Abstract
Two years after the initial investigation of a Japanese case with a mixed fluent/non-fluent progressive aphasia, the patient still had a relatively pure language disorder which had deteriorated sharply. Her spontaneous speech at follow-up consisted almost entirely of repetitive jargon; but the single-word speech production tasks of object naming and reading aloud revealed two striking phenomena: (i) she sometimes produced the first one or several syllables of a target word correctly before tailing off into jargon; (ii) both the occurrence and the size of these correct phonological fragments were task-related, following the gradient of kana word reading < kanji word reading > picture naming. Differential performance in the three tasks is attributed to differences amongst the three stimulus types in the degree to which they provide constraints specifying the correct phonological representation.