effet du nombre des graphèmes en Anglais - Aix Marseille Université

Appendice I 227reading system is interpreted as an interactive, dynamical system 4 (Grossberg & Stone, 1986 ; Korte, 1923 ;Rumelhart, 1977 ; Stone et al., 1997 ; Van Orden & Goldinger, 1994), models, methods, and measuresmust be developed that adequately reflect and help analyze the functioning of such a complex nonlinear system.According to our view (e.g., Ziegler, Van Orden, & Jacobs, in press b), reading is a two-way system :Phonological information and phonological skills influence orthographic processing, and orthographic informationand orthographic skills also influence phonological processing (Jakimik, Cole, & Rudnicky,1985 ; Dijkstra, Roelofs, & Fieuws, 1995 ; Wagner and Torgesen, 1987). From this perspective, singletaskapproaches to reading, measures of orthographic neighborhood (Coltheart, 1978), metrics of spelling-tosoundconsistency (Rosson, 1985 ; Treiman, Mullennix, Bijeljac-Babic, & Richmond-Welty, 1995 ;Venezky & Massaro, 1987), or monotask models of performance must necessarily remain incomplete approximations.The study of bidirectional, two-way consistency effects (Stone et al., 1997 ; Ziegler et al.,1996, in press a ; Ziegler, Montant, & Jacobs, in press c) which we will discuss below, and our present attemptto model them represent a step beyond this one-way approach to reading.A PRINCIPLED APPROACH TO COGNITIVE MODELINGIt is useful to note here that our approach to understanding the reading process by help of formal cognitivemodels follows a set of pragmatic stratagems and principles that are outlined in several recent papers(Grainger & Jacobs, 1996 ; Jacobs, 1994 ; Jacobs & Grainger, 1994) and further discussed in Grainger andJacobs (present volume). The most relevant stratagem for the present paper is nested modeling, that is theidea that a new model should either include the old one as a special case by providing formal demonstrationsof the inclusion, or dismiss with it, after falsification of the core assumptions of the old model. The developmentof our M ultiple R ead- O ut M odel (MROM) gives a detailed example of nested modeling in the domainof orthographic processing in lexical decision and perceptual identification tasks (Grainger & Jacobs,1996). In the present paper, we further pursue our efforts concerning nested modeling of visual word recognitionby including elementary p honological processes into the MROM that thus becomes the MROM-P.SUBJECT AREAIn this section two empirical phenomena that are considered as evidence for phonological and bidirectionalinfluences on visual word recognition, the pseudohomophone effect and the bidirectional consistencyeffect, are used as empirical touchstones to test the MROM-P.THE PSEUDOHOMOPHONE EFFECTThe first phenomenon is the classical pseudohomophone effect. As regards the LDT, it refers to the observationthat nonwords that sound like words when read aloud (e.g., BRANE) are more difficult to rejectthan nonpseudohomophonic control stimuli (e.g., FRANE ; Rubenstein, Lewis, & Rubenstein, 1971).Since the precursor of our present model, the MROM (which is briefly discussed in the next section) doesnot include any phonological processes, it should not be able to simulate the pseudohomophone effect in theLDT, if the effect is a genuine phonological one. Thus, the first critical test for the MROM-P is to evaluateits ability to capture the pseudohomophone effect. As a testing ground, we chose the classical set of data byColtheart, Davelaar, Jonasson, and Besner (1977), and a set of data from a more recent replication study bySeidenberg, Petersen, MacDonald, and Plaut (1996).THE FEEDFORWARD AND FEEDBACK CONSISTENCY EFFECTSThe second phenomenon, only very recently discovered, provides perhaps the strongest experimental evidencefor an interaction between orthographic and phonological processes in visual word recognition. Itcombines two effects : the traditional feedforward consistency effect and the newly discovered feedback consistencyeffect (Stone et al., 1997 ; Ziegler & Jacobs, 1995 ; Ziegler et al., in press c).Effects of spelling-to-sound (feedforward) consistency have been studied extensively in the naming task,i.e., a task requiring overt pronunciation. The standard result is that naming latencies are longer and/or spellingerrors more frequent for inconsistent words that have multiple spelling-to-sound mappings than for consistentwords whose spelling bodies are always pronounced the same. Thus, an inconsistent word whosebody has several possible pronunciations like _OUGH in COUGH, DOUGH, THROUGH, BOUGH,TOUGH will be harder to pronounce than a consistent word like DUCK that has a unique spelling body(_UCK). The known conditions favorable to obtaining this effect (see Ziegler et al., in press c) include alphabeticwriting systems (e.g., absence of consistency effects in a logographic writing system such as JapaneseKanji ; Wydell, Butterworth, & Patterson, 1995), inconsistent words of low frequency (Andrews, 1982; but see Jared, 1995), and words with a consistency ratio smaller than .5 (i.e., the ratio given by the4 A view now shared by extant former representants of traditional modular, noncomputational, feedforwardmodels of the reading process (e.g., Coltheart et al., 1993 ; Coltheart & Rastle, 1994 ; Norris, 1994 ; for adifferent view see Massaro & Cohen, 1994).