In most alphabetic systems the mapping between spelling and sound is quasi-systematic. Many graphemes have multiple pronunciations, among which one is considered as regular. One claim of most dual-route models of reading compared to connectionist models is that the knowledge of print-to-sound correspondences is restricted to rules encoding the most frequent phonemic mapping of each orthographic unit. This assumption was tested in two experiments. In the first experiment, a replication and extension of Ziegler & Jacobs (1995) letter detection experiments, we examined whether the soft and hard pronunciation of G and C are activated. We compared performance on letter-absent pseudohomophones (J in GEUDI) and control pseudowords (J in BEUDI) with letter-absent "false pseudohomophones" (J in BONGOUR) and their respective controls. Longer detection times were observed for G items compared to their orthographic controls for both real pseudohomophones ans false pseudohomophones suggesting that the contextually appropriate and inappropriate pronunciations are activated. In a second experiment, we examined naming latencies for pseudowords varying in grapheme frequency and grapheme entropy (an uncertainty measure capturing the degree of predictibility of grapheme-phoneme correspondences). A significant effect of both variables was found, demonstrating the graded nature of the representations in the print-to-sound conversion system.
Versions of the submitted paper are available on-line:
Full version (HTML)