On the basis of the good predictions for phonemes correct, we conclude that closed-set feature identification may successfully predict phoneme identification in an openset word recognition task. For word recognition, however, the PCM model underpredicted observed performance, and the addition of a mental lexicon (ie, the SPAMR model) was needed for a good match to data averaged across 7 adults with CIs. The predictions for words correct improved with the addition of a lexicon, providing support for the hypothesis that lexical information is used in openset spoken word recognition by CI users. The perception of words more complex than CNCs is also likely to require lexical knowledge (Frisch et al, this supplement, pp 60-62). In the future, we will use the performance of individual CI users on psychophysical tasks to generate predicted vowel and consonant confusion matrices to be used to predict open-set spoken word recognition.
ASJC Scopus subject areas