In theory, of course, Cohort would predict that nonwords would be recognised as nonwords as soon as they stopped being like real words, based on a 'left-to-right' (through time) analysis of their sounds. So blorsilize (see exercise 8.3) should be recognised at the first vowel, since this is where it no longer maps onto existing words. In the experiment mentioned in this exercise, however, 3-syllable nonwords like blorsilize took longer to identify as nonwords than 2-syllable nonwords with matched deviation points (i.e. in this case also at the first vowel) like shrortile.
A possible reason for this result is that even though the sequence of sounds no longer maps onto a real word starting with the same sequence, we are used to hearing people mispronounce words, and so we do not straightaway opt to say that this is not a word, in case there has been a mispronunciation, and listening to the rest of the stimulus might help decide this. Since there is more of the stimulus still to come in the 3-syllable nonwords than in the 2-syllable nonwords with matched deviation point, the 3-syllable ones take longer to reject as not being words.
The additional finding that rejection times neasured from the end of the stimulus were shorter for these 3-syllable nonwords than for the 2-syllable nonwords indicates that listeners did not have to listen to the entire stimulus before they could start to make their decision.