The Generalization of Exceptional Knowledge in Word Pronunciation Tasks

Abstract

Models of word pronunciation have tended to emphasize either generalized knowledge in the form of letter-to-sound correspondence rules or item specific knowledge in the form of rote associations. Simple formulations of both types of models have been found to be clearly insufficient to account for the pattern of results obtained with three types of items: regular words, exceptional words and pseudowords. The general findings are: 1-pseudowords take longer for response initiation than words; 2-although slower, pseudowords are pronounced quite easily by most readers; 3-exception words take longer for response initiation than regular words. Even a dual-process formulation, based on item specific knowledge for some type of items and generalized correspondence rules for other items, fails to account for some of the differences in pronunciation latencies which have been observed between regular and exceptional words. Glushko (1979a) has proposed that the regular-exceptional distinction should be replaced by a consistent-inconsistent distinction. The thrust of his argument is based on his finding that regular words like MINT have the same pronunciation latencies as exceptional words like PINT. Hence when one controls for a certain type of similarity between words, there is no difference between regular and exceptional words. However, a difference is found between words whose final three letters share a consistent pronunciation {such as MINK and PINK) and words whose terminal letter groups have more than one pronunciation {such as MINT and PINT). Glushko's model, the activation-synthesis model, is based on the elimination of the regular-exceptional distinction. On the basis of this elimination, it is possible to make a number of extensions to the activation-synthesis model. Empirical verifications of these potential extensions were attempted using manipulations intended to increase or decrease the amount of conflict present when target items were presented for pronunciation. Inconclusive results were obtained from a first manipulation which used a repeated list paradigm. A second manipulation revealed that it was possible to speed up the pronunciation of a pseudoword by priming with regular words but not by priming with exceptional words. However, priming with exceptional words increased the number of exceptional pronunciations which were emitted for the subsequent pseudoword. The pattern of results is consistent with the conclusion that the regular-irregular distinction should be maintained and superimposed on the consistent-inconsistent distinction. The argument is made that the inconsistency effect may be due to the activation of inconsistency detectors by words which contain specially coded letter groups. It is further suggested that the activation of inconsistency detectors would modify the usual response generation or retrieval process. This modified process would allow access to the exceptional information which would be somewhat resistant to the usual generalization phenomena observed in word pronunciation. Suggestions are also made as to the nature of the cues which could activate the inconsistency detectors.