Use of spatial frequencies information in normal readers and a letter-by-letter dyslexic patient

Abstract

We employed the Bubbles technique (Gosselin & Schyns, 2001) to examine the use of spatial frequencies in normal (N = 10) and letter-by-letter (N = 1) reading. Each stimulus sampled either an individual letter, or one of 2,000 four-to-seven-letter words (one letter spanned 0.38 deg of visual angle × 0.6 deg of visual angle) by dot multiplying their Fourier spectrum with white Gaussian noise convolved with a Gaussian function (Std = 0.156 of the Nyquist frequency). Accuracy was 97% on average for normal readers and of 71 % for LH, a so-called letter-by-letter dyslexic. Multiple linear regression was performed on reading latency and sampling noise. We found that normal letter identification is mediated by a single frequency band centered at 3 cycles per letter (see also Solomon & Pelli, 1994), and normal word reading, by another, partly overlapping, frequency band peaking at 1.6 cycles per letter. We found a different pattern of results for LH: in letter recognition, he uses a frequency band centered at 1.9 cycles per letter and, in word recognition, a somewhat narrower frequency band centered at 1.6 cycles per letter. Our results suggest that letter-by-letter reading is caused by a deficit in using the spatial frequencies normally associated with letter recognition and that this deficit is exacerbated in word recognition. We will argue that spatial frequencies around 3 cycles per letter are necessary for the explicit word identification, and that lower spatial frequencies are responsible for the implicit effects found in normal and pathological reading.