In two experiments, we examined whether context information can affect the activity of the nodes at the character level. Chinese readers viewed two Chinese characters; one was intact, but the other (the target) was embedded in a rectangle of visual noise and increased in visibility over time. The two characters constituted a word in one condition but did not in the other condition. The task was to press a button to indicate whether the character in the noise was at the top or bottom of the rectangle. (They did not have to identify the character.) Response times were faster in the word condition than in the nonword condition. Because the "wordness" of the stimulus was logically irrelevant to judging the location of the target character, the data indicate that processing at the word level can feed back to fairly low-level judgments, such as where a character is.
Keywords Word recognition . ReadingModels of word processing in reading usually assume that the processing has multiple levels: a feature level, a character (or letter) level, and a word level (Massaro, 1998;McClelland & Rumelhart, 1981). In all such models, the feature level encodes visual feature information that feeds into character (letter) identification, which in turn feeds into word identification. However, word superiority effect (WSE) studies have demonstrated that character identification is facilitated (even when guessing is controlled for) when the character is part of a word (Reicher, 1969;Wheeler, 1970) as compared with when it is embedded in a series of nonword letters or when it is shown in isolation. Such a phenomenon clearly casts doubt on whether word recognition is merely the result of the feed-forward mechanism sketched previously.Perhaps the most influential current model of word recognition is the interactive activation model (IAM, McClelland & Rumelhart, 1981), which assumes that word processing is an interactive process. (The IAM model was largely proposed to explain phenomena such as the WSE.) Through the interconnections between the nodes in a threelevel network (feature level, character level, and word level), the activity of a node can affect the activity of the nodes in the same level and the other levels, including "lower" levels. Most notably, to explain the word superiority effect, IAM assumes that the activity of the nodes at the character level is affected by the activity of nodes at the word level. The activation of a character node increases faster when the character is part of a word (or even part of a pseudoword) than when it is not. Hence, a character belonging to a word is identified faster than when it is not part of word or when it is shown in isolation.However, not all of the models of word processing that can explain the WSE assume that context affects the activity in the character level. For example, the fuzzy logic model of perception (FLMP;Massaro, 1998;Massaro & Cohen, 1991) assumes that context information in the word level does not feed back to the character level and affects the activity of relat...