effet du nombre des graphèmes en Anglais - Aix Marseille Université

258Appendice IIIHowever, an unsolved question remains with respect of both models : how does the reading system parsethe letter sequence into graphemes ? The dual-route model suggests that a set of rule correspondances is usedfor the graphemic segmentation. The remaining question here is : which representational format or cognitivecode does the reading system use to implement this set of rules given that rules belong mainly to high-levelcognitive processes and seem less plausible for fast and automatic perception processes ? Similarly, Plaut etal. (1996) did not specify how does the reading system perform the graphemic segmentation. In their model,input letter strings are parsed into onset-nucleus-coda graphemes by turning on the corresponding graphemenodes. Thus, it appears that these authors implicitly assume that graphemes are basic units of the readingsystem which leads paradoxically to the conclusion that graphemes are considered in their model as localrepresentations (Grainger &Jacobs, 1998a).A solution to the graphemic segmentation problem can be found in a boxological model proposed byLaberge and Samuel (1974). These authors suggested that during the acquisition of reading, intermediate readingunits (units situated between the letter and word levels) are developped within the reading system. Thefunctional role of these units would be to increase the automaticity and rapidity of skilled reading and to allowprogressively the processing of letter strings as wholes. The establishment of these intermediate units wouldlikewise be a manner to push the reading system in the direction of an increased parallel processing of writtenwords. The idea that graphemes could be such intermediate units was already suggested by Gibson, Pick,Osser, & Hammond (1962). Another argument in favour of this assumption comes from the fact that graphemeshave a functional role with respect to phonological units. Graphemes could thus be viewed in suchcontext as phonographic reading units (Peereman & Content, 1997).How does the « reading unit » hypothesis solve the graphemic segmentation problem? If one assumesthat the reading system develops representations of intermediate reading units, then the functional segmentationof a letter sequence could be done by activating a set of matching intermediate reading units. This solutionis implicitly used in the Plaut et al. model (1996) despite that these authors do not explicitly assume thatgraphemes are reading units or localist representations of the reading system. Also, this « reading unit »solution could be a more functionally plausible alternative in order to perform the grapheme-to-phonemeconversion in the Coltheart et al. model (1993).The purpose of the present study is to test experimentally the assumption that graphemes are processed asperceptual units by the reading system. Indeed, the psychological validity of graphemes has been assumed bymany models of word recognition despite only few empirical data are available to support such assumption.As a window to explore unitization processes, a large number of studies used the letter detection procedure (fora review, see Healy, 1994). The preliminary idea in these manipulations is that the reading system processparticular groups of letters as wholes or reading units. Therefore, searching for a letter embedded in a readingunit appears costly since subjects have to split the unit into its constituent letters in order to perform thetask. Letter detection studies reported, for example, missing detections for letters embedded in high frequencyshort words such as THE (the target letter being E). Since these words are probably processed as wholes bythe reading system, they consequently seem to mask the identification of their constituent letters (e.g., Corcoran,1966 ; Drewnowski & Healy, 1977).We followed the same logic in the present study. Indeed, if graphemes are processed as units by the readingsystem, then detecting a letter embedded in a multi-letter grapheme should be harder than detecting a letterbeing a single-letter grapheme itself. As an illustration, searching for the letter A in the word BEACH (Abeing embedded in the multi-letter grapheme EA -> /i/) should be harder than searching for A in the wordGRASS (A being here a single-letter grapheme).We tested this hypothesis in two experiments done in English (Experiment 1) and French (Experiment 2).In both experiments, we used the same experimental design. Subjects were presented a target letter followedby a word and had to decide whether or not the letter was in the word. In present trials, letters were eitherembedded in a multi-letter grapheme (e.g., A in BEACH) or were equivalent to a single-letter grapheme (e.g.,A in GRASS). In addition, we performed a frequency manipulation of the words. In half of the graphemeconditions, words were of high frequency and for the other half, words were of low frequency. This manipulationwas intended to test whether unitization is pre-lexical as suggested by the reading unit assumption. Theprediction in this case is that if there is any evidence of a pre-lexical unitization procedure, then letter detectionlatencies should be independant of the lexical status of the target word.EXPERIMENT 1ParticipantsParticipants were 21 psychology students from Macquarie University. All were native Australian-Englishspeakers and had normal or corrected to normal vision.Stimuli and ApparatusThe stimulus set was composed of 60 target-present trials and 60 target-absent trials. The 60 target-presenttrials were divided into four lists of 15 monosyllabic, 5-letter English words (see Appendix A). Each list wasmatched between items for the target letter and for its position in the word. Two lists were composed of lowfrequencywords (F < 10 occurences per million) and two lists of high frequency words (F > 50 occurences permillion). Each frequency list was matched between items for word frequency. Frequency was estimated usingthe CELEX frequency count (Baayen, Piepenbrock, & van Rijn, 1993). For one list of words in each fre-