In a previous post I have questioned whether we need to explicitly represent phonemes in speech perception. Massaro and others have raised this issue in the past. Phonemes, the line of thinking goes, are only really important for production. There are linguistic arguments for this that I won't detail here. There is also well-known speech error data which shows that phoneme size units can break off and dislocate themselves. Here I want to highlight some evidence from aphasia. A reviewer of one of my papers pointed me to this study by Lindsey Nickels and David Howard.
A group of aphasics who exhibited speech production errors were asked to repeat words that varied in terms of the number of phonemes, number of syllables, or syllable complexity (defined in terms of consonant clusters). These variables are, of course, highly correlated, but the stimuli were carefully designed so that the contribution of each of these factors could be examined using logistic regression analyses.
The main result was that only number of phonemes in a word predicted correct repetition (see graph below derived from their Table 4) and once this variable was taken into account, the number of syllables or syllable complexity did not explain any additional variance.
Phonemes seem to matter in speech production. I have to say, though, that I'm not fully convinced that the others factors aren't also important.
Nickels, L., & Howard, D. (2004). Dissociating Effects of Number of Phonemes, Number of Syllables, and Syllabic Complexity on Word Production in Aphasia: It's the Number of Phonemes that Counts Cognitive Neuropsychology, 21 (1), 57-78 DOI: 10.1080/02643290342000122