ASR in a human word recognition model: generating phonemic input for shortlist

The current version o f the psycholinguistic model o f human word recognition Shortlist suffers from two unrealistic con­ straints. First, the input o f Shortlist must consist o f a single string o f phoneme symbols. Second, the current version o f the search in Shortlist makes it difficult to deal with insertions and deletions in the input phoneme string. This research attempts to fully automatically derive a phoneme string from the acoustic signal that is as close as possible to the number o f phonemes in the lexical representation o f the word. We optimised an Automatic Phone Recogniser (APR) us­ ing two approaches, viz. varying the value o f the mismatch pa­ rameter and optimising the APR output strings on the output of Shortlist. The approaches show that it will be very difficult to satisfy the input requirements o f the present version o f Shortlist with a phoneme string generated by an APR.