitation and filtering processes have an important and nuanced role in serial dependence.