ABSTRACT
Oral reading of new words requires assembly of phonology from constituent parts. One view of this operation involves lexical analogies between items sharing an orthographic segment, an operation which may be constrained by the size of the shared segment, its position in the string, its frequency of occurrence in English words, and the frequency of occurrence of those words.
Extant “analogy” models are not capable of predicting the outcome of assembly operations for all possible strings. However, using nonwords for which pronunciation options are limited and relevant neighbours specifiable, the analogy view can be forced to predict the probability of each optional pronunciation. Such probabilities can also be estimated empirically.
In four experiments, subjects were asked to read aloud nonwords for which pronunciation variability was confined to two contiguous letters: CC or OU. The dependent variable was the frequency across subjects of each major pronunciation variant for each stimulus type. Approximately 1250 subjects were tested; each pronounced only 1 string. No model based strictly on the orthographic parameters mentioned above can account for the data. A model in which rules produce phonological “frames” which guide search through a phonological lexicon is more successful in explaining the results.
