Word perception and misperception in context.

Abstract

When reading lists of words and nonwords at 100 ms/word, Ss reported words accurately but frequently converted nonwords such as dack into similarly spelled words such as duck or deck. In sentences, both nonwords and anomalous words were misread as appropriate words, but the bias was greater for nonwords. Word associations in lists (e.g., sailor-dack-vessel) produced a similar bias, but when sentence meaning was pitted against such associations the lexical effect was largely overridden. Sentences in which biasing context appeared only after the critical item reduced but did not eliminate the context effect, suggesting that multiple word candidates remained active while at least the next 3 words were processed. These results support a 2-stage modular interactive model: The first stage is stimulus driven and emits multiple weighted candidates that are combined interactively with contextual information in a second stage. In reading or listening to discourse, up to five or six words may be perceived and understood per second: a remarkable cognitive feat. Although one might think that processing language simultaneously at several levels—discriminating letters or phonemes, identifying words, parsing and interpreting sentences, and constructing a representation of the discourse—would be more difficult than focusing attention on just the lowest one or two levels, the opposite has been shown to be the case. In general, letter perception is more accurate when the letters are embedded in words (the word superiority effect, Reicher, 1969; Wheeler, 1970; see Estes & Brunn, 1987, for a recent reconsideration), and word perception is more accurate when the words are embedded in sentences (e.g., for speech; Miller, Heise, & Lichten, 1951). Yet the detailed mechanisms underlying the positive effects of higher level context on word perception have been a matter of theoretical and experimental debate since the earliest days of experimental psychology. The debate centers on the way in which sensory-driven, bottom-up processing is combined with background knowledge. Everyone agrees that by the time a listener or reader has reached a final interpretation of a word, sentence, or text, both lower level and higher level knowledge will have been taken into account. The issue is when, during processing, various sorts of knowledge become available. In the present study, the question is how sentence context influ