I can see there is still a little confusion, so let me make a few clarifications. I think this should clear up a few things.
1. Here's a pretty good definition of "white noise" from Wilkepedia:
White noise is a random signal (or process) with a flat power spectral density. In other words, the signal's power spectral density has equal power in any band, at any centre frequency, having a given bandwidth. White noise is considered analogous to white light which contains all frequencies.
An infinite-bandwidth white noise signal is purely a theoretical construct. By having power at all frequencies, the total power of such a signal is infinite. In practice, a signal can be "white" with a flat spectrum over a defined frequency band.
2. There is nothing "unknown" in the experiment. To say that it is unknown what the 20,000 slides contained is wrong, or "probably random" is a mis-characterization. They contained static white noise.
3. There are no Ss in any of the slides, only small systematic correlations with what each subject's knowledge of an S is to them individually, on average that is. As jrich pointed out, this could have been done with an "N", "L", "P", etc., and the results would have been similar. Are we to believe that the slides contain all of these letters, in both upper and lower case, in 30 different font types (see paper for description of statistical analysis with all the font types) in each or some slides? Well, maybe if we think about it terms of each individual pixel and how they might match up, but not in the sense of what a letter is to us. There were no letters. Only imaginary ones.
4. Finally, from the paper:
"To test that the S s in the classification images were not simply the result of our own superstitious perceptions, we correlated these classification images with the 26 letters of the alphabet from 31 fonts, 3 in 7 styles (normal, italic, bold, underline, outline, condense, extend) and in upper- and lowercase, for a total of 11,284 Pearson correlations. All letters were resized and cut to fill a square window in the two possible ways (i.e., with the width of the letter occupying the whole width of the window or the height of the letter occupying the whole height of the window). The highest correlations were obtained with the following letters (see Fig. 1, inset c, for each observer): for R.G., an uppercase Courier New bold S scaled horizontally (r=.557); for N.L., a lowercase Verdana regular-style S scaled horizontally (r=.553); and for M.L., an uppercase Arial bold S scaled vertically (r=.704). On average, confounding font, style, case, and observer, the largest correlation between the classification images and the 26 letters of the alphabet was found for S (see Fig. 2), a pattern true for each observer.
In sum, we induced superstitious perceptions of Ss by instructing 3 observers to detect this letter in noise. They did not know that the stimuli never comprised the letter, but only white noise. Thus, if the observers had performed only according to the stimulus (i.e., in a bottomup manner), their classification images should have had the same properties as averaged white noise—that is, constant energy across all spatial frequencies. However, there were marked peaks of energy below 3 cycles per letter. These must have arisen from top-down influences on the interpretation of white noise—very low correlations between input noise and the memory representations of the letter. Further analyses revealed the shape of the letters that the observer s thought they saw." (Superstitious, Schyns et al.)
rd
