Endel Põder

Change in feature space is not necessary for the flash-lag effect.

To achieve perceptual match between a flashed target and a gradually changing one, the flashed target should have the feature value corresponding to the value to be obtained by the gradually changing target only later. Flashed target should be positioned ahead of the continuously moving one in order to be perceived as aligned (Nijhawan (1994). Nature, 370, 256-257); with continuously changing colour, spatial frequency, pattern entropy or luminance, the flashed target should have feature value which changing target obtains only later (Sheth, Nijhawan, & Shimojo (2000). Nature Neuroscience, 3, 489-495). It was found that flash-lag effect is present even if the continuously accumulating pre- and post-target input consists in spatially and featurally invariant stimulation. The perceptual precedence of the target in stream over its synchronous single-flashed replica may result from perceptual acceleration where newly arriving visual signals are facilitated by the locally preceding stimulation.

Attentional gating models of object substitution masking.

Di Lollo, Enns, and Rensink (2000) proposed the computational model of object substitution (CMOS) to explain their experimental results with sparse visual maskers. This model supposedly is based on reentrant hypotheses testing in the visual system, and the modeled experiments are believed to demonstrate these reentrant processes in human vision. In this study, I analyze the main assumptions of this model. I argue that CMOS is a version of the attentional gating model and that its relationship with reentrant processing is rather illusory. The fit of this model to the data indicates that reentrant hypotheses testing is not necessary for the explanation of object substitution masking (OSM). Further, the original CMOS cannot predict some important aspects of the experimental data. I test 2 new models incorporating an unselective processing (divided attention) stage; these models are more consistent with data from OSM experiments. My modeling shows that the apparent complexity of OSM can be reduced to a few simple and well-known mechanisms of perception and memory.

Illusory reversal of temporal order: the bias to report a dimmer stimulus as the first

When two objects are presented in rapid succession, observers find it difficult to discriminate their temporal order. Below certain limit (e.g., 20-70 ms), the rate of correct temporal order judgement is reported to be about 50% (i.e., close to chance level). However we have found stimulus conditions where order discrimination drops significantly below chance level: the stimulus that is presented as the second is reported as the first. It is necessary that the stimuli are very brief, spatially overlapping, clear-cut backward and forward masking is absent, stimulus onset asynchronies are very short, and luminance contrast of the following stimulus is considerably lower than luminance contrast of the first stimulus. The higher the contrast ratio, the stronger the order reversal effect. However, because also in the conditions where the two stimuli are presented synchronously, the dimmer target is perceived as the first, the effect should be attributed to some implicit bias which enforces subjects to regard a more contrasted stimulus as the one that appears subsequent to the less contrasted stimulus.

Perceptual acceleration of objects in stream: Evidence from flash-lag displays

An object in continuous motion is perceived ahead of the briefly flashed object, although the two images are physically aligned (Nijhawan, 1994), the phenomenon called flash-lag effect. Flash-lag effects have been found also with other continuously changing features such as color, pattern entropy, and brightness (Sheth, Nijhawan, & Shimojo, 2000) as well as with streamed pre- and post-target input without any change of the feature values of streaming items in feature space (Bachmann & Põder, 2001a. 2001b). We interpret all instances of the flash-lag as a consequence of a more fundamental property of conscious perception in general: acceleration of the speed with which samples of perceptual information become represented in explicit format immediately after the stimulation onset. Decreased visual latency of the samples of stimulus information from the streamed input leads to the relative perceptual lag for the separately flashed stimulus because it is not preceded by adjacent sensory input that would have accelerated its perception. Experimental support for the notion of perceptual acceleration is reviewed.

Variations in backward masking with different masking stimuli: II. The effects of spatially quantised masks in the light of local contour interaction, interchannel inhibition, perceptual retouch, and substitution theories.

In part I we showed that with spatially non-overlapping targets and masks both local metacontrast-like interactions and attentional processes are involved in backward masking. In this second part we extend the strategy of varying the contents of masks to pattern masking where targets and masks overlap in space, in order to compare different masking theories. Images of human faces were backward-masked by three types of spatially quantised masks (the same faces as targets, faces different from targets, and Gaussian noise with power spectra typical for faces). Configural characteristics, rather than the spectral content of the mask, predicted the extent of masking at relatively long stimulus onset asynchronies (SOAs). This poses difficulties for the theory of transient-on-sustained inhibition as the principal mechanism of masking and also for local contour interaction being a decisive factor in pattern masking. The scale of quantisation had no effect on the masking capacity of noise masks and a strong effect on the capacity of different-face masks. Also, the decrease of configural masking with an increase in the coarseness of the quantisation of the mask highlights ambiguities inherent in the re-entrance-based substitution theory of masking. Different masking theories cannot solve the problems of masking separately. They should be combined in order to create a complex, yet comprehensible mode of interaction for the different mechanisms involved in visual backward masking.

Search for feature and for relative position: measurement of capacity limitations.

Palmer and his associates (Palmer, Ames & Lindsey (1993). Journal of Experimental Psychology: Human Perception and Performance, 19, 108-130; Palmer (1994). Vision Research, 34, 1703-1721) have confirmed that searches for simple feature targets are not limited by perceptual processing capacity and the effect of set size on performance can be accounted for by integration stage processes only. In this study I used a similar difference threshold method with target and distractor stimuli defined by the relative position of their elements (line drawings of bisected squares) and found clear capacity limitations. Feature search condition, however, with nearly comparable bisected square stimuli did replicate the results of Palmer and associates. This experiment demonstrates that a search for targets defined by relative position in the set of line drawing type of stimuli is fundamentally different from a search for more simple (feature) stimuli and may conform to a strict capacity limited model.

Let's correct that small mistake

A huge number of informal messages are posted every day in social network sites, blogs, and discussion forums. Emotions seem to be frequently important in these texts for expressing friendship, showing social support or as part of online arguments. Algorithms to identify sentiment and sentiment strength are needed to help understand the role of emotion in this informal communication and also to identify inappropriate or anomalous affective utterances, potentially associated with threatening behavior to the self or others. Nevertheless, existing sentiment detection algorithms tend to be commercially oriented, designed to identify opinions about products rather than user behaviors. This article partly fills this gap with a new algorithm, SentiStrength, to extract sentiment strength from informal English text, using new methods to exploit the de facto grammars and spelling styles of cyberspace. Applied to MySpace comments and with a lookup table of term sentiment strengths optimized by machine learning, SentiStrength is able to predict positive emotion with 60.6p accuracy and negative emotion with 72.8p accuracy, both based upon strength scales of 1–5. The former, but not the latter, is better than baseline and a wide range of general machine learning approaches.

Spatial-frequency spectra of printed characters and human visual perception

It is well known that certain spatial frequency (SF) bands are more important than others for character recognition. Solomon and Pelli [Nature 369 (1994) 395-397] have concluded that human pattern recognition mechanism is able to use only a narrow band from available SF spectrum of letters. However, the SF spectra of letters themselves have not been studied carefully. Here I report the results of an analysis of SF spectra of printed characters and discuss their relationship to the observed band-pass nature of letter recognition.

Size-based selection in rapid serial visual presentation.

Several recent studies [Farell, B., & Pelli, D. (1993). Can we attend to large and small at the same time? Vision Research, 33, 2757-72; Shih, S., & Sperling, G. (1996). Is there feature-based attentional selection in visual search? Journal of Experimental Psychology: Human Perception and Performance, 22, 758-79] have found that visual selection based on the size of stimuli is impossible when the stimuli are presented in rapid succession in overlapping positions (RSVP paradigm). In the present study effective size-based selection is demonstrated in several conditions with RSVP stimuli. Attention to specific size is highly efficient when stimuli are presented in a single location (at fixation point) and may be possible also with a few (2-4) locations. When overlapping small and large characters are presented without abrupt onsets, then selection by size is effective at least over six locations. The results are explained by certain mandatory properties of spatial attention.

Variations in backward masking with different masking stimuli: I. Local interaction versus attentional switch.

The types of stimuli used as targets and masks considerably change the masking functions in a way that requires us to abandon any single mechanism of masking as the sole explanation of backward masking. In the first of two reports in which the problem of the mask-dependence of masking is addressed, we explore the role of the relative spatial positioning of targets and masks in order to differentiate between local interaction and attentional models. If single letters were masked by double-letter masks then the relative spatial arrangement of the letters, which was changed in order to vary the involvement of metacontrast-like processes, had an effect at shorter SOAs, but not at longer SOAs where strong masking still persisted. This poses difficulties for proposing local contour interaction as the main mechanism of masking. Similarly, crowding effects alone cannot explain the results. Backward masking also involves attention being directed to working-memory processing of the succeeding object while abandoning the preceding object.