Context-conditioned generalization in adaptation to distorted speech

People were trained to decode noise-vocoded speech by hearing monosyllabic stimuli in distorted and unaltered forms. When later presented with different stimuli, listeners were able to successfully generalize their experience. However, generalization was modulated by the degree to which testing stimuli resembled training stimuli: Testing stimuli's consonants were easier to recognize when they had occurred in the same position at training, or flanked by the same vowel, than when they did not. Furthermore, greater generalization occurred when listeners had been trained on existing words than on nonsense strings. We propose that the process by which adult listeners learn to interpret distorted speech is akin to building phonological categories in one's native language, a process where categories and structure emerge from the words in the ambient language without completely abstracting from them.