Abstract
It is widely accepted that we recognize letters by first detecting the presence of letter features, followed by integrating the available features to form an overall percept of the letter. Does the processing of letter features depend on spatial and temporal factors? Here, we used a data-driven approach to investigate how letter features are processed amongst letters that are in close proximity to one another, when their spatial content is restricted, and for two stimulus durations. We presented trigrams (sequences of three lowercase letters) to five observers whose task was to identify all three letters on each trial. Trigrams were presented for 50 or 200 ms at 10° in the lower visual field. Letters were rendered in lowercase Arial font (x-height=1.2°, center-to-center separation=1.3°), and were band-pass filtered with a center frequency of 1.35 or 5 c/letter. Each observer completed 5200 trials for each combination of conditions. Letter features for each response and stimulus letter were classified as round, descenders, ascenders, oblique and arch (Bouma, 1971). In general, letter recognition error rates were higher for the middle than the two outer letter positions, for the 50- than the 200-ms condition, and for filtered than unfiltered letters (although the error rates were similar for the 1.35 and 5 c/letter conditions). For the error trials, letter mislocations occurred only sparingly (averaging ~5%). Instead, we observed significant rates of occurrences of feature interactions amongst neighboring letters, such as mislocations, gains, losses, doubling, and sometimes masking of one another. The rates of occurrences of these feature interactions differed for the two durations, and exhibited a dependence on the filter condition. Our findings provide strong evidence for the contingent nature of spatio-temporal processing of letter information when letters are presented in close proximity to one another, and will help constrain models of visual crowding.
Acknowledgement: NIH Research Grant R01-EY012810