effet du nombre des graphèmes en Anglais - Aix Marseille Université

260Appendice IIIgrapheme compared to letters having the status of single-letter graphemes. However, this effect was marginalfor subjects, F1(1,18) = 3.34, .05 < p .1.Finally, error scores were not affected by word frequency, all Fs < 1.DISCUSSIONThe results of the present studies can be summarized as follows. First, in two letter search experiments,we obtained longer response times when the target letter was embedded in a multi-letter grapheme (A inBEACH) compared to when it was itself a single-letter grapheme (A in GRASS). This effect appeared both inthe English and French data. Second, letter detection latencies were not affected by the frequency of the targetword neither in the English nor in the French data.The present results support the view that graphemes are automatically processed by the reading system asperceptual units. Therefore, accessing the letter A in BEACH appears more costly than accessing A inGRASS. Indeed, in BEACH, A is embedded in the multi-letter grapheme EA that seems to be processed as awhole by the reading system. That is, the identification of A is masked by the automatic processing of thehigher-order unit EA. Following this interpretation, the locus of the present effect can be considered as prelexical.That is, it results from a competition between a graphemic and a letter level of representation. Afurther argument in favor of such interpretation comes from the absence of any word-frequency effect. Indeed,lexical factors appear to not influence letter detection latencies.However, one may propose an alternative description of these results in terms of phonemic mismatch.Indeed, one may argue that the sound of letter A is closer to the sound of A in GRASS than to the sound of Ain BEACH. Thus, the difference in letter detection latencies would be due to phonemic distance (for a reviewon the role of phonetic factors in letter detection, see Healy, 1994). Therefore, the locus of the present effectwould not be perceptual but rather, post-lexical. That is, subjects would compare the name of the target letterto the phonemic representation of the word, and because of the phonemic distance of the letter’s name and theletter’s sound in the word, they would need more time to generate a present-response for letters embedded in amulti-letter grapheme. However, this post-lexical phonemic interpretation is not congruent with the absenceof a word frequency effect. Indeed, if the phonemic representation of a word is crucial for letter detection, thenthere should be an advantage of high frequency over low frequency words in the multi-letter grapheme condition,since the phonemic representation of high frequency words should be accessed more rapidly. Consequently,given the absence of a word-frequency effect on letter detection latencies, the present data do not supportthe post-lexical phonemic interpretation.Another empirical evidence underlying the critical role of graphemes during reading comes from a studydone by Pring (1981). This author used the pseudohomophone effect for her demonstration (Rubenstein,Lewis, & Rubenstein, 1971). The pseudohomophone effect is the fact that, in a lexical decision task, whennonwords are constructed to be pronounced like words (e.g., CHERCH), participants are slower to reject theseitems (i.e., pseudohomophones) compared to spelling controls (e.g., CHIRCH). Pring showed that this effectdisappeared when the graphemes in the stimulus were disrupted through case alternation (i.e., the pseudohomophoneeffect disappeared for CheRcH where the graphemes are disrupted, whereas the effect remains forCHerCH, where case alternation does not split graphemes).Grouping letters into graphemesFollowing the assumption that graphemes are perceptual units leads to the conclusion that some letters areautomatically grouped into multi-letter graphemes during word processing. This grouping process is, in fact,highly functional to perform an efficient orthography-to-phonology computation since it allows to retrievethe correct sequence of phonemes (which would not be the case if the unit of the reading system was the letter).However, recent studies showed that the presence of multi-letter graphemes in a word seem to slow downits processing. In a non-word reading experiment, Rastle and Coltheart (1998) reported faster naming latenciesfor non-words composed of 5 graphemes and 5 letters compared to non-words composed of 3 graphemes and 5letters. Given that the number of letters was constant, 3 graphemes non-words were thus composed of multilettergraphemes which was correlated with longer naming latencies. In a similar manipulation, Rey, Jacobs,Schmidt-Weigand and Ziegler (1998) obtained longer identification times for words having a smaller numberof graphemes (the number of letters being constant). This effect was observed in English and French for lowfrequency5-letter monosyllabic words, but was not obtained in French for high-frequency 5-letter monosyllabicwords. Together, these results indicate that grouping letters into graphemes, despite being automatic andfunctional, requires to avoid or inhibit a non-functional letter-by-letter processing.Reading unitsThe present data support the view according to which graphemes can be considered as minimal functionalreading units. This reading unit assumption is congruent with a model proposed by Laberge an Samuel (1974)in which word processing is mediated from letters to words through different levels of spelling units. However,Laberge and Samuel did not specify the nature of these units. Also, the present data suggest that graphemesmay be represented in this framework at an early level of representation. Furthermore, a large set ofempirical data recorded in different langages (Englsih, Spanish or French) indicates that larger functional spellingunits may also be represented in the reading system such as onset, rimes or syllables (for onset/rime