Does masked onset priming provide evidence for serial left to right letter to sound conversion?

When a target word is preceded by a masked prime which has the same onset as the target, naming is faster than when a different onset prime precedes the target. This effect is specific to the onset. Overlap at a later word position does not produce priming (Forster & Davis, 1991; Kinoshita, 2000). This has been explained as resulting from left to right grapheme to phoneme conversion (Forster & Davis, 1991; Coltheart, Woollams, Kinoshita, & Perry 1999). If this is so then priming is expected in tasks which require grapheme to phoneme conversion but do not involve a spoken response. This prediction was tested using a pseudohomophone decision task, which requires participants to indicate whether a non-word target sounds like a word. Masked onset priming was not found in this task, but was found in naming using the same materials, and priming was found in pseudohomophone decision where the words from which pseudohomophones were derived were used as primes. This indicates that the onset specificity of masked onset priming is not due to a serial process in reading, but arises as a consequence of a beginning of word to end of word process in the preparation of a spoken response.