Abstract
Recent studies have emphasized the importance of letters as units in word recognition (e.g. Pelli et al., 2003). However, recent findings from our lab suggest different feature extraction strategies for single letters (Fiset et al., 2008; 2009) and words (Blais et al., 2009). To explore these surprising findings further, we used a psychophysical method derived from Huey (1908). Twenty-seven observers were asked to recognize lowercase Arial letters, trigrams, 3-letter words, and 5-letter words of which only a proportion (from 1/6 to 5/6) of either the top or the bottom was visible. For each stimulus type, for both top and bottom conditions, we fitted identification performance with a Gaussian cumulative function and located the intersection of the two functions. Akin to Huey (1908; see also Blais et al., 2009), we found a clear and significant bias for the upper parts of words. Interestingly, no such effect was found for single letters (neither on average, nor taking letter frequencies into account) and trigrams (which eliminates crowding as a potential explanation). To understand the source of this upper bias and its specificity to lexical letter strings, we first submitted an ideal observer to identical tasks with added white Gaussian noise—the ideal observer was essentially unbiased. Second, we attempted to explain the upper bias for words by using human biases for each individual letters (e.g., human observers show a bias for the top part of ‘t’). However, we found that the individual letters' biases could not account for the words' bias—even assuming that every word is recognized solely by its most upper-biased letter is insufficient to reproduce the words' bias. Finally, we discovered a bias in human syllabic trigrams, comparable to that of words. We conclude that the visual system bypasses single letter identification on the road to word recognition.
Meeting abstract presented at VSS 2012