Whether the process of lexical access in silent reading is mediated by an internal phonological representation or only by visual representations has been a topic of empirical investigation since before the turn of the century. To date, research has generated contradictory findings. More recent evidence suggests that factors such as difficulty of material, frequency of occurrence of items, subjects' fluency, and task demands-which are seldom manipulated in typical studies of word recognition-play important roles in determining the type of mediation used. As a result, interest has shifted to a family of dual access models in which both phonological mediation and visual mediation are ordinarily used in parallel. The data point to a dual access model in which high-frequency words enjoy highspeed access via a visually based representation, whereas low-frequency words are accessed using a slower, phonological receding process.

Predictions for word recognition latencies were generated for the independent parallel model of word recognition based on letter-naming latencies in two display conditions. In one condition, the inside two letters of four-letter words were presented 50 msec in advance of the presentation of the whole word, whereas in the other condition, the outside two letters were presented 50 msec in advance. The model predicts that the two conditions should yield roughly equal recognition latencies, but in each of four experiments the prior presentation of the outside letters led to faster recognition. The implications of these results for the family of parallel models (nonindependent as well as independent) are discussed.

The phonological recoding model of lexical access was tested in two experiments. In Experiment 1, college students were presented words and nonwords at recognition threshold for lexical decision. Nonwords homophonous with real words (homophonous nonwords] and nonwords nonhomophonous with real words (nonhomophonous nonwords) were used. The phonological recoding model predicts more errors on homophonous nonwords as a result of false matches in the subjects' internal lexicons. Blocks of items with homophonous nonwords led to significantly poorer performance as measured by percent correct and d'. Experiment 2 was a replication of Experiment 1 with the addition of a spelling test to eliminate poor spellers. The d' measure again supported the phonological recoding model. These data were compared to those of Stanovich and Bauer (1978) which led to a conclusion against phonological recoding. This discrepancy was considered in light of the task differences, and the presence of a mask was thought to be important. The possible masking effect of successive fixations in reading was considered.The spoken and written stimuli which comprise an alphabetic language have, for the most part, no inherent meanings. Contrary to this view is the theory of phonetic symbolism (see Taylor, 1963), which maintains that the assignment of phonetic symbols to referents is not arbitrary but purposeful. Thus, the word god represents its referent because the word's acoustic and articulatory properties (e.g., deepness, roundness) parallel the worldly properties of its referent. There exist a few words which exhibit what might be called orthographic symbolism. The word bed, for instance, looks like it has been appropriately paired with its referent. But as this does not seem to be true of refrigerator or ¡ove or somewhat or the vast majority of English words, readers cannot depend on orthographic symbolism as a means of word recognition, just as listeners cannot depend on phonetic symbolism for speech perception. Thus readers can only arrive at the meaning of a written word by mapping the print onto the entries in their mental lexicons. This "mapping problem" has been called "the fundamental problem of reading," (Gough, 1972, p. 335), and is the concern of this paper.at

Guttentag's two questions about our review of the phonological receding literature are addressed. Following Guttentag's suggestion, we now consider the possibility of phonological mediation without the use of grapheme-phoneme correspondence rules. It is recognized as a logically possible supplement to phonological receding, but no empirical support is discovered. Next, we address Guttentag's discussion of how dependence on phonological receding might change with increased age and reading skill.

No abstract