Moving beyond the monosyllable in models of skilled reading: Mega-study of disyllabic nonword reading

Most English words are polysyllabic, yet research on reading aloud typically focuses on monosyllables. Forty-one skilled adult readers read aloud 915 disyllabic nonwords that shared important characteristics with English words. Stress, pronunciation, and naming latencies were analyzed and compared to data from three computational accounts of disyllabic reading, including a rule-based algorithm (Rastle & Coltheart, 2000) and connectionist approaches (the CDP++ model of Perry, Ziegler, & Zorzi, 2010, and the print-to-stress network of Ševa, Monaghan, & Arciuli, 2009). Item-based regression analyses revealed orthographic and phonological influences on modal human stress assignment, pronunciation variability, and naming latencies, while human and model data comparisons revealed important strengths and weaknesses of the opposing accounts. Our dataset provides the first normative nonword corpus for British English and the largest database of its kind for any language; hence, it will be critical for assessing generalization performance in future developments of computational models of reading.