118 K. Rastle, M. Brysbaert / Cognitive Psychology 53 (2006) 97–1455.1.2. Stimuli and apparatusFor the YES response of the visual lexical decision task, pseudohomophone primes,graphemic controls, and word targets were taken from Experiment 1. These were divided<strong>in</strong>to counterbalanc<strong>in</strong>g conditions <strong>in</strong> exactly the same manner as <strong>in</strong> Experiment 1.For the NO response, one hundred and twelve pseudohomophone targets, each withthree phonemes, were selected from the ARC Nonword Database (Rastle et al., 2002).Pseudohomophone targets were equated to word targets on number of letters (words:mean 4.25 letters, range 3–6; pseudohomophones; mean 4.25 letters, range 3–6,t(222) = 0.00), and neighborhood size (words: mean 8.39 neighbors, range 0–21; pseudohomophones:7.83 neighbors, range 0–21, t(222) < 1).Pseudohomophone primes and graphemic controls were created <strong>for</strong> each of thesepseudohomophone targets <strong>in</strong> the same way as was accomplished <strong>for</strong> the word targets.For 56 of the pseudohomophone targets, a <strong>phonological</strong>ly identical pseudohomophoneprime was generated that differed by one (e.g., phite–FITE), two (e.g., phib–FIBB), orthree (e.g., wrighs–RIZE) graphemes. With<strong>in</strong> the sets of pseudohomophone prime–targetpairs derived by a one- or two-grapheme change, we further varied the position at whichthe graphemic change was made <strong>in</strong> exactly the same manner as was accomplished <strong>for</strong> wordprime–target pairs. For the other 56 pseudohomophone targets, a <strong>phonological</strong>ly dissimilarnonword control prime was generated that differed from the target by one (e.g., biss–BUSS), two (e.g., leeth–LAIM), or three (e.g., marf–BEED) graphemes. With<strong>in</strong> the sets ofcontrol–target pairs derived by a one- or two-grapheme change, we manipulated the positionat which that graphemic change was made <strong>in</strong> the same manner as was done <strong>for</strong> wordcontrol–target pairs. Prime–target pairs <strong>for</strong> the NO response are conta<strong>in</strong>ed <strong>in</strong> Appendix B.5.1.3. ProcedureThe procedure <strong>for</strong> runn<strong>in</strong>g the experiment was identical to that used <strong>in</strong> Experiment1, with one alteration. To ascerta<strong>in</strong> whether participants could identify explicitly thepseudohomophone status of the masked primes, stimuli were presented to each participanta second time immediately follow<strong>in</strong>g the ma<strong>in</strong> experiment. Targets rema<strong>in</strong>ed onscreen <strong>for</strong> 416 ms, and were preceded by the 58 ms masked primes. At this po<strong>in</strong>t, participantswere told of the existence of the prime, and were asked to decide as quicklyand as accurately as possible whether that prime sounded like an <strong>English</strong> word. Participants<strong>in</strong>dicated their decisions on a two-button response box, as <strong>in</strong> the ma<strong>in</strong>experiment.5.2. ResultsReaction time and error data were collected and cleaned <strong>in</strong> the same manner as <strong>in</strong>Experiment 1. Six participants with unusually slow and/or error prone responses relativeto the rest of the sample were removed: Five participants were removed because of a rateof false positive responses exceed<strong>in</strong>g 25%, and one participant was removed because of anaverage NO response RT greater than 1500 ms. Six word targets were removed becausethey produced an error rate exceed<strong>in</strong>g 35%: NORSE, BADE, DUES, FOB, WAIF, andVAT. F<strong>in</strong>ally, 17 outly<strong>in</strong>g data po<strong>in</strong>ts exceed<strong>in</strong>g 2400 ms (0.21% of the rema<strong>in</strong><strong>in</strong>g datapo<strong>in</strong>ts) were removed from the YES data. (Note that respond<strong>in</strong>g <strong>in</strong> this experiment wasgenerally slower than <strong>in</strong> Experiment 1, and this is reflected <strong>in</strong> the trimm<strong>in</strong>g criteria). Itemdata are presented <strong>in</strong> Appendix A.
As <strong>in</strong> Experiment 1, we assessed <strong>phonological</strong> <strong>prim<strong>in</strong>g</strong> <strong>effects</strong> on RT and error data byconduct<strong>in</strong>g by-subjects and by-items ANOVAs which treated prime type (2 levels: <strong>phonological</strong>prime versus graphemic control) and list version (2 levels) as factors. In analysesby-subjects and by-items, prime type was treated as a repeated factor and list versionwas treated as an unrepeated factor. Analyses of the latency data revealed that responsesto target words were facilitated by the presence of a masked <strong>phonological</strong> prime (634 ms,by items) relative to a graphemic control prime (643 ms, by items), F 1 (1,78) = 4.31,p < .05, MSE = 773.59, F 2 (1,104) = 3.75, p = .056, MSE = 1144.90. Effect size calculation(us<strong>in</strong>g the <strong>for</strong>mulas described <strong>in</strong> the Introduction) yielded r values of .23 and .19 <strong>for</strong> theanalyses by subjects and by items, respectively. Although the <strong>prim<strong>in</strong>g</strong> advantage was alsoseen numerically <strong>in</strong> the error data (primed: 5.7% errors; unprimed 6.4% errors, by items),this difference did not reach statistical significance, F 1 (1,78) = 1.77, n.s., F 2 (1,104) = 1.46,n.s.To ascerta<strong>in</strong> whether participants were able to identify the pseudohomophone status ofthe masked primes, we exam<strong>in</strong>ed error data <strong>for</strong> the second presentation of the stimulusmaterials (<strong>in</strong> which participants were asked if the prime sounded like a word). Therewas no evidence that participants could identify the pseudohomophone status of the primestimuli under these conditions (48.24% errors, by items, where chance per<strong>for</strong>mance isequal to 50%).6. SimulationHav<strong>in</strong>g confirmed the existence of the masked <strong>phonological</strong> <strong>prim<strong>in</strong>g</strong> effect on lexicaldecision, we are now <strong>in</strong> a position to consider its theoretical implications <strong>for</strong> weak <strong>phonological</strong>theories of visual word recognition. For several reasons outl<strong>in</strong>ed <strong>in</strong> the <strong>in</strong>troduction,we chose to evaluate the weak <strong>phonological</strong> theory of visual word recognitionexpressed by the DRC model (Coltheart et al., 2001) <strong>in</strong> this context. It may <strong>in</strong> future beimportant to complement this <strong>in</strong>vestigation of the DRC model with an <strong>in</strong>vestigation ofthe triangle model of Harm and Seidenberg (2004)—another computational implementationof a weak <strong>phonological</strong> theory. We were not able to pursue this route because, unlikethe DRC model, this model is not presently available <strong>for</strong> public evaluation. Even if it wereavailable, however, <strong>in</strong>vestigat<strong>in</strong>g its per<strong>for</strong>mance would have required us to make crucialdecisions about the simulation of lexical decision. Though the authors of this model havedrafted some ideas about the source of <strong>in</strong><strong>for</strong>mation used to make lexical decisions (e.g.,activation of semantic feature units, semantic stress, orthographic stress, orthographic distance),they have not made any commitments concern<strong>in</strong>g this issue. Further, there hasbeen considerable scepticism about whether this model (and those related to it) could per<strong>for</strong>mthe lexical decision task <strong>in</strong> the manner <strong>in</strong> which human readers accomplish the task(e.g., Borowsky & Besner, <strong>in</strong> press; Coltheart, 2004; Rastle & Coltheart, <strong>in</strong> press). Forthese reasons, it seems that much further work on this model may be necessary be<strong>for</strong>eit can be evaluated <strong>in</strong> the context of masked <strong>phonological</strong> <strong>prim<strong>in</strong>g</strong> <strong>effects</strong>.6.1. Approaches to simulationK. Rastle, M. Brysbaert / Cognitive Psychology 53 (2006) 97–145 119The DRC model of visual word recognition (Coltheart et al., 2001) is represented <strong>in</strong>Fig. 1. Lexical decisions <strong>in</strong> the model are made on the basis of an analysis of activity <strong>in</strong>units of the orthographic lexicon. As shown, the activation of these orthographic units
- Page 5 and 6: K. Rastle, M. Brysbaert / Cognitive
- Page 7 and 8: Table 1Studies of English phonologi
- Page 9 and 10: Table 2Studies of English phonologi
- Page 11 and 12: Table 4Studies of English phonologi
- Page 13 and 14: Table 5Studies of English phonologi
- Page 15 and 16: K. Rastle, M. Brysbaert / Cognitive
- Page 17 and 18: K. Rastle, M. Brysbaert / Cognitive
- Page 19 and 20: K. Rastle, M. Brysbaert / Cognitive
- Page 21: Phonological priming effects on RTs
- Page 25 and 26: simulation of masked priming. It is
- Page 27 and 28: items yielded this pattern; and at
- Page 29 and 30: K. Rastle, M. Brysbaert / Cognitive
- Page 31 and 32: Despite improvement in the analysis
- Page 33 and 34: K. Rastle, M. Brysbaert / Cognitive
- Page 35 and 36: K. Rastle, M. Brysbaert / Cognitive
- Page 37 and 38: K. Rastle, M. Brysbaert / Cognitive
- Page 39 and 40: K. Rastle, M. Brysbaert / Cognitive
- 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 45 and 46: K. Rastle, M. Brysbaert / Cognitive
- Page 47 and 48: K. Rastle, M. Brysbaert / Cognitive
- Page 49: K. Rastle, M. Brysbaert / Cognitive