The authors found that listeners’ perception of vocoded speech did not improve in the nonword condition. They thus concluded that although learning occurs at the phonological level (since improved performance extended to words that were not encountered in the training set), it is mediated by lexical representations.