Neighborhood and Position Effects Interact in Naming Latency

Naming latency studies have recently shown a position-ofirregularity effect (words with early irregularities seem slowed compared to those with late irregularities), for which DualRoute models of reading can account. Milostan & Cottrell (1998) showed that the initial studies contained a confound between irregularity position and friend/enemy ratio, and that the statistical confound could be captured by connectionist networks which then show the supposed position effect. This paper presents work to disentangle the position/regularity confound through a subject study and additional connectionist explorations. The latency data show that, once friend/enemy ratios are controlled for, the supposed position effect is driven entirely by high-enemy words in the first position. Further, connectionist network simulations show that network error at the first phoneme position only is a better match for naming latency, while overall network error produces a better match to subject error counts.