Masked phonological priming effects in English - Center for Reading ...

More documents

Recommendations

Info

110 K. Rastle, M. Brysbaert / Cognitive Psychology 53 (2006) 97–145Table 6Mean effect sizes and 95% confidence limits for masked phonological priming effects reported in five experimentalparadigmsTask Mean effect size (r) .05 Lower limit .05 Upper limitForward-masked perceptual identification .240 .116 .356Backward-masked perceptual identification .193 .020 .354Forward-masked reading aloud .312 .158 .451Forward-masked lexical decision .204 .057 .341Text reading .234 .036 .472rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi4:385r ¼¼ .433.4:385 þ 19The same value would be obtained for a study in which the difference between the phonologicaland graphemic control conditions is reported as F(1,19) = 4.385. When the data werenot reported as a main effect between the two conditions but as an omnibus test involvingmore conditions, the F contrast was calculated as described in Rosnow and Rosenthal (1996,pp. 336–337) and used to calculate the effect size. Finally, when no further indication was givenother than F < 1, we used values of r = +.05 for a trend in the expected direction andr = .05 for a trend in the opposite direction. Standardized effect sizes were calculated onlywhen there were independent samples of participants. Studies were weighted for their numbersof participants to calculate their overall effect size and confidence limits.Table 6 reports the average effect sizes for the five paradigms, together with their .05confidence intervals. As shown, all effect sizes varied around r = .22 (5% varianceexplained), which indicates a small to medium effect. Whenever possible, we also calculatedthe effect sizes for analyses over items. On average, these were comparable but slightlysmaller than the ones presented in Table 6.3. Some insights from the literature reviewOf the published literature identified, it appears as if the processing of a target word is facilitatedby the presentation of a masked homophone (word or nonword) prime relative to thepresentation of a graphemic control. The influence of a masked phonological prime on targetprocessing is small (particularly in backward masking), but statistically reliable across studieswithin every experimental paradigm considered. At the same time, however, the literaturereview unearthed a number of methodological issues, which in our view limit the persuasivenessof the existing findings and call for further empirical evidence to substantiate the effect.3.1. A prime–response correlation in lexical decisionIn all but one of the lexical decision experiments reviewed in Table 4, pseudohomophoneprimes were always followed by word targets requiring the ‘‘YES’’ response. 22 The exception is Berent (1997, Experiment 2), in which pseudohomophone foils were preceded by identityprimes. This aspect of Berent’s (1997) study may be undesirable nevertheless, because there is evidence thatparticipants can exploit information disclosed by primes and targets in combination (e.g., Ratcliff & McKoon,1988). In the experimental situation reported by Berent (1997), the lexical decision response was ‘‘NO’’ wheneverthe overlap between primes and targets was maximal.
K. Rastle, M. Brysbaert / Cognitive Psychology 53 (2006) 97–145 111It is possible that this stimulus arrangement created the conditions for particularly rapid‘‘YES’’ responses, which may have had nothing to do with the phonological overlapbetween primes and targets. Rather, any information accessible to participants thatdistinguished pseudohomophone primes from other primes in these experiments—forexample, the total activation that these primes produce in the phonological lexicon (cf.Grainger & Jacobs, 1996)—could have cued a rapid ‘‘YES’’ response, and thus theappearance of a phonological priming effect. This possibility is not unreasonable,especially given the body of research demonstrating response congruity effects forsubliminally presented primes (e.g., Dehaene et al., 1998; Forster, 2004; Reynvoet,Caessens, & Brysbaert, 2002).This problem with the existing lexical decision data is very important to us, because weprefer that task to the others for several reasons. For one, we consider the lexical decisiontask to be the most appropriate with which to examine weak phonological theories of visualword recognition implemented as computational models. For example, whereas theDRC model has implemented procedures for carrying out visual lexical decision, neitherperceptual identification nor gaze duration has been considered. Even apart from theDRC model, though, we prefer the forward-masked visual lexical decision task to theother tasks for both theoretical and empirical reasons. Showing a phonological primingeffect in reading aloud, for example, is theoretically less interesting than in lexical decision,because there is no disagreement between weak and strong phonological theories that aphonological representation must be computed prior to the reading aloud response. Similarly,showing the effect in forward or backward masked perceptual identification may beempirically less interesting than in lexical decision, because we do not know to what extentthese tasks involve off-line guessing (see e.g., Brysbaert & Praet, 1992; and Perry & Ziegler,2002 for evidence that masked perceptual identification may be subject to undesirableeffects of guessing). Finally, the eye-movement data deviate from the other findings,because the effect seems to be particularly robust with high-frequency word primes(Lee et al., 1999a).3.2. What makes an adequate graphemic control?Phonological primes typically overlap their targets on both phonological and orthographicdimensions (e.g., kake–CAKE). The influence of phonological overlap alone istherefore obtained by comparing the phonological priming condition to a graphemic controlpriming condition (e.g., pake–CAKE), the logic being that any additional primingobserved in the phonological priming condition must be due to phonological overlapalone. This logic hinges on the requirement that phonological primes and graphemic controlprimes share equivalent orthographic similarity with their targets. If targets are moreorthographically similar to their phonological primes than to their graphemic controlprimes, then any benefit yielded by the phonological primes can be ascribed to the orthographicsimilarity between primes and targets instead of their phonological similarity.Conversely, if targets are more orthographically similar to their graphemic control primesthan to their phonological primes, then any additional benefit yielded by phonologicaloverlap may be hidden.How, then, might we ensure that graphemic controls and phonological primes haveequivalent orthographic similarity to targets? The vast majority of articles that we surveyedreported constructing graphemic controls such that they preserved the number of
Page 5 and 6: K. Rastle, M. Brysbaert / Cognitive
Page 7 and 8: Table 1Studies of English phonologi
Page 13: Table 5Studies of English phonologi
Page 21 and 22: Phonological priming effects on RTs
Page 23 and 24: As in Experiment 1, we assessed pho
Page 25 and 26: simulation of masked priming. It is
Page 27 and 28: items yielded this pattern; and at
Page 31 and 32: Despite improvement in the analysis
Page 41 and 42: flu phlue slaur DSD 727 669 0.15 0.
Page 43 and 44: nerve nurve narve SDS 563 547 0.10
Page 49: K. Rastle, M. Brysbaert / Cognitive

Masked phonological priming effects in English - Center for Reading ...

Create successful ePaper yourself

Delete template?

Save as template?