This paper reviews two hypotheses for how the psychological correlates of
phoneme categories are acquired. The authors argue against the hypothesis that
phonemes are learned via a comparison of minimal pairs, and provide evidence
in favor of the hypothesis that phonemes are learned from the distribution of
sounds in a language. English-speaking adults were presented with words from
an artificial language for 9 minutes, without any information about word
meaning. Participants were then tested on their categorization of phonemes in
the language. Participants trained on a bimodal distribution of the experimental
stimuli were more likely than participants trained on a monomodal distribution,
to indicate during the test phase that the experimental stimuli corresponded to
two phoneme categories. These results support a distribution-based model of
phoneme learning.