ИСТИНА

Visual working memory (VWM) is a topic of vibrant discussions since the seminal study by Luck and Vogel measuring its capacity for visual features and objects1. Among the currently discussed problems is the influence of bottom-up and top-down factors on the information storage in VWM2. It has been shown to be influenced by perceptual organization3. However, less is known about the VWM temporal structure. Of interest is the issue of chunking, or unitization4,when an observer is processing temporally distributed stimuli belonging to a larger whole, e.g. letters forming a word. To probe into the nature of information buffering and chunking in VWM, we used a dual-task rapid serial visual presentation (RSVP), a procedure traditionally used to study the attentional blink phenomenon5. In our experiment, 24 participants were presented with 210 Russian words in a letter-by-letter manner (120 ms per letter with no interstimulus interval). All letters, presented in the center of a computer display, were printed in black, however, among them was a letter printed in white, followed by 4 or 5 letters and a mask (#). The participants were asked to report on the shape of the white letter (which could be either printed or handwritten) and to read a word beginning with it. The words used were mid-frequency 5- or 6-letter nouns. If a participant failed to report a word, (s)he was encouraged to name aloud all letters (s)he could identify. The procedure produced quite a number of errors classifiable into various types. The dichotomies we used for error classification were reporting either words or nonwords, skipping or not skipping letters in the reported strings (complete vs. incomplete report), adding or not adding letters not presented in the RSVP string, changing or not changing letter order, and reporting or not reporting on letters presented before the 1st target letter coloured in white. Distribution of errors was analyzed. With most errors to be letter omissions or misidentifications, two groups of errors were of special interest as regards VWM. First were anagram errors such as reporting on all letters but in a wrong order forming either a nonword or a different meaningful word (e.g. “Thread” presented, “Hatred” reported), 25.1% of all errors. Second were reporting on letters presented before the white target letter as a part of the target string, also producing either a nonword or a meaningful word, 24.6% of all errors (e.g. “…dfrSpite” presented, “Sprite” reported). The analysis of these two groups of errors allows assuming that the order of temporally distributed stimuli is not encoded during their presentation (a result agreeing with some previous findings6), and the to-be-reported larger units are first to be constructed. The data also support the hypothesis that VWM is subject to top-down strategic regulation such as maintaining the maximum number of stimuli accompanying the 1st target (the white letter) as a reference mark, but not allowing to distinguish between stimuli preceding and following it.