| The extant literature in visual word recognition has been largely dominated by the study of monosyllabic words in factorial experiments, computational models, and most recently, megastudies. Although this literature has accumulated a rich description of behavioral effects for monosyllabic words, the extent to which these results generalize to more complex multisyllabic words remains unclear. This dissertation presents a large-scale study of pronunciation and lexical decision latencies for 9,639 monomorphemic mono- and multisyllabic words. Through hierarchical regression techniques, the influence of standard variables like phonological onsets, length, orthographic neighborhood size, phonological neighborhood size, and word frequency were explored. Importantly, the unique influence of other variables, including: (a) syllable length, (b) syllabically-defined feedforward (spelling-to-sound) and feedback (sound-to-spelling) phonological consistency, (c) new orthographic and phonological similarity measures, (d) stress typicality, and (e) semantics were also investigated. Using the full set of predictors, it was possible to account for substantially more variance in both pronunciation (65.9%) and lexical decision (65.6%) performance than any previous large-scale study. The results are consistent with a highly interactive, cascadic word recognition system where semantics exert early top-down effects. More importantly, the present findings provide a unified, well-specified set of benchmark phenomena which can be used for constraining emerging models of multisyllabic visual word recognition.;The secondary objective of the dissertation was to examine the role of individual differences in word recognition performance. Word recognition studies and models typically average performance across participants, despite strong theoretical reasons to expect between-participant variability. Using the large-scale databases afforded by the English Lexicon Project, the reliability and prevalence of individual differences were examined. The results indicate that over two test sessions, there was surprisingly high test-retest reliability in word recognition performance. Specifically, reliability was demonstrated in both the distributional characteristics of response time data, as well as in the extent to which participants were influenced by different underlying lexical dimensions. The analyses also considered the relationships between lexical processing strategies, word recognition performance, and verbal proficiency. Interestingly, participants with higher verbal proficiencies were associated with lexical processing strategies that were better calibrated to the specific demands of the task at hand. |
