INTRODUCTION
Levels of Processing in Visual Word Recognition

Visual word recognition is a complex process that in-volves several cognitive operations, such as visual encod-ing of letters, translation of the letters’ shapes into asequence of graphemes and orthographic patterns, andactivation of lexical/phonological structures and theirmeanings. All these processes have been shown to beinvolved in reading words through many experimentswith normal subjects as well as by neuropsychologicalinvestigations of patients with different types of dyslexia.For example, the importance of visual processing forword recognition was highlighted by patients with ne-glect dyslexia, who have difficulty identifying letterswhile keeping track of their order in the word (e.g., Ellis,Flude, & Young, 1987) and by patients with attentionaldyslexia who correctly identify the letters while misplac-ing them within or across words (e.g., Shallice & War-rington, 1977). The need for efficient orthographicintegration is demonstrated by patients with simultag-nosia who are letter-by-letter readers (e.g., Patterson &Kay, 1982). The phonological dyslexia syndrome indi-cates that reading without phonology is deficient and, inconjunction with surface dyslexia and deep dyslexia,demonstrates the importance of the lexical access fornormal reading. (For a detailed discussion of these syn-dromes, see Coltheart, Patterson, & MArshall, 1980, andPatterson, MArshall, & Coltheart, 1985.) On the basis ofsuch evidence, the model used as a framework for thepresent study posits that visual word recognition in-volves different levels at which printed information isprocessed. These levels are (1) an orthographic level atwhich visual features are integrated to represent ortho-graphic patterns, (2) a lexical level at which thephonological (and possibly the whole-word ortho-graphic) representation of the printed word is activated,and (3) a semantic level at which the meaning of theword is accessed. In addition, tasks in which attention isdirected to the phonetic features of the words (such asrhyming judgments) may induce phonetic activity thatmay or may not be involved in the word recognitionprocess.Although the exact nature of the processes involvedin visual word recognition is still a matter of debate, thenotion of levels (either of processing or of repre-sentation) is accepted and incorporated into most theo-ries (e.g., Ellis & Young, 1996, ch. 8; McClelland &Rumelhart, 1981; Seidenberg & McClelland, 1989). More-over, there is evidence that the level at which a word isprocessed is task-dependent and can be controlled. Forexample, several studies did not find semantic primingwhen the prime was processed at a letter level, at leastif the stimulus onset asychrony (SOA) between theprime and the target was longer than a few hundredmilliseconds (e.g., Henik, Friedrich, & Kellogg, 1983;Smith, Theodor, & Franklin, 1983). In fact, semantic prim-ing was absent even at a short SOA (200 msec) if theprime task was letter-search (Henik, Friedrich, Tzelgov, &Tramer, 1994). This result suggests that it is possible tocontrol the putatively automatic activation of the seman-tic system by directing subjects’ attention to the letterlevel. Other studies, however, challenged this interpreta-tion, suggesting that the absence of priming in the Heniket al. (1994) experiment was an epiphenomenon causedby the difficulty of the task, which prevented the activa-tion of the prime’s semantic representation within theshort SOA time range. When easier letter-level tasks wereused, semantic priming was obtained, suggesting that theactivation of meaning, although not resource-free, is thedefault in visual word perception (Smith, Bentin, &Spalek, submitted). It appears, therefore, that the ques-tions of whether the processing of printed words maybe restricted to a shallow level and whether the cogni-tive system involved in visual word recognition can beinfluenced and shaped by the purpose of reading thewords are still open. A related question addresses theword-related information that is processed at each level.For example, although traditional models of word recog-nition assume the existence of a mental lexicon in whichword-related information is represented (but see Hinton& Shallice, 1991, and Seidenberg & McClelland, 1989, foralternative views), there is no consensus regarding thecharacteristics of this representation. According to somemodels, the lexicon contains only structural word-relatedinformation (phonologic and orthographic), whereasothers see no evidence requiring separation betweenthe structure of the word and its meaning.An additional, major question concerning the levels ofprocessing printed words is how the level-specific proc-esses interact among themselves. One traditional viewsuggested a series of stages. Accordingly, the printedword should be processed first at the orthographic level.The output of this stage addresses a visual lexicon acti-vating a word pattern and, subsequently, its semanticrepresentation (e.g., Morton, 1969). Other models sug-gested that the various visual word perception opera-tions are exerted in “cascade;” that is, a processing stagecan begin before the previous stage is finished (McClel-land, 1979). More recent models of reading suggest par-allel, interactive processes by which the visual stimulusis processed in parallel at all levels and different wordsare represented by different patterns of activity in aneural network (Carr & Pollatsek, 1985; Coltheart, 1985;McClelland & Rumelhart, 1981; Seidenberg & McClel-land, 1989; see also Jared & Seidenberg, 1991). Investi-gating the dynamics of visual word recognition has beenpartly hampered by the difficulty of disentangling proc-esses by the use of discrete measures of performancesuch as the reaction time (RT). Some of these impedi-ments can be overcome by studying the neurophys-iological mechanisms that subserve this cognitivefunction. In addition to providing ways to distinguishbetween cognitive mechanisms by relating them to theneuroanatomically distinct structures that mediate them,some neurophysiological measures (such as ERPs) pro-vide an on-line and time-continuous index of processing.