Perception of speech segments can be shifted by the surrounding phonetic context in a manner that appears to compensate for acoustic effects of coarticulation. Different explanations for this effect have implicated the phonemic, articulatory or acoustic content of the context sounds as being the effective cause of these identification shifts. To test the relative importance of these factors, phonemic and articulatory content were stripped from context sounds. Context effects were still obtained for contexts that only resembled speech in some spectral details. Learning phonemic labels for these non-speech sounds did not result in a change in the results. These findings along with previous results with non-human animals suggest that much of "compensation for coarticulation" may be accomplished by general perceptual mechanisms operating on auditory representations.