Letter Position Information And Printed Word Perception: The Relative-position Priming Constraint

Transcript

Journal of Experimental Psychology: Human Perception and Performance 2006, Vol. 32, No. 4, 865– 884 Copyright 2006 by the American Psychological Association 0096-1523/06/$12.00 DOI: 10.1037/0096-1523.32.4.865 Letter Position Information and Printed Word Perception: The Relative-Position Priming Constraint Jonathan Grainger Jean-Pierre Granier University of Provence and Centre National de la Recherche Scientifique University of Provence Fernand Farioli Eva Van Assche University of Provence and Centre National de la Recherche Scientifique University of Ghent Walter J. B. van Heuven University of Nottingham Six experiments apply the masked priming paradigm to investigate how letter position information is computed during printed word perception. Primes formed by a subset of the target’s letters facilitated target recognition as long as the relative position of letters was respected across prime and target (e.g., “arict” vs. “acirt” as primes for the target “apricot”). Priming effects were not influenced by whether or not absolute, length-dependent position was respected (e.g., “a-ric-t” vs. “arict”/“ar-i-ct”). Position of overlap of relative-position primes (e.g., apric-apricot; ricot-apricot; arict-apricot) was found to have little influence on the size of priming effects, particularly in conditions (i.e., 33 ms prime durations) where there was no evidence for phonological priming. The results constrain possible schemes for letter position coding. Keywords: letter position coding, masked priming, relative-position priming, orthographic processing, word recognition position priming. Relative-position priming is a form of orthographic priming in which priming effects depend on shared letters having the same relative position in prime and target stimuli. Thus, for example, given the target word “apricot,” the prime stimulus “arct” is said to preserve relative letter positions (i.e., the letter “r” is after “a” and before “c”) but to violate absolute, lengthdependent position. Most studies of orthographic priming to date have used substitution primes (e.g., adricot) where shared letters preserve absolute letter position (e.g., Forster, Davis, Schoknecht, & Carter, 1987; Grainger & Jacobs, 1993). With brief prime exposures (50 – 60 ms) and forward masking of prime stimuli, substitution primes generally facilitate target word recognition compared to unrelated control primes (see Grainger & Jacobs, 1999, for a review). Priming effects obtained with substitution primes were taken as evidence in favor of orthographic coding schemes that use length-dependent, position-specific letter detectors, such as in McClelland and Rumelhart’s (1981) interactiveactivation model. However, there are an increasing number of behavioral results that are incompatible with such an approach (see Grainger & van Heuven, 2003; and Perea & Lupker, 2003, for reviews). Relative-position priming is one such phenomenon. Humphreys, Evett, and Quinlan (1990) were the first to study relative-position priming. They used a four-field masking procedure that involved brief presentations of both prime and target stimuli. In this paradigm, primes (nonwords) and targets (words) are briefly presented one after the other. Immediately before the prime and after the target, two masking patterns are displayed, and There is a general consensus today among researchers investigating printed word perception that abstract letter identities, which are independent of type font and case, represent one particularly relevant source of information for the word recognition process (e.g., Besner, Coltheart, & Davelaar, 1984; Evett & Humphreys, 1981; Rayner, McConkie, & Zola, 1980). However, in mapping letter identities onto whole-word representations in memory, it is clear that information about letter position must also be computed. The importance of this type of information is obvious given that languages that use alphabetic orthographies, such as English and French, have large numbers of anagrams. We are able to distinguish words containing the same letters (e.g., BALE-ABLE) on the basis of the different position of the letters in the string, just as we are able to recognize that BLAE is not an English word. The question to be addressed in the present study is, therefore, how do skilled readers code such position information during printed word perception? In an attempt to answer this question, the present study provides a further investigation of a phenomenon referred to as relativeJonathan Grainger and Fernand Farioli, University of Provence, France, and Centre National de la Recherche Scientifique, France; Jean-Pierre Granier, University of Provence; Eva Van Assche, University of Ghent, Belgium; Walter J. B. van Heuven, University of Nottingham, United Kingdom. Correspondence concerning this article should be addressed to Jonathan Grainger, Laboratoire de Psychologie Cognitive, Universite´ de Provence, 3 place Victor Hugo, 13331, Marseille, France. E-mail: grainger@up .univ-mrs.fr 865 866 GRAINGER ET AL. the participant’s task is to recognize the target word presented in upper case letters. The exposure durations of the four fields are adjusted so that participants correctly reported about 40% of targets. The results showed that priming effects varied as a function of both the number and the position of letters shared by primes and targets of the same length. Greater degrees of orthographic overlap produced larger priming effects, but only when shared letters occupied the same position in primes and targets. In the same study however, priming effects were also obtained when primes and targets differed in length. So, for example, the sequence “bvk” facilitated the identification of the word “black” with respect to the neutral condition “ovf,” just as well as the sequence “btvuk” with respect to the neutral condition “otvuf.” In the case of “bvk” primes, letters shared by prime and target are said to preserve their relative position (ordinal information) and not their length-dependent absolute position (as is the case with “btvuk” primes). Peressotti and Grainger (1999) replicated the relative-position priming reported by Humphreys et al. using the masked priming technique of Forster and Davis (1984). In their study, Peressotti and Grainger (1999) found that modifying the absolute position of letters shared between prime and target while maintaining their correct relative-position (e.g., blcn-balcon) did not affect priming relative to unrelated primes (e.g., tpvf-balcon). Inserting filler letters or characters (e.g., bslcrn, b-lc-n) to provide absolute position information never led to significantly larger priming effects. Furthermore, the results of one specific experiment showed no priming relative to an unrelated prime condition when the relativeposition of two out of four letters is violated in orthographically related primes and 6-letter target words (e.g., bcln-balcon). On the other hand, if all four prime letters maintain their relative-position in the target string, but not their absolute position in the target, significant priming is obtained. These results therefore add support to the work of Humphreys et al. (1990) showing that some form of relative-position coding operates on printed strings of letters. The relative-position priming effects observed by Humphreys et al. (1990) and Peressotti and Grainger (1999) show very clearly the limits of length-dependent, position-specific coding schemes, such as implemented in the interactive-activation model (McClelland & Rumelhart, 1981). Furthermore, the vast majority of attempts to increase the flexibility of such coding schemes (e.g., Coltheart, Curtis, Atkins, & Haller, 1993; Coltheart, Rastle, Perry, & Ziegler, 2001; Plaut, McClelland, Seidenberg, & Patterson, 1996; Seidenberg & McClelland, 1989) have not lead to any greater success in capturing the basic relative-position priming effects. For example, none of the above-cited coding schemes can account for Peressotti and Grainger’s (1999) Experiment 2, which used 4-letter primes and 6-letter targets. In this experiment, priming was observed for “1346” and not for “1436” primes. According to both the IAmodel and Seidenberg and McClelland’s wickelgraph coding scheme, there is no overlap between prime and target stimuli for either of these prime conditions. According to the DRC (dual route cascaded) model, only the first letter overlaps, and so the two prime conditions are matched in terms of their orthographic similarity with targets. In vowel-centered coding schemes for monosyllabic words (e.g., Plaut et al., 1996), letter identities are assigned to one of three positions that correspond to the orthographic onset, nucleus, and coda of the word. Relative-position primes often violate such structure and therefore these schemes cannot account for these data (see Davis & Bowers, 2004, for a similar analysis with related phenomena). With these considerations in mind, Grainger and van Heuven (2003) recently argued that relative-position priming is one of the single most constraining pieces of evidence for models of orthographic processing. Relative-position priming clearly reflects a stage of processing where retinotopic information is lost, and some form of word-centered, location-invariant orthographic coding operates (Caramazza & Hillis, 1990). The main objective of the present study is to provide a larger empirical foundation for this theoretically important phenomenon. The larger empirical foundation will be provided by 1) examining the limits of relativeposition priming in terms of degree of orthographic overlap across prime and target in words of different lengths (7 and 9-letters); 2) evaluating possible sequential biases and the relative importance of outer and inner letters in relative-position priming; 3) comparing relative-position priming effects in two different tasks (lexical decision and perceptual identification); 4) evaluating the influence of masking (presence of a mask and length of mask) on relativeposition priming; 5) examining priming effects at different prime exposure durations; and 6) comparing relative-position priming with phonological priming. Table 1 summarizes the different priming conditions to be tested in the present study. In the general discussion we will present some recent theoretical proposals for letter position coding, and examine how these fare in accounting for the results of the present experiments. Experiment 1 As a further test of relative-position versus absolute-position priming, Experiment 1 uses longer words than tested in prior research (Humphreys et al., 1990; Peressotti & Grainger, 1999). It is possible that the use of length-dependent position information becomes more useful as word length increases. (Since the number of different words of a given length decreases with word length, length-specific information becomes more informative.) Experiment 1 tests 7-letter words, and Experiments 2–5 include 9-letter stimuli. Furthermore, Experiment 1a includes a new manipulation Table 1 Summary of the Priming Conditions Tested in the Main Experiments Experiment 1a: Experiment 1b: Experiment 2: 7-letter: 9-letter: Experiments 3–5: 7-letter: 9-letter: Experiments 6: RP: PH: 1-345-7 1-345-7 13-4-57 1-543-7 13457 7-345-1 ddddd d-ddd-d 12345 12345 34567 56789 13457 14569 ddddd ddddd 1234 1234 4567 6789 1357 1469 dddd dddd 1357 P⫹O⫹ 1537 P⫺O⫹ 7351 P⫹O⫺ dddd P⫺O⫺ Note. The numbers in the example primes refer to letters shared between prime and target in that specific position. Hyphens refer to the use of this symbol in the prime stimulus and the letter “d” refers to the presence of a different letter at a given position in the prime and target. Experiment 6 tests relative-position priming (RP) and phonological priming (PH) with different levels of phonological (P⫹/P⫺) and orthographic (O⫹/O⫺) overlap across primes and targets. RELATIVE-POSITION PRIMING of absolute versus relative-position in prime stimuli of the same length. The prime stimuli 1-345-7 and 13-4-57 both have the same length as target stimuli, and letters respect relative position in both cases. However, in the second prime condition absolute, lengthdependent position is incorrect for two of the five letters. These priming conditions are compared to a relative-position prime Condition 13457 where all prime letters are concatenated. Method Participants. Forty students at the University of Provence took part in Experiment 1a and 44 in Experiment 1b. They all reported being native speakers of French with normal or corrected-to-normal vision. Stimuli and design. Sixty French words, seven letters long, with printed frequencies ranging from 7 to 175 per million (Imbs, 1971) were selected. The words were nouns, adjectives, or verbs in infinitive form. Sixty pronounceable, orthographically regular nonwords were created that were all seven letters long. These 120 items formed the targets. Four priming conditions were tested in each subexperiment with different participants. In Experiment 1a each related prime was made up of the two outer letters and of the central triplet of the corresponding target (respectively letters 3, 4 and 5). Two hyphen marks were used as filler characters in place of the missing letters (i.e., letters 2 and 6). In Condition 1, the central triplet of the prime was in the same absolute position as in the target (1-345-7)1. In Condition 2, absolute position information was disrupted by inserting hyphen marks within the triplet (13-4-57), while in Condition 3 it was disrupted by removing the hyphen marks (13457). Finally, in the fourth condition, the primes were formed by five unrelated letters (ddddd). In Experiment 1b, the first prime condition was the same as in Experiment 1a (1-345-7). In Condition 2, the order of the letters in the central triplet was reversed (1-543-7), and in Condition 3 the two outer letters were reversed (7-345-1). In Condition 4, the prime string was formed by five unrelated letters and two hyphen marks (d-ddd-d). Counterbalancing with a Latin Square design allowed each target to be tested in the four priming conditions, across four lists associated with four independent groups of participants in each subexperiment. Procedure. Each trial consisted of three stimuli presented one after the other at the center of a computer screen. The first was a row of seven hash marks (#######) that served as a forward pattern mask and remained in view for 500 ms. The second was the prime stimulus, which was displayed for 50 ms, and was immediately followed by the third stimulus, the target, which lasted until participants’ response or for a maximum time of 1000 ms. Stimulus presentation and response collection were controlled using Psyscope software (Cohen, MacWhinney, Flatt, & Provost, 1993) on a Macintosh Centris computer. Primes and targets were strings of centerjustified lower-case letters presented in fixed-width courier font. They had different sizes in order to minimize physical overlap: courier 14 for primes and courier 24 for targets. Each letter of the prime was approximately 0.3 cm wide, and each letter of the target approximately 0.5 cm wide. Participants sat in front of the computer at a viewing distance of approximately 60 cm. At that distance, the horizontal visual angle subtended by each letter of the prime was approximately 0.3 degrees, and for each letter of the target it was approximately 0.5 degrees. Participants were instructed to attend to the center of the string of hash marks when they appeared, and to decide as rapidly and as accurately as possible whether the following string of letters was or was not a French word. The presence of a prime was not mentioned. They responded “Yes” by pressing one response button with the forefinger of the preferred hand and “No” by pressing the other response button with the forefinger of the other hand. The response buttons selected were the two outer buttons of the Psyscope button box. The intertrial interval was 800 ms. Stimulus presentation was randomized within each block with a different order for each participant. Response times (RTs) were measured to the nearest millisecond. 867 Results RTs for correct responses were analyzed after removing all values smaller than 300 ms and greater than 1500 ms. In all cases this resulted in removal of less than 1% of the data for correct responses. Analyses of variance (ANOVAs) were performed with participants (F1) and items (F2) as random variables. The mean RT and percent error for each condition are shown in Table 2. Given the particular design of the present set of experiments, as well as main effects, pairwise comparisons with the unrelated prime condition were performed in order to examine priming effects in the various related prime conditions. All comparisons against the unrelated prime condition were performed using Dunnett’s test (Dunnett, 1955). Other pairwise comparisons that were critical in a specific experiment were tested using planned comparisons. In the by-participant analyses, a list was included as a betweenparticipants factor in order to extract the variance associated with this (Pollatsek & Well, 1995). Experiment 1a. An ANOVA on mean correct RTs to word targets revealed a main effect of prime type (F1(3,108) ⫽ 6.72, p ⬍ .001; F2(3,177) ⫽ 6.23, p ⬍ .001). Pairwise comparisons using Dunnett’s test indicated that the unrelated prime condition produced significantly longer RTs than the 1-345-7 condition (t1(39) ⫽ 2.86, p ⬍ .05; t2(59) ⫽ 2.74, p ⬍ .05), than the 13-4-57 condition (t1(39) ⫽ 3.61, p ⬍ .01; t2(59) ⫽ 3.60, p ⬍ .01), and the 13457 condition (t1(39) ⫽ 4.10, p ⬍ .01; t2(59) ⫽ 3.22, p ⬍ .01). No significant effects were revealed in ANOVAs performed on the percentages of error to word targets, nor on the mean correct RTs and percentages of error to nonword targets. Experiment 1b. An ANOVA on mean correct RTs to word targets showed a main effect of prime type (F1(3,120) ⫽ 7.89, p ⬍ .001; F2(3,177) ⫽ 14.27, p ⬍ .001). Pairwise comparisons using Dunnett’s test revealed a significant difference between the 1-345-7 condition and the all different letter primes (t1(43) ⫽ 3.77, p ⬍ .01; t2(59) ⫽ 5.61, p ⬍ .01). The other comparisons were not significant. No significant effects were revealed in ANOVAs performed on the percentages of error to word targets, nor on the mean correct RTs and percentages of error to nonword targets. Discussion The results of Experiment 1a show that disrupting absolute position information by inserting hyphens in the wrong place (13-4-57) or removing the hyphens (13457) does not modify priming effects compared with the absolute position condition (1-345-7). All of these priming conditions produced significantly faster RTs to target words than the unrelated prime condition. This is therefore further evidence in favor of some form of relativeposition coding of letter strings. The results of Experiment 1b are perfectly in line with those reported by Peressotti and Grainger (1999, Experiment 2). In the Peressotti and Grainger study using 6-letter target words, only 1346 primes produced significant effects on RT. The 1436 and 6341 prime conditions did not differ 1 Related priming conditions are described using the numbered letters of the target (e.g., 12345 for a 5-letter target) to indicate which of the target letters appeared in the prime and in which location they appeared. So the prime “135” indicates that primes were composed of the first, third, and fifth letters of the target. GRAINGER ET AL. 868 Table 2 Mean Response Times (RT in ms) and Percentages of Error (PE) for the Different Priming Conditions Tested in Experiment 1 Priming condition Experiment 1a Word targets RT PE Nonword targets RT PE 1-345-7 13-4-57 13457 ddddd 560* 2.8 558* 1.9 555* 2.2 580 2.9 616 2.5 615 3.0 622 3.1 626 3.5 Experiment 1b Word targets RT PE Nonword targets RT PE 1-345-7 1-543-7 7-345-1 d-ddd-d 565* 1.8 591 2.9 581 2.0 589 1.5 659 3.6 653 3.5 661 2.5 669 3.9 * Significantly different from the ddddd condition, p ⬍ .05 by participant using Dunnett’s test. significantly from the unrelated prime condition. In Experiment 1b, the same pattern was observed with 7-letter target words, such that priming effects disappeared when the order of letters shared by prime and target no longer matched (the 1-543-7 and 7-345-1 prime conditions). Furthermore, the results of Experiment 1b showed no sign of an advantage when outer letters were maintained in their correct position compared to when they were not. Indeed, the 1-543-7 priming condition generated numerically slower RTs and more errors than the 7-345-1 condition. The following experiments will provide further evidence on this point. Experiment 2 Experiment 2 examines relative-position priming with 7-letter and 9-letter targets with primes sharing the first five letters of target words (initial primes), the last five letters (final primes), or the first and last letter plus three central letters (outer-central primes). In all the related prime conditions, the prime letters respected their order in the corresponding target stimulus (i.e., the relative position of letters was respected across prime and target, but not absolute position). This manipulation allows us to examine possible positional biases in relative-position priming. Whether word-initial overlap generates stronger priming than word-final overlap, and whether primes sharing both of the targets’ outer letters are more effective than primes with just a single outer letter. Concerning sequential biases in masked orthographic priming, the evidence at present from both perceptual identification experiments (Humphreys et al., 1990) and lexical decision experiments (Grainger & Jacobs, 1993) suggests that word-initial letters do not have any privileged status over word-final letters. However, these experiments were performed on short (4 and 5-letter) words, so it remains to be seen whether increasing word length will allow sequential biases to emerge. On the other hand, there is evidence that a word’s outer letters enjoy a privileged status relative to inner letters (e.g., Jordan, Thomas, Patching, & Scott-Brown, 2003; Humphreys et al., 1990). This stands in contrast to the results observed in Experiment 1. Since masks were longer than the prime stimuli in Experiment 1, and primes were presented in a smaller font size than the targets, this might have led to less accurate coding of the prime’s outer letters. A mask that is longer than the prime could generate visual noise in the space before and after the prime stimulus, and therefore hinder the assignment of letter identities to the first and last positions. Furthermore, the fact that targets were larger than primes would also contribute to rendering the boundaries of the prime stimulus less detectable. This should not be the case when the forward mask is the same size as the prime stimulus, and when the prime is bigger than the target. These two types of masking conditions are compared in Experiment 2. Method Participants. Fifty-six students at the University of Provence took part in this experiment. They all reported being native speakers of French with normal or corrected-to-normal vision, and had not participated in the previous experiment. Stimuli and design. A new set of sixty 7-letter words was selected using the Lexique database (New, Pallier, Ferrand, & Matos, 2001). Their printed frequencies ranged from 7 to 290 per million (mean 82 per million). A set of sixty 9-letter words was selected with printed frequencies ranging from 4 to 179 per million (mean 40 per million). Only nouns were used in this experiment in order to reduce possible variance caused by different grammatical categories. One hundred twenty pronounceable, orthographically regular nonwords were created, half of which were 7-letters long and the other half 9-letters long. Four priming conditions were constructed. In the first condition the prime was made up of the first five letters of the target (initial letter primes), in the second condition it was formed by the last five letters of the target (final letter primes), and in the third condition by both outer letters and the central triplet of the target (outer-central primes). In the baseline condition, the prime string consisted of five unrelated letters. The size of the forward mask and prime stimulus was manipulated. In one condition, masks were composed of 11 hash marks that extended either one or two spaces to the left and to the right of prime stimuli, with mask and primes presented in Arial 12 point and targets in Arial 16. In the other condition, masks were the same length as primes and both were presented in a bigger font size (Arial 16) than the targets (Arial 12). Mask, prime, and target size was manipulated between participants. Prime-target pairs were counterbalanced across four experimental lists following a Latin-Square design. Procedure. This was the same as in the previous experiment except that DMDX software (Forster & Forster, 2003) was used on a PC computer. In one condition, prime stimuli were presented in Arial 16 font and the targets in Arial 12 font, and in another condition primes were presented in Arial 12 and targets in Arial 16. In Arial 16 font each letter was approximately 0.3 cm wide, and in Arial 12 font each letter was about 0.2 cm wide (respectively about 0.3 degrees and 0.2 degrees of horizontal visual angle for a viewing distance of 60 cm). Results Mean RTs and percentages of error per experimental condition are shown in Table 3. An ANOVA on mean correct RTs to word targets showed a main effect of prime type (F1(3,144) ⫽ 39.76, p ⬍ .0001; F2(3,354) ⫽ 40.45, p ⬍ .0001), an effect of length (F1(1,48) ⫽ 124.04, p ⬍ .0001; F2(1,118) ⫽ 36.4, p ⬍ .0001), no effect of mask size, and no interactions (all Fs ⬍1). Pairwise comparisons against the unrelated condition using Dunnett’s test revealed significant differences for all three related prime conditions, respectively: initial primes (t1(55) ⫽ 8.69, p ⬍ .01 and RELATIVE-POSITION PRIMING Table 3 Mean Response Times (RT in ms) and Percentages of Error (PE) for the Different Priming Conditions Tested in Experiment 2 Priming condition 7-letter words RT PE 9-letter words RT PE 7-letter nonwords RT PE 9-letter nonwords RT PE 12345 531* 1.1* 12345 567* 1.9 12345 638 1.4* 12345 692 4.4 34567 539* 1.6* 56789 571* 1.3* 34567 643 2.3 56789 688 4.5 13457 547* 0.7* 14569 585 2.7 13457 631 1.1* 14569 686 4.4 ddddd 576 3.7 ddddd 597 2.7 ddddd 632 3.4 ddddd 694 5.2 * Significantly different from the ddddd condition, p ⬍ .05 by participant using Dunnett’s test. 869 should generate incorrect coding of the end letters of the prime. This result is also evidence against a privileged role played by outer letters in printed word perception as previously reported in experiments using degraded target presentation (e.g., Jordan et al., 2003; Humphreys et al., 1990). In line with this reasoning, we also failed to observe an influence of forward mask length on relativeposition priming effects in Experiment 2. Given a privileged role for outer letters, it was expected that forward masks that were the same length as primes ought to facilitate the processing of such letters compared to a condition with masks that extended beyond the prime stimulus. There was no evidence for such an influence in Experiment 2. Experiment 3 Experiment 3 tests a set of priming conditions similar to those tested in Experiment 2, but with a lower level of orthographic overlap (primes share four letters with 7-letter and 9-letter targets words). In the extreme condition (4-letter primes for 9-letter targets) the orthographic overlap is less than 50%. t2(119) ⫽ 9.96, p ⬍ .01); final primes (t1(55) ⫽ 9.18, p ⬍ .01 and t2(119) ⫽ 7.81, p ⬍ .01); outer-central primes (t1(55) ⫽ 5.12, p ⬍ .01 and t2(119) ⫽ 5.89, p ⬍ .01). Furthermore, planned comparisons showed that outer-central primes differed significantly from both initial, F(1, 48) ⫽ 36.61, p ⬍ .01; F2(1,118 ⫽ 24.2, p ⬍ .01), and final primes, F(1, 48) ⫽ 8.75, p ⬍ .01; F2(1,118 ⫽ 8.22, p ⬍ .01). Priming generated by initial and final primes did not differ significantly (F(1, 48) ⫽ 2.54; F2(1,118) ⫽ 3.5). An ANOVA performed on the percentages of error to word targets showed an effect of prime type, F(3, 144) ⫽ 6.8, p ⬍ .001, no effect of word length, no effect of mask size, and no interactions. An analysis of the mean correct RTs to nonword targets showed no effect of prime type, no effect of mask size, but a significant effect of length (F1(1,48) ⫽ 112.36, p ⬍ .0001; F2(1,118) ⫽ 41.24, p ⬍ .0001). None of the interactions between these factors were significant (all Fs ⬍1). Significant effects were observed in the error data to nonword targets for both prime type, in the analysis by participants (F1(3,144) ⫽ 2.85, p ⬍ .05), and length (F1(1,48) ⫽ 22.94, p ⬍ .00025; F2(1,118) ⫽ 22.83, p ⬍ .0001). Method Discussion Mean RTs and percentages of error per experimental condition are shown in Table 4. An ANOVA on mean correct RTs to word targets showed a main effect of prime type (F1(3,120) ⫽ 8.69, p ⬍ .001; F2(3, 354) ⫽ 8.40, p ⬍ .001), an effect of length (F1(1,40) ⫽ 42.70, p ⬍ .0001; F2(1,118) ⫽ 28.30, p ⬍ .0001), and no effect of list. Length did not interact with the effects of prime type (Fs ⬍ 1) and the list factor did not interact with these two factors. Pairwise comparisons against the unrelated conditions using Dunnett’s test revealed significant differences for initial primes (t1(43) ⫽ 4.92, p ⬍ .01; t2(119) ⫽ 4.75, p ⬍ .01), and final primes (t1(43) ⫽ 3.11, p ⬍ .01; t2(119) ⫽ 3.47, p ⬍ .01). On the other hand, outer-central primes (1357 or 1469) did not differ significantly from the unrelated condition. Planned comparisons between related prime conditions showed that initial primes differed significantly from outer-central primes (F1(1,40) ⫽ 10.63, p ⬍ .01; F2(1,118) ⫽ 7.83, p ⬍ .01). None of the other planned comparisons reached significance. The results of Experiment 2 are clear-cut. There was very little evidence for any beginning-to-end positional bias in relativeposition priming. The clearest evidence against a beginning-to-end processing bias is the fact that primes that did not contain the target’s initial letters (final letter primes) actually produced significantly faster RTs than one set of primes containing the target’s first letter (outer-central primes). Another important result obtained in Experiment 2 is that primes that contained both of the targets’ outer letters did not produce more priming than primes containing only a single outer letter. Indeed, the former condition actually produced slower RTs than the other two conditions. This result contradicts coding schemes that use the word center as an anchor point for relative-position coding (e.g., Caramazza & Hillis, 1990). According to this type of scheme, asymmetrical primes such as 12345 for a 7-letter target, Participants. Forty-four students at the University of Provence took part in this experiment. They all reported being native speakers of French with normal or corrected-to-normal vision, and had not participated in the previous experiments. Stimuli and design. The words and nonwords were the same as in Experiment 2. A new set of 4-letter prime stimuli were generated for the four priming conditions. In the first condition the prime was composed of the first four letters of the target, in the second condition it was composed of the last four letters of the target, and in the third condition it was composed of the two outer letters plus two inner letters (the two letters flanking the target’s central letter). Finally, in the baseline condition the prime string was formed by four unrelated letters. These four priming conditions were crossed with target length (7-letter or 9-letter targets) in a 2 (word length) ⫻ 4 (prime type) factorial design (see Table 1 for a summary of the priming conditions). Procedure. This was the same as in Experiment 2 except that primes were always presented in Arial 12 font and targets in Arial 16 font. (Font and mask size were not manipulated in this experiment.) Results GRAINGER ET AL. 870 Table 4 Mean Response Times (RT in ms) and Percentages of Error (PE) for the Different Priming Conditions Tested in Experiment 3 Priming condition 7-letter words RT PE 9-letter words RT PE 7-letter nonwords RT PE 9-letter nonwords RT PE 1234 562* 1.9 1234 591* 1.3 1234 668 1.2 1234 723 2.1 4567 573 1.6 6789 594* 1.2 4567 675 1.5 6789 727 1.9 1357 577 1.2* 1469 606 1.9 1357 664 1.8 1469 716 2.1 dddd 585 3.6 dddd 614 1.2 dddd 669 2.5 dddd 725 1.9 * Significantly different from the ddddd condition, p ⬍ .05 by participant using Dunnett’s test. An ANOVA performed on the percentages of error to word targets showed no significant main effects or interactions. An analysis of the mean correct RTs to nonword targets showed no effect of prime type, but a significant effect of length (F1(1,40) ⫽ 79.73, p ⬍ .0001; F2(1,118) ⫽ 120, p ⬍ .0001). There was no interaction between these factors. No effects were observed in the error data to nonword targets. Discussion The most important result of Experiment 3 is the fact that robust priming can be obtained with very low levels (less than 50%) of orthographic overlap between prime and target (4-letter primes and 9-letter targets). In terms of positional biases, the results of Experiment 3 perfectly replicate those of Experiment 2, showing very little evidence of any beginning-to-end bias (no significant difference between initial and final prime conditions), and no evidence in favor of primes containing both of the target’s outer letters. On the contrary, primes containing the two outer letters of target words produced smaller priming effects than the other related prime conditions, and were not significant relative to the alldifferent prime condition. This therefore once again counters prior observations of an outer letter advantage in visual word recognition (Humphreys et al., 1990; Jordan et al., 2003). However, the fairly meager evidence obtained so far for positional biases in relative-position priming could possibly reflect the conjoint influence of other factors that differ across the priming conditions we tested. The following post hoc analyses were designed to examine this possibility. Post Hoc Analyses of Experiments 2 and 3 In priming conditions similar to those tested in the present experiments, prior research has provided evidence that structural variables, such as morphemic and syllabic structure, can influence priming. For example, Ferrand and Grainger (1993) reported phonological priming that emerged with prime exposures of around 50 ms, and Diependaele, Sandra, and Grainger (2005) found morphological priming with prime exposures of 43 ms. Hence, part of the relative-position priming effects reported here could be due to such structural influences. In order to rule out this possibility, a series of post hoc analyses were performed on the results of Experiments 2 and 3. The hypothesized relevant variables were: syllable structure, morphological structure, consonant/vowel (CV) status of letters shared by prime and target, and amount of phonological overlap across prime and target. We also examined the possible influence of target word frequency and the bigram frequency of prime stimuli on priming effects in Experiments 2 and 3. Syllable Structure In this post hoc analysis we separated out cases where prime stimuli formed one of the target word’s syllables (e.g., lencesilence). Obviously, this situation only arose in the contiguous priming conditions (initial and final primes), and therefore could explain the nonsignificant priming for outer-central primes observed in Experiments 2 and 3. Averaging across experiments and word length (240 items), we found a quite stable pattern of priming for the no-syllable (average priming effect ⫽ 28 ms, N ⫽ 73), initial-syllable (word-initial priming effect ⫽ 34ms, N ⫽ 64), and final-syllable (word-final priming effect ⫽ 27ms, N ⫽ 95). An ANOVA with Prime Type and Syllable Structure (prime is a syllable of the target or not) showed that syllable structure did not affect priming (all Fs ⬍ 1). Morphological Structure and Embedded Words In the same manner as the preceding analysis, here we separated out cases where prime stimuli formed one of the target word’s morphemes (either a root or an affix). Prime stimuli never formed a free root, but some of the word-final primes (N ⫽ 26) did form a derivational suffix in Experiments 2 and 3. An analysis showed that primes that were suffixes of the target word actually produced numerically less priming. However, the interaction with priming was not significant (F ⬍ 1), and an analysis limited to only cases without suffix primes showed exactly the same pattern as the overall analysis. We also checked for a possible influence of the lexical status of prime stimuli that arose on a small number of occasions. In Experiment 2, removing the seventeen 9-letter targets for which one prime condition did form a word, produced a pattern that was almost identical to the complete analysis. The same was true for the 7-letter (N ⫽ 13) and 9-letter (N ⫽ 13) targets in Experiment 3. Thus the fact that a small number of prime stimuli formed embedded words or derivational affixes had no significant impact on the priming effects we observed. CV Status and Phonological Overlap In order to check whether priming effects were modulated by the number of vowels versus consonants shared by prime and target, these values were correlated with priming effect size across items. The average proportion of shared vowels was quite stable across the three related priming conditions (0.67, 0.55, and 0.66 for the initial, final, and outer-central primes respectively). The overall correlation between priming effect (three priming effects for each word target) and proportion of vowels was not significant (r ⫽ ⫺0.02, N ⫽ 720) for Experiments 2 and 3 together, and doing RELATIVE-POSITION PRIMING separate correlations for each word-length and experiment never produced a significant effect. Phonological overlap across primes and targets was estimated by counting the number of phonemes of the target word that were present in the prime stimulus. For the French target word “silence” for example, the prime stimulus “sile” shares three phonemes (/s/, /i/, and /L/) out of five (60%). Averaging across all items, the amount of phonological overlap was 65% for initial primes, 56% for final primes, and 41% for outer-central primes. T tests showed that all pairwise comparisons of degree of phonological overlap were highly significant (all ps ⬍.001). The overall correlation between priming effect size and degree of phonological overlap was significant (r ⫽ .17, p ⬍ .01, N ⫽ 720) for Experiments 2 and 3 (and approximately the same size in each experiment). Primes that shared more phonemes with targets (for a fixed number of shared letters) produced stronger priming effects. The significant correlation suggests that part of the observed differences in priming effects across the different prime conditions tested in Experiments 2 and 3 is due to differences in the level of phonological overlap across primes and targets in these different conditions. Word Frequency and Bigram Frequency In this analysis we examined whether the pattern of priming effects varied as a function of either target word frequency or the mean positional bigram frequency of prime stimuli. It is possible that priming effect sizes, especially in the outer-central prime conditions of Experiments 2 and 3, vary as a function of the familiarity of the letter sequences that they are composed of. This might explain why initial letter and final letter primes (composed of completely contiguous letter sequences from the target word) were more effective than outer-central primes (composed of noncontiguous letter sequences), since the latter will generally have lower bigram frequencies. The mean positional bigram frequency of prime stimuli did not correlate with priming effect size across Experiments 2 and 3 (r ⫽ .05, N ⫽ 720), and none of the correlations done separately for Experiment or word length were close to being significant. Similarly, there was a nonsignificant correlation between target word frequency and priming effect size (averaged across the three prime conditions) for both experiments (r ⫽ .08, N ⫽ 240), and again this correlation was not significant when calculated separately for Experiment or word length. These post hoc analyses clearly show that neither the bigram frequency of prime stimuli nor target word frequency is having any significant influence on priming effects sizes in Experiments 2 and 3. As a further test of any possible influence of target word frequency on priming effects in Experiments 2 and 3, the target words were split in two halves according to their printed frequencies (mean frequency ⫽ 21 per million and 97 per million respectively, New et al., 2001). An ANOVA was performed on RTs to word targets (both 7-letter and 9-letter targets) with prime type and target frequency as within-participant factors and experiment as a between-participants factor. There was an effect of frequency, F(1, 118) ⫽ 26.5, p ⬍ .01, an effect of prime type, F(3, 177) ⫽ 18.34, p ⬍ .01, and no interaction between prime type and frequency (F ⬍ 1). The ANOVA therefore confirms the correlational analysis showing no significant influence of target word frequency on priming effect sizes in Experiments 2 and 3. 871 Discussion The results of these post hoc analyses therefore suggest that the stronger priming observed for the initial and final primes conditions relative to the outer-central prime condition, observed in Experiments 2 and 3, may be due to the higher levels of primetarget phonological overlap in the former conditions. The post hoc analyses showed no significant influence of the other variables we examined, including the relative proportion of consonants and vowels present in the prime stimulus. However, Perea and Lupker (2004) found an influence of CV status on priming from nonadjacent transposed-letter primes (e.g., caniso-CASINO), observing nonsignificant priming when the two transposed letters were vowels. Our correlation, although not significant, goes in the same direction, with weaker priming as the proportion of vowels increases. Clearly, a direct manipulation of the proportion of consonants and vowels shared by prime and target is necessary to clarify the role of this factor in relative-position priming. Given the significant correlation between prime-target phonological overlap and mean RT per item in Experiments 2 and 3, it is possible that it is phonological and not orthographic overlap that is the main driving force behind the pattern of results obtained with this task. We believe that this is unlikely given that Peressotti and Grainger (1999) found strong relative-position priming with 33 ms prime durations, in conditions where Ferrand and Grainger (1992, 1994) had systematically failed to find any evidence for phonological priming in French. In order to be completely sure about the pattern of results obtained with the priming conditions tested in Experiments 2 and 3, we ran two further experiments using reduced (33 ms) prime exposures. Here we do not expect to observe any correlation with phonological overlap and priming effect size. Furthermore, since prior experiments showing an outer letter advantage have used degraded target stimuli (e.g., Humphreys et al., 1990, used the perceptual identification task), an obvious next step in our investigation was to test the same priming conditions in a masked priming experiment where targets are masked as well as primes. However, it could also be argued that it is the presence of pattern masking in the present studies that reduces the inherent advantage of outer letters. Thus, in order to test these different possibilities, Experiment 4 uses the four-field masking perceptual identification paradigm of Humphreys et al. (1990), and Experiment 5 uses the same paradigm as our previous experiments but with a 33 ms prime duration, and including a condition with pattern masking of the prime stimulus and a condition where no pattern mask is present. Experiment 4 Experiment 4 uses the same stimuli and priming conditions as in Experiment 3 with a perceptual identification task, as used in the work of Humphreys et al. (1990). Method Participants. Thirty-two students at the University of Provence took part in this experiment. They all reported being native speakers of French with normal or corrected-to-normal vision, and had not participated in the previous experiments. Stimuli and design. These were the same as in Experiment 3. GRAINGER ET AL. 872 Procedure. A four-field masking procedure was employed similar to that used by Humphreys et al. (1990). On each trial, participants first saw a forward mask (a row of 11 hash marks) for 500 ms, immediately followed by a prime for 33 ms, then by the target word for 33 ms, and finally by another row of hash marks for another 500 ms. Participants were instructed to identify word stimuli presented in uppercase letters and they typed in their response using the computer keyboard. Primes and targets were displayed in the same font and size as in Experiment 3 except that primes were in lower case and targets in upper case in the present experiment. Results Mean percent correct target identification for each experimental condition is given in Table 5. An ANOVA on these percentage scores showed a main effect of prime type (F1(3,84) ⫽ 13.99, p ⬍ .0001; F2(3,354) ⫽ 7.11, p ⬍ .0001), and an effect of length (F1(1,28) ⫽ 48.53, p ⬍ .0001; F2(1,118) ⫽ 16.86, p ⬍ .0001). There was no effect of list and no interactions between these three factors. Pairwise comparisons using Dunnett’s test revealed significant differences between the unrelated condition and the initial prime condition (t1(31) ⫽ 4.09, p ⬍ .01; t2(119) ⫽ 5.07, p ⬍ .01), the final prime condition (t1(31) ⫽ 4.86, p ⬍ .0; t2(119) ⫽ 3.08, p ⬍ .01), and the outer-central prime condition (t1(31) ⫽ 5.23, p ⬍ .01; t2(119) ⫽ 3.78, p ⬍ .01). Planned comparisons across the related prime conditions showed no significant differences. Discussion In the perceptual identification task of Experiment 4, there is no evidence for any positional biases in relative-position priming effects. Priming effects were very similar for the three priming conditions and the two target word lengths, and all highly significant compared with the all-different letter primes. The perceptual identification task has generated stronger priming effects for the outer-central primes compared to Experiment 3, since these are now significant relative to the unrelated condition. However, outer-central primes (containing both of the targets’ outer letters) still did not generate stronger priming than either initial or final primes (containing only one of the target’s outer letters). A correlational analysis on priming effects and degree of prime-target phonological overlap showed that this was not significant in Experiment 4 (r ⫽ .03). Comparing the results of Experiments 3 and 4, where the same set of stimuli were tested in a lexical decision and a perceptual identification task, it seems that only the outer-central priming Table 5 Mean Percent Correct Word Identification Scores for the Different Priming Conditions Tested in the Perceptual Identification Task of Experiment 4 With 33 ms Prime Exposures 9-letter words 1234 52.4* 1234 41.1* 4567 50.8* 6789 39.8* 1357 52.3* 1469 41.1* Experiment 5 Method Participants. Eighty students at the University of Provence took part in this experiment. Thirty-six were tested in the condition where primes were masked, and 44 tested in the condition where no masks were used. They all reported being native speakers of French with normal or corrected-tonormal vision, and had not participated in the previous experiments. Stimuli and design. These were the same as in Experiment 3 except for the masking factor. In one condition prime stimuli were masked, as in Experiment 3, and in another condition no pattern mask was presented. Two independent groups of participants were tested in these two masking conditions. Procedure. This was the same as in Experiment 3 except for the use of a prime exposure duration of 33 ms and the absence of any masking stimulus in one condition. Results Priming condition 7-letter words condition was affected by task. This priming condition produced nonsignificant effects in the lexical-decision task, and robust priming in perceptual identification. It is possible that the perceptual identification task is more sensitive to the influence of outer letters in prime stimuli, as attested by the effects of this factor reported by Humphreys et al. (1990), and this compensates for the disadvantage for these primes relative to the initial and final prime conditions. There are two possible sources of this hypothesized disadvantage for outer-central primes compared to initial and final prime conditions. These are 1) as shown above, the lower level of phonological overlap across primes and targets in the outer-central prime condition; and 2) the lower level of contiguity (the number of adjacent letter combinations in the target that are also adjacent in the prime stimulus) in the outer-central prime condition. In the General Discussion we will present preliminary explorations of contiguity effects in one recent account of letter position coding (Grainger & van Heuven, 2003). However, at present the most straightforward and most conservative conclusion is that when phonological influences on priming effects are minimized then the different priming conditions tested in Experiments 2– 4 show statistically equivalent effects. The reason why phonological influences were neutralized in Experiment 4 is likely due to the shorter prime exposure duration used in this experiment rather than the particular task used. In order to test this, the following experiment uses the procedure of Experiment 3 (with lexical decision on target words), but with a shorter (33 ms) prime duration. The possible influence of pattern masking on relative-position priming effects is further studied in Experiment 5 by manipulating the presence or absence of a forward mask. dddd 43.1 dddd 31.5 * Significantly different from the dddd condition, p ⬍ .01 by participant using Dunnett’s test. The condition means are given in Table 6. An ANOVA on mean correct RTs to word targets showed a main effect of prime type (F1(3,216) ⫽ 16.52, p ⬍ .00001; F2(3,336) ⫽ 15.00, p ⬍ .00001), an effect of length (F1(1,72) ⫽ 117.12, p ⬍ .00001; F2(1,112) ⫽ 28.73, p ⬍ .00001) and an effect of masking (F2(1,112) ⫽ 23.75, p ⬍ .00001). There was an interaction between prime type and masking (F1(3,216) ⫽ 5.25, p ⬍ .01; F2(3,336) ⫽ 4.34, p ⬍ .01) and between length and masking (F1(1,72) ⫽ 4.17, p ⬍ .05; F2(1,112) ⫽ 6.59, p ⬍ .05). RELATIVE-POSITION PRIMING Table 6 Mean Response Times (RT in ms) and Percentages of Error (PE) for the Different Priming Conditions Tested in Experiment 5 With 33 ms Prime Exposures With Masking Priming condition 7-letter words RT PE 9-letter words RT PE 1234 536 1.5 1234 574 1.3 Without Masking 4567 547 0.7 6789 566 1.5 1357 535 0.7 1469 565 2.6 dddd 552 1.9 dddd 572 3.5 Priming Condition 7-letter words RT PE 9-letter words RT PE 1234 513* 0.8 1234 555* 1.7 4567 519* 1.7 6789 564* 1.2 1357 518* 1.2 1469 558* 1.5 dddd 549 2.3 dddd 581 2.0 * Significantly different from the dddd condition, p ⬍ .05 by participant using Dunnett’s test. In the mask condition, there was a main effect of length (F1(1,32) ⫽ 64.98, p ⬍ .00001; F2(1,112) ⫽ 16.37, p ⬍ .001), no effect of prime type, and no interaction. Comparisons against the unrelated condition using Dunnett’s test showed that only the outer-central condition produced significantly faster RTs in the analysis by participants (t1(35) ⫽ 2.47, p ⬍ .05) when primes were masked. In the no-mask condition, on the other hand, there was a main effect of prime type (F1(3,120) ⫽ 21.76, p ⬍ .00001; F2(3,336) ⫽ 19.51; p ⬍ .00001), an effect of length (F1(1,40) ⫽ 67.17, p ⬍ .00001; F2(1,112) ⫽ 36.23, p ⬍ .00001), and no interaction. Dunnett’s test indicated that the unrelated prime condition produced significantly longer RTs than the initial condition (t1(43) ⫽ 6.22, p ⬍ .01; t2(119) ⫽ 6.19, p ⬍ .01), the final condition (t1(43) ⫽ 4.65, p ⬍ .01; t2(119) ⫽ 4.18, p ⬍ .01), and the outer-central condition (t1(43) ⫽ 5.79, p ⬍ .01; t2(119) ⫽ 5.49, p ⬍ .01). An ANOVA on percentages of error to word targets revealed a significant main effect of prime type (F1(3,216) ⫽ 5.25, p ⬍ .01; F2(3,336) ⫽ 4.02, p ⬍ .01) and an effect of length by participant (F1(1,72) ⫽ 4.19, p ⬍ .05) that failed to reach significance in the item analysis (F2(1,112) ⫽ 3.58, p ⬍ .07). There was no effect of masking and there were no significant interactions. Dunnet’s test revealed significant differences between the unrelated condition and the initial prime condition (t1(79) ⫽ 2.86, p ⬍ .05; t2(119) ⫽ 2.64, p ⬍ .05), the final prime condition (t1(79) ⫽ 2.93, p ⬍ .05; t2(119) ⫽ 2.61, p ⬍ .05), and the outer-central prime condition in the analysis by participants (t1(79) ⫽ 2.54, p ⬍ .05). An analysis of the mean correct RTs to nonword targets showed a main effect of length (F1(1,72) ⫽ 160.46, p ⬍ .0001; F2(1,112) ⫽ 56.06, p ⬍ .00001) and an effect of masking in the item analysis (F2(1,112) ⫽ 41.44, p ⬍ .00001). RTs were slower with the longer nonwords and in the condition where primes were masked. An ANOVA on percentages of error to nonword targets revealed a significant main effect of length (F1(1,72) ⫽ 15.64, p ⬍ .001; F2(1,112) ⫽ 4.00, p ⬍ .05). There was no effect of prime type or masking, and no interaction. 873 Discussion Experiment 5 successfully replicated the pattern of priming effects obtained with the perceptual identification task in Experiment 4 in conditions where prime stimuli were presented very briefly (33 ms) and with no pattern masking. In line with the results of Experiment 4, the outer-central primes now show just as much priming as initial and final prime conditions. Also in line with the results of Experiment 4 is the fact that there was no hint of a correlation between prime-target phonological overlap and priming effect size in either the mask or no-mask conditions tested in Experiment 5 (r ⫽ .006 and 0.015, respectively). This implies that it is not the particular masking conditions nor the type of task that was essential for obtaining the pattern of priming effects in Experiment 4. It appears that presentation conditions that allow orthographic effects to emerge in the absence of phonological effects are sufficient for obtaining priming effects that are unaffected by positional biases. The presence of a pattern mask significantly affected the amount of priming observed in Experiment 5. In the RT analysis, priming effects were only robust in the absence of a pattern mask, although priming effects in the error data were not affected by masking. This is contrary to previous reports of relative-position priming at such short prime durations (Peressotti & Grainger, 1999). However, the level of orthographic overlap across prime and target was lower in the present study, suggesting that this is one factor that determines whether or not significant priming effects will be obtained. The present experiment also shows that whether or not the prime stimulus is masked is another critical factor, and other visual factors such as the type of pattern mask, and the size and luminance of mask, prime, and target stimuli, could also be critical (e.g., Frost, Ahissar, Gotesman, & Tayeb, 2003). With all other factors held constant, the presence of a forward mask significantly affects the amount of orthographic priming that is obtained. Combined Analysis of Experiments 2, 3, and 5 Although Experiment 5 once again failed to show a significant difference between initial primes and the other related prime conditions, this is the fourth experiment to show a small nonsignificant advantage for initial primes compared with final primes. We therefore performed an ANOVA by participants combining the data of Experiments 2, 3, and the unmasked condition of Experiment 5. (Experiment 4 was not included because a different dependent variable had been used.) This combined analysis showed that initial primes generated RTs that were on average 7 ms faster than RTs to targets following final primes, F(1, 141) ⫽ 7.74, p ⬍ .01, and this effect was stable across experiments (F ⬍ 1 for partial interaction). On the other hand, the 13 ms advantage for targets following initial primes compared to outer-central primes, F(1, 141) ⫽ 34.13, p ⬍ .001, did interact with Experiment, F(2, 141) ⫽ 4.27, p ⬍ .05. Final primes were also significantly faster than RTs to targets following outer-central primes, F(1, 141) ⫽ 5.17, p ⬍ .05, and this also interacted with Experiment, F(2, 141) ⫽ 4.04, p ⬍ .05. Therefore, there is evidence for a small but systematic advantage when primes are formed of a contiguous sequence of beginning letters compared to when they are formed of a contiguous sequence of final letters. Differences relative to the noncontiguous (outer-central) prime condition were, however, not stable GRAINGER ET AL. 874 across experiments, probably as a result of the influence of prime duration on these particular priming effects. Finally, since the three priming conditions differ in terms of their phonological overlap with targets, we performed an ANCOVA on the data of Experiments 2, 3, and 5, with percent phonological overlap as a covariate. Only the three related prime conditions were included in this analysis as a further test of differences among these three priming conditions. In the by-items ANOVA including these three conditions (i.e., without the unrelated prime condition), there was a main effect of prime type, F(2, 708) ⫽ 5.34, p ⬍ .01 that did not interact with Experiment, F(4, 708) ⫽ 1.42. This therefore confirms the significant differences across priming conditions found in the planned comparisons in the by-participants ANOVA described above. However, in the ANCOVA this main effect of prime type was no longer significant, F(2, 706) ⫽ 2.69, and again did not interact with Experiment, F(4, 706) ⫽ 1.28. Furthermore, the critical pairwise comparison between word-initial and word-final primes was not significant in the ANCOVA, F(1, 353) ⫽ 1.14. This result strongly suggests that differences across the three related priming conditions, including the word-initial prime advantage, are being driven by differences in the amount of phonological overlap between primes and targets. Experiment 6 Experiment 5 has provided evidence for relative-position priming effects in conditions where it is unlikely that phonological prime-target overlap is significantly affecting processing. Experiment 6 was designed to test for the absence of phonological processing using a traditional measure of such influences: priming from pseudohomophones (e.g., “brane” as a prime for the target word “brain”). Prior research in French has systematically shown effects of orthographic priming and no pseudohomphone priming at very brief prime durations (⬍50 ms), while pseudohomophone priming effects emerge at longer prime durations (Ferrand & Grainger, 1992, 1993, 1994; Grainger, Diependaele, Spinelli, Ferrand, & Farioli, 2003; Ziegler, Ferrand, Jacobs, Rey, & Grainger, 2000). Experiment 6 tests relative-position priming and pseudohomophone priming at 33 ms and 83 ms prime durations with no pattern masking. Experiment 6 also provides a further exploration of the relative importance of outer versus inner letters in relative-position priming using a manipulation similar to Experiment 1b but with 4-letter primes. Method Participants. Twenty-eight students at the University of Provence took part in this experiment. They all reported being native speakers of French with normal or corrected-to-normal vision, and had not participated in the previous experiments. Stimuli and design. A new set of sixty 7-letter French words with printed frequencies ranging from 1 to 313 occurrences per million (M ⫽ 29; New et al., 2001) was generated with the constraint that two different pseudohomophones could be generated from each word: one that maximized orthographic overlap with its corresponding baseword (one different letter, e.g., silance-silence), and one that minimized orthographic overlap (at least two different letters, e.g., cylance-silence). This represents the “orthographic overlap” factor (minimal vs. maximal). Each pseudohomophone prime was matched to an orthographic control prime in terms of the number of shared letters and the position of these letters (e.g., “silunce” for the pseudohomophone “silance”), and every prime stimulus was tested at two different prime durations (33 ms and 83 ms). Thus orthographic overlap was crossed with prime type (pseudohomophone vs. orthographic control) and prime duration in a 2 ⫻ 2 ⫻ 2 factorial design that forms the “phonological priming” subpart of Experiment 6. The same set of target words and nonwords were also tested with relative-position primes formed of the first letter, last letter, and the 3rd and 5th letters of targets in three different orders (1357, 1537, 7351) and an unrelated 4-letter prime (dddd), also tested at two prime durations (see Table 1). Thus the “relative-position priming” subpart of Experiment 6 involved the manipulation of prime type and prime duration in a 4 ⫻ 2 factorial design. Given the strong constraints on stimulus selection and the need for both a within-participants and a within-items design (i.e., the same targets tested in both the phonological priming and the relative-position priming conditions), target repetition could not be avoided in Experiment 6. The counterbalancing procedure applied in Experiments 1–5 was again used to form four lists presented to four independent groups of participants. Within each list, each target appeared four times paired with one of the four prime types associated with each subpart of the experiment (i.e., one of the four possible primes in the phonological priming subpart, and one of the four possible primes in the relative-position priming subpart, determined by the standard counterbalancing procedure), and at both prime durations. Sixty 7-letter nonword targets were constructed such that an orthographically similar (one letter different) and an orthographically dissimilar (at least two letters different) pseudohomophone prime could be created for each target. The nonwords were tested in exactly the same priming conditions as the word targets. Thus each participant received 480 trials, half of which had word targets. Procedure. This was the same as the condition with no pattern mask in Experiment 5 except for the use of two prime exposure durations: 33 ms and 83 ms. Results The condition means are given in Table 7. Separate ANOVAs were performed for the relative-position priming conditions and the pseudohomophone priming conditions. Relative-position priming. The ANOVA on mean correct RTs to word targets showed a main effect of prime duration (F1(1,27) ⫽ 40.72, p ⬍ .00001; F2(1, 58) ⫽ 44.24, p ⬍ .00001), no effect of prime type (F1 ⬍ 1; F2 ⬍ 1), and a trend to an interaction in the item analysis (F1(3,81) ⫽ 1.58; F2(3,174) ⫽ 2.58, p ⫽ .056). As can be seen in Table 7, prime type was only influencing RT at the 33 ms prime duration, F(3, 81) ⫽ 6.78, p ⬍ .01; F2(3,174 ⫽ 6.89, p ⬍ .01), but not at the 83 ms prime duration (both Fs ⬍1). The 1357 prime condition was significantly faster than the three other prime conditions with prime exposures of 33 ms, F(1, 27) ⫽ 8.94, p ⬍ .01; F2(1,58 ⫽ 4.85, p ⬍ .05). This priming effect did not interact with target word repetition, and the triple interaction between prime type, prime duration, and target repetition was not significant (all Fs ⬍1). None of the main effects or interactions were significant in an analysis of the percent errors to word targets. An analysis of the mean correct RTs to nonword targets revealed an effect of prime type (F1(3,81) ⫽ 6.04, p ⬍ .001; F2(3,177) ⫽ 5.34, p ⬍ .01) that interacted with prime duration (F1(3,81) ⫽ 7.02, p ⬍ .001; F2(3,177) ⫽ 6.48, p ⬍ .001). Significant priming effects only appeared at the 83 ms prime duration (1357 ⫽ 606 ms, 1537 ⫽ 635 ms, 7351 ⫽ 626 ms, dddd ⫽ 655 ms). There were no significant effects in the error data for nonword targets. RELATIVE-POSITION PRIMING Table 7 Mean Response Times (RT in ms) and Percentages of Error (PE) for the Different Priming Conditions and the Two Prime Durations (33 ms and 83 ms) of Experiment 6 Testing for Relative-Position Priming and Phonological Priming With Different Levels of Phonological (P⫹/P⫺) and Orthographic (O⫹/O⫺) Overlap Across Primes and Targets Relative-Position priming 33 ms RT PE 83 ms RT PE Priming Condition 1357 577* 3.3 1537 589 3.1 7351 590 4.8 dddd 597 4.1 613 4.5 613 4.5 619 2.9 610 5.0 Priming Condition O⫹ Phonological priming 33 ms RT PE 83 ms RT PE O⫺ P⫹ P⫺ P⫹ P⫺ 581 3.3 584 2.4 590 4.3 593 4.1 564 4.8 570 4.3 604 5.0 633 4.1 * Significantly different from the dddd condition, p ⬍ .05 by participant using Dunnett’s test. Phonological priming. The ANOVA on mean correct RTs to word targets showed a main effect of prime type in the analysis by participant (F1(1,27) ⫽ 4.27, p ⬍ .05; F2(1,58) ⫽ 1.62), with pseudohomophone primes generating faster RTs than their orthographic controls. Orthographically similar primes generated faster RTs than the dissimilar primes (F1(1,27) ⫽ 57.29, p ⬍ .0001; F2(1,58) ⫽ 40.87, p ⬍ .0001), and there was a trend to an effect of prime duration in the items analysis (F1(1,27) ⫽ 2.53; F2(1,58) ⫽ 3.11, p ⬍ .10). The effects of orthographic overlap interacted with prime duration (F1(1,27) ⫽ 23.42, p ⬍ .0001; F2(1,58) ⫽ 27.61, p ⬍ .0001), since the effects of this factor were much larger at the 83 ms prime duration. There was a significant triple interaction in the participants’ analysis (F1(1,27) ⫽ 4.29, p ⬍ .05; F2(1,58) ⫽ 2.46), reflecting the fact that the two-way interaction between prime duration and prime type was significant (in the analysis by participants) for the orthographically dissimilar pseudohomophones (F1(1,27 ⫽ 4.18, p ⬍ .05) but not for the orthographically similar pseudohomophones (both Fs ⬍ 1). Pairwise comparisons revealed that only orthographically dissimilar pseudohomophones generated a significant priming effect, and only at the 83 ms prime duration (F1(1,27) ⫽ 11.06, p ⬍ .01; F2(1,58) ⫽ 6.51, p ⬍ .05). None of the other comparisons were significant. Neither the effects of prime type nor the effects of orthographic overlap interacted with target word repetition (both Fs ⬍ 1). None of the main effects or interactions were significant in an analysis of the percentages of error to word targets. An analysis of the mean correct RTs to nonword targets showed a trend to an effect of prime duration (F1(1,27) ⫽ 4.13, p ⬍ .10; F2(1,58) ⫽ 5.66, p ⬍ .05).There was a main effect of orthographic overlap (F1(1,27) ⫽ 24.52, p ⬍ .0001; F2(1,58) ⫽ 23.62, p ⬍ 875 .0001), with orthographically more similar primes generating slower RTs. There were no significant effects in the error data for nonword targets. Discussion The results of Experiment 6 replicate those of Experiment 5, showing relative-position priming effects with short prime durations (33 ms) and no pattern masking, while demonstrating that phonological priming effects are absent in the same testing conditions. This therefore adds direct support to the evidence provided by nonsignificant correlations between prime-target phonological overlap and priming effect sizes found with short prime durations (Experiments 4 & 5). In line with this reasoning, phonological priming did emerge with longer prime durations (83 ms) in Experiment 6. This in itself is an important replication of prior research showing priming from pseudohomophone stimuli, particularly since Experiment 6 tested pseudohomophone priming with longer words than typically tested in phonological priming experiments. On the other hand, the effects of relative-position primes actually diminished and were nonsignificant at the longer prime duration. The level of phonological overlap of the noncontiguous relative-position primes tested in Experiment 6 was clearly not high enough to generate priming at the longer prime duration. However, the fact that higher levels of orthographic overlap generated stronger priming effects at this prime duration suggests that some form of orthographic priming continues to develop with increasing prime exposures. It appears that single-letter substitution primes (the O ⫹ primes tested in the phonological priming subpart of Experiment 6) are generating maximum facilitation (relative to the O- primes) that grows with increasing prime exposure duration. The relative-position primes, on the other hand, generate a weaker and more short-lived facilitation. One possible explanation for the distinct time-course of relativeposition and substitution priming effects is in terms of the different degrees of phonological overlap with target words. Relativeposition primes formed of noncontiguous letter sequences (as was the case in Experiment 6, e.g., slne-silence) will tend to generate phonological codes that are incompatible with the target word’s phonology, while this is not the case with single-letter substitution primes (e.g., silunce-silence). Assuming that the phonological code for a word requires precise order information, then the prime stimulus “slne” will only have one phoneme (/s/) that is compatible with the French target word “silence,” whereas the substitution prime stimulus “silunce” will have four out of five compatible phonemes. Therefore, as prime duration increases, it appears that orthographic prime-target overlap continues to facilitate target word processing only to the extent that there is also a minimal level of compatibility between the phonological code generated by the prime stimulus and the target’s phonological representation. We return to this point in the section on contiguity effects in the General Discussion. The results of Experiment 6 show once again no sign of an outer letter advantage in the testing conditions of the present study. Primes that maintained the target’s outer letters (1537 primes) did not differ from primes in which the two outer letters were transposed (7351) primes. Finally, Experiment 6 provided further evidence (following the post hoc analyses of Experiments 2 and 3) that letter cluster frequency is not influencing the size of relative- GRAINGER ET AL. 876 position priming effects. The mean positional bigram frequency of the 1357, 1537 and 7351 primes were respectively 689, 922, and 832 occurrences per million (New et al., 2001). The observed advantage for primes respecting relative-position information cannot therefore be attributed to prime stimuli having higher bigram frequencies in this particular condition. General Discussion The present results speak to several key issues in printed word perception. First, they provide further evidence for the role of individual letter representations in this process (Besner, et al., 1984; Evett & Humphreys, 1981; Rayner et al., 1980). Any possible influence of global shape information from lower-case primes would be greatly reduced in the relative-position primes compared to absolute position primes. Probably one of the most critical sources of global information about a word is its length, and this information was incorrect in the relative-position prime condition. This is in line with recent eye-movement research suggesting that length information is used essentially for nonlinguistic purposes (visual object selection and saccade specification) during reading (Inhoff, Radach, Eiter, & Juhasz, 2003). Furthermore, relative-position primes modified the visual configuration formed of ascending, descending, and neutral letters, thus again providing information that was incompatible with the upcoming target word’s global shape. Our results show significant priming from very briefly presented, pattern-masked prime stimuli when they are composed of a subset of the upcoming target’s component letters. This subset of letters can be quite small. In Experiments 3–5, significant priming was obtained when primes shared only four letters with 9-letter target words. However, one key result of the present study is that when primes are composed of only a subset of the target’s letters, then letter order is critical for obtaining significant priming. Experiment 1 replicated the basic findings of Humphreys et al. (1990) and Peressotti and Grainger (1999). With primes that shared five letters with 7-letter target words, significant priming arose only when the prime letters respected their order in the target word, and inserting hyphens to provide length-dependent, absolute position information did not increase the amount of priming. This pattern was replicated with a different set of target words in Experiment 6 (relative-position priming) with no pattern masking. However, relative-position priming was only robust at the shortest prime duration (33 ms) in this experiment, and no longer influenced performance to target words at the 83 ms prime duration. These results provide a solid empirical foundation for relative-position priming that will be important for developing our understanding of how letter position information is coded during printed word perception. Positional Biases in Masked Priming Position of overlap, in terms of whether the prime letters were mainly from the beginning or from the end of target words, did not have a major influence on the results obtained in the present experiments. In each individual experiment, primes containing the initial letters of the target word did not provide significantly greater facilitation than primes containing the target’s final letters. Nevertheless, a combined analysis of three different experiments did reveal a significant word-beginning advantage, and this advantage, relative to word-final primes, was systematic across the three experiments. This pattern fits with the results obtained with eyemovement recordings (parafoveal priming) that showed a systematic advantage when parafoveal previews share the target word’s initial letters compared to when only word-final letters are shared (e.g., Briihl & Inhoff, 1995; Inhoff, 1989). However, further analyses of this word-initial advantage showed that it is likely due to a confound with level of prime-target phonological overlap. Including this variable as a covariate removed the significant advantage of word-initial primes. Therefore, the most parsimonious account of the present results is that positional biases are minimal in conditions where the influence of phonological factors is minimized. This fits with a model of orthographic processing that allows the parallel identification of a string of letters (Grainger & van Heuven, 2003). However, the results do not exclude the possibility of serial (beginning-to-end) processing of phonological information. Our correlational analyses of priming effects obtained at 50 ms prime durations in Experiments 2 and 3 suggested a small but significant phonological influence on priming effect size. (Priming effects increased with an increase in the number of phonemes shared by prime and targets.) Indeed, in other studies using slightly longer prime durations, robust effects of phonological priming have been systematically found in the masked priming paradigm (e.g., Ferrand & Grainger, 1992, 1994; Grainger et al., 2003; Ziegler et al., 2000). This was confirmed in Experiment 6 where phonological priming was evaluated by comparing effects of pseudohomophone primes with nonhomophonic orthographic control primes. Phonological priming was absent at the shorter (33 ms) prime duration, and emerged at the longer (83 ms) prime duration in this experiment. Recently, Carreiras, Ferrand, Grainger, and Perea (2005) found evidence for sequential processing of phonology using longer words than traditionally tested in studies of masked phonological priming. Carreiras et al. found priming for initial phonological overlap (the first syllable of bisyllabic words) and no priming for final phonological overlap (the last syllable of bisyllabic words), and this pattern was obtained in both the naming and lexical decision tasks. Since the length of words tested by Carreiras et al. was similar to the length of words in the present study, it can be tentatively concluded that fast parallel orthographic processing precedes the sequential computation of a sublexical, phonological code. The other positional bias investigated in the present study concerns a possible advantage for outer letters compared with inner letters. Contrary to a number of studies, the present work provided no evidence in favor of an outer letter advantage in visual word recognition. Using a masked priming paradigm similar to that used in the present study, Humphreys et al. (1990) found that primes containing both of the target’s outer letters tended to produce stronger priming than primes that contained only one outer letter or neither of the target’s outer letters. However, these differences were always marginal and not always in the appropriate direction. For example, in Humphreys et al.’s Experiment 1a, the ssds prime condition was actually numerically worse than the sssd and dsss prime conditions (where “s” is a letter shared by prime and target, and “d” is a different letter). It is in their Experiment 1b that Humphreys et al. found the strongest evidence for an outer letter advantage with prime condition sdds generating significantly RELATIVE-POSITION PRIMING higher levels of target identification than either of the following conditions: ssdd, dssd, ddss. Given that Experiments 2 and 3 of the present study showed exactly the opposite effect (in some conditions primes containing both of the targets’ outer letters produced less priming than primes containing a single outer letter), it was tentatively hypothesized that the perceptual identification task might be more sensitive to outer letter effects. However, using the same paradigm as Humphreys et al. (1990), Experiment 4 of the present study found no evidence for an outer letter advantage. We would therefore suggest that the type of priming (substitution priming in Humphreys et al.’s experiments vs. relative-position priming in our experiments) could be the source of the discrepancy. Substitution primes include unrelated letters (that replace one or more of the target’s letters), and it may well be the position of the two unrelated letters in Humphreys et al.’s (1990) Experiment 1b that determined the pattern they observed. However, robust and systematic advantages for outer letters have been obtained with other paradigms. For example, in a recent study Jordan et al. (2003) presented participants with words that had two of their letters degraded by spatial filtering (producing a blurred version of the letter). In an otherwise normal reading situation, they found that reading rate was significantly faster when two inner letters were degraded compared to when the two outer letters were degraded. Nevertheless, an analysis of the complete pattern of data presented by Jordan et al. (2003) shows that their data only partly support the hypothesis that outer letters enjoy a privileged status in visual word recognition. The condition where both outer letters were intact (interior letters degraded) did not improve reading rate compared to a condition where only one outer letter was intact (when the first two letters were degraded). So these results are compatible with the present data and are not compatible with a straightforward outer letter advantage in visual word recognition.2 Letter Position Coding The central aim of the present study was to provide further critical data to inform our ongoing attempt to describe how letter position information is coded during printed word perception. Some standard letter position coding schemes were mentioned in the introduction. The length-dependent position-specific coding scheme of the interactive-activation model (McClelland & Rumelhart, 1981) codes the position of each letter at a specific position in a string of a given length. This is a highly efficient but very uneconomical means of coding letter position, requiring a large number of duplications of letter representations for each lengthdependent position in a word. The number of duplications can be reduced by introducing one or more anchor points in the coding scheme. Thus in Coltheart et al.’s (1993; 2001) DRC model, the beginning of a word forms the anchor point for coding letter position independently of word length. In the multiple read-out model with phonology (MROM-p) Jacobs, Rey, Ziegler and Grainger (1998) used the first and last letters as anchor points for position coding, such that inner letters are either coded as being n positions from the beginning, or n positions from the end of the string. Another possible means of coding letter-in-string position was adopted by Seidenberg and McClelland (1989). In this scheme, words are coded as unordered sets of letter triples where the space before and after the word is included as a character. Thus 877 the word “MAKE” is coded as #MA, MAK, AKE, KE#, where the hash mark represents a space. The order of the letters in each triplet is critical, but no order information is provided for the triplets themselves. As noted in the introduction, none of these approaches to letter position coding can account for the relative-position priming effects that were replicated and extended in the present study. A number of more recent models of letter position coding can account for the basic phenomenon of relative-position priming (see Davis & Bowers, 2004; Grainger & van Heuven, 2003; & Perea & Lupker, 2004, for more detailed descriptions of recent coding schemes). In spatial coding schemes, the relative position of a spatially distributed set of items is coded in terms of their relative activation level. This is best achieved when the items in the list form a monotonically increasing or decreasing set of activation values, referred to as an activation gradient (Grossberg, 1978). For the purposes of letter position coding, the activation gradient must form a monotonically decreasing activation function across letter position with the highest value for initial letters and the lowest value for the final letter of the string. This is the case in the SOLAR model (Davis, 1999), and the SERIOL model of Whitney (2001), where a set of position-independent letter detectors are activated by the orthographic input, and the relative position of these letter identities is coded by the relative activation levels of the corresponding detectors. In the SERIOL model, relative activation values at the letter level are used to activate noncontiguous bigram units, to be discussed in more depth below. In the SOLAR model, information about letter identity and letter order serves as input to an equation that determines the match between incoming information and whole-word orthographic representations in longterm memory3. This match calculation, which is admittedly only part of the complex set of computations used by the SOLAR model, actually operates like a fuzzy slot-based coding scheme. In this particular case, letter order is given from beginning-to-end (as in the DRC model, Coltheart et al., 2001), with Gaussian noise added at each position. The match value is a function of the degree of displacement of letters shared across input and the bestmatching stored representation. In the Appendix we provide the match calculations generated by the SOLAR and SERIOL models for the priming conditions tested in the presented experiments. The central idea in Grainger and van Heuven’s (2003) approach is that higher-level sublexical representations code for the presence of contiguous and noncontiguous letter sequences at lower levels of processing (i.e., the alphabetic array in Figure 1). The principle of coding the relative position of nonadjacent letters is also used in Whitney’s (2001) model where, for example, the word “CART” is coded as the following set of bigrams: CA, CR, CT, AR, AT, RT. Thus bigrams are formed across adjacent and nonadjacent letters in the correct order, the basis of what Grainger and van Heuven 2 Ongoing work in our laboratory is currently exploring possible differences in outer versus inner letters in relative-position priming. The results at present suggest that primes with no outer letter (e.g., 23456 for a 7-letter target) are at a disadvantage compared with primes that contain at least one outer letter (e.g., 13456, 23457). 3 Information about the match calculation equation, the SOLAR model, and the match calculator program used to calculate the match values for the SOLAR model can found at http://www.maccs.mq.edu.au/⬃colin. 878 GRAINGER ET AL. Figure 1. Functional architecture for orthographic processing. A letter string is first processed by a bank of alphabetic character detectors (the alphabetic array). The next level of processing combines information from different processing slots in the alphabetic array to provide a relative position code for letter identities. (The example shows an unconstrained version of Grainger & van Heuven’s open-bigram coding.) These relativeposition coded letter identities control activation at the level of whole-word orthographic representations (O-words) via bidirectional excitatory connections with all units at the relative position level. (2003) later referred to as “open-bigram” coding. The critical difference between our account and the one developed by Whitney (2001), concerns the mechanism used to activate the appropriate open-bigrams. Whitney (2001) described a method for translating acuity-dependent activations of letter representations into a monotonically decreasing gradient. Relative activation at the letter level is then transformed into activation of ordered bigram units. Our approach is more in the tradition of Mozer (1987) using a hierarchical, parallel activation mechanism, as shown in Figure 1. The first stage of orthographic processing involves a specialized bank of letter detectors (the alphabetic array) that simultaneously receive input about visual feature information at a particular retinal location along the horizontal meridian (i.e., there are different letter detectors for different retinal locations). Each letter detector in the alphabetic array signals the presence of one out of 26 possible letters at a given retinal location. Open-bigram units receive activation from the alphabetic array such that a given letter order A-B, which is realized at any of the possible combinations of location in the alphabetic array, activates the open-bigram for that sequence (with possible constraints on the distance separating the two component letters, Grainger & van Heuven, 2003). Thus, in recoding information represented in the alphabetic array, a location-specific map of letter identities is replaced by a wordcentered, location-invariant code. All of the more recent proposals for letter position coding during printed word perception can accommodate the basic relativeposition priming effect that was replicated and extended in the present study (see Appendix). This is because all these coding schemes (Davis, 1999; Grainger & van Heuven, 2003; Whitney, 2001) code for the relative position of letters in a string somewhat independently of the absolute, length-dependent position of each letter in the string. This increased flexibility in the way letter position is computed has endowed such coding schemes with the ability to explain another equally important empirical phenome- non: transposed-letter priming. Open-bigram coding, for example, was developed explicitly to account for relative-position priming effects, yet this theoretical approach provides a principled account of transposed-letter priming (Grainger & Whitney, 2004). Transposed-letter primes are formed by changing the order of letters in a word without removing any of the letters. In standard masked priming with the lexical-decision task and relatively long target words, Forster et al. (1987) found that effects of transposed letter primes (e.g., salior-SAILOR) were practically the same as identity primes (e.g., sailor-SAILOR). This finding has been replicated and extended by Perea and Lupker (2003, 2004), and Schoonbaert and Grainger (2004), using more appropriate substitution primes as the baseline for evaluating transposed-letter priming (e.g., “sateor” as a control for “salior”). However, with shorter words (Humphreys et al., 1990), or when primes do not contain all of the target’s letters (Peressotti & Grainger, 1999), then transposition priming is greatly diminished. Using their 4-field masking procedure and perceptual identification responses to targets, Humphreys et al. (1990) found only a nonsignificant 3.1% increase in response accuracy for transposed letter primes (e.g., snad-SAND) compared to primes sharing two out of four letters with targets (e.g., smed-SAND). Similarly, Peressotti and Grainger (1999, Experiment 2a) observed a nonsignificant 5 ms advantage relative to all different primes when the two central letters were transposed in primes sharing four out of six letters with targets (e.g., bclnBALCON). This is in line with the results of Experiments 1 and 6 of the present study, showing no priming relative to an unrelated baseline for subset primes with transposed letters. Although at first sight contradictory, the robust effects of transposed-letter primes when primes contain all of the target’s letters, and the absence of such effects when primes are formed of a subset of the target’s letters, are both predicted by the models tested in the present study. Indeed, these two phenomena (transposed-letter priming and relative-position priming) provide the strongest evidence in favor of such coding schemes. More fine-grained manipulations are required to begin the difficult task of selecting among these models, and one such manipulation involves the level of contiguity of letters shared by prime and target stimuli. Contiguity and Orthographic Priming In relative-position primes, the letters shared by prime and target can vary in terms of the level of contiguity of these letters in the target stimulus. Thus, in Experiments 2, 3, 4, and 5 of the present study, primes could be completely contiguous letter sequences (e.g., 12345, 34567) or noncontiguous sequences of letters (e.g., 13457). The models presented above (except for an unconstrained version of open-bigram coding, see Appendix) all predict an influence of this factor on the size of relative-position priming effects. However, the way contiguity affects orthographic processing varies across the different models as a function of the specific mechanism used to code for letter position. To complete the present investigation, we will examine the possible role of letter contiguity as a factor determining the efficacy of orthographic primes. In an open-bigram coding scheme, contiguity can influence performance by having bigram activation vary as a function of the distance separating the component letters. This is the case in RELATIVE-POSITION PRIMING Whitney’s (2001) SERIOL model where more contiguous bigrams receive greater activation. In Grainger and van Heuven’s (2003) simulations, a simple binary weighting was used such that bigrams that involved more than two intervening letters (e.g., the bigram TE in the word TABLE) were ignored (see description of constrained open-bigram model in Appendix). At a more general level of theorizing, one can draw a distinction between models that deliberately code for noncontiguous letter combinations (e.g., Grainger & van Heuven, 2003; Whitney, 2001), and an approach in which noncontiguous letter combinations are coded as a result of positional coding errors in more traditional contiguous bigram coding schemes. If letter detector units in the alphabetic array are assumed to have overlapping receptive fields (see Figure 2), as in the overlap model of Perea, Go´mez, and Ratcliff (2003), then coding for adjacent letter combinations will lead to certain noncontiguous combinations being coded as well as letter transpositions. The amount of overlap in the receptive fields of letter detectors determines the limits of open-bigram coding via this mechanism (see Dehaene, Cohen, Sigman, & Vinckier, 2005, for a similar proposal). Thus, coding for adjacent bigrams in a fuzzy slot-based coding scheme generates a graded activation in contiguous and noncontiguous bigrams, and provides a natural means of constraining the degree of separation of letters forming an openbigram. Furthermore, the differential weighting of open-bigrams as a function of contiguity provides more precise positional information in such a coding scheme (see Appendix for more details). These two different coding schemes can be seen as trading off local versus more global position information in opposite ways. The contiguous coding scheme optimizes local combinations at the cost of an increase in the risk of making position coding errors, while the noncontiguous scheme optimizes accurate relativeposition coding (knowledge that a given letter is positioned left or Figure 2. Description of the overlap open-bigram model. Letter detectors in the alphabetic array have large overlapping receptive fields (RFs) such that for a given letter at a given retinal location one letter identity will be maximally activated, and other letter identities falling within the receptive field of the letter detector will also receive some activation (see Appendix for details). Bigrams are computed across adjacent locations in the alphabetic array on the basis of all letter identities (several at any given location) activated above a criterion value. Thus bigrams are formed from adjacent letters in the correct order, nonadjacent letters in the correct order (openbigrams), and adjacent letters in the incorrect order (transposed bigrams). 879 right of another letter) at the cost of not directly specifying local information (knowledge that a given letter is next to another letter). More generally, contiguous and noncontiguous letter combinations (whatever the mechanism that generates them) can be seen as providing two different types of orthographic code, possibly with different time-courses. Noncontiguous combinations would provide a fast, approximate orthographic code useful for providing an initial constraint on word identity, whereas contiguous letter combinations would provide a more fine-grained orthographic representation that would be useful for deriving a phonological code. Therefore, under this view the contiguous code should start to dominate processing as soon as phonology starts to exert an influence during visual word recognition. On the other hand, the influence of noncontiguous orthographic codes will diminish as phonological codes start to dominate processing, since they map poorly onto phonology (many noncontiguous letter combinations will not have a corresponding phonological representation). The observed difference in the time-courses of noncontiguous relative position primes, single letter substitution primes, and pseudohomophone primes in Experiment 6 is in line with this proposition. The results of match calculations for the different versions of open-bigram coding and the SERIOL and SOLAR models are given in the Appendix. Not surprisingly, a model in which contiguity does not affect orthographic processing (i.e., the unconstrained open-bigram model) does the worse in conditions where we suspect phonological prime-target overlap is partly driving priming effects (see Table A2 in Appendix). This is because it is precisely in these conditions that we found the clearest evidence for an influence of contiguity: Primes formed of completely contiguous sequences of letters from the target (e.g., 12345, 34567) were more effective than primes formed of noncontiguous letter sequences (e.g., 13457). All models, except for the unconstrained open-bigram model, predict that the level of contiguity of prime letters in the target stimulus will influence priming, hence the relatively strong correlations between the match values generated by these models and priming effects sizes. However, what is more important is that even in conditions where phonology was not found to significantly affect priming, the unconstrained openbigram model continues to generate the poorest predictions (see Table A1 in Appendix). This is because the model overestimates the amount of priming generated by the transposition of inner letters (the 15437 condition of Experiment 1 and the 1537 condition of Experiment 6). The SOLAR model also has difficulty with these particular priming conditions, again overestimating the size of priming effects compared to certain conditions where letter order is respected (e.g., 1469 prime for 9-letter targets). Therefore, the match calculations shown in the Appendix reveal that models of orthographic processing that include contiguity constraints (particularly the overlap open-bigram model and the SERIOL model) actually do a better job in capturing the overall pattern of results in testing conditions where contiguity was not expected have a strong influence. However, these models do fail to account for the robust priming effect found with extreme cases of noncontiguous primes (i.e., the 1469 primes in Experiment 5). Clearly more experimentation is required on effects of contiguity in relative-position priming in order to provide more discriminatory data with respect to current accounts of letter position coding. These experiments should provide more information concerning GRAINGER ET AL. 880 the precise time-course of effects of contiguous and noncontiguous orthographic primes. Conclusions The present study replicates and extends the relative-position priming effects first demonstrated by Humphreys et al. (1990) and Peressotti and Grainger (1999). Relative-position primes contain a subset of the target’s letters that are concatenated such that while letter order information is respected across prime and target stimuli, absolute, length-dependent information is violated. Experiments 1 and 6 showed that order information is important, since priming effects disappeared when the subset of letters forming the prime stimulus did not respect their order in the target. Experiments 2 and 3 compared relative-position priming using different subsets of a target’s letters: initial letters, final letters, and a combination of the first, last, and central letters (outer-central primes). Initial and final letter primes produced significant priming and these two conditions did not differ significantly, while the outer-central primes produced less priming. One possibility is that the reduced priming for the outer-central primes is due to the lower level of contiguity in this prime condition, and match calculations using Grainger and van Heuven’s (2003) constrained open-bigram model, the overlap open-bigram model, the SERIOL model (Whitney, 2001), and the SOLAR model (Davis, 1999) indeed predict such an effect of contiguity (see Appendix). However, post hoc analyses showed that differences in priming effect sizes could be due to differences in the level of prime-target phonological overlap across the different priming conditions. Experiments 4 and 5 confirmed this hypothesis by showing statistically equivalent priming effects for initial, final, and outer-central primes in conditions where prime-target phonological overlap did not influence priming (i.e., with short, 33 ms prime exposures). Experiment 6 further confirmed that phonological priming only emerges with longer prime exposures. On the basis of these results, it was concluded that when phonological influences on priming effects are minimized, then the level of contiguity of relative-position primes does not influence priming. Whether or not orthographic priming is indeed affected by the level of contiguity of letters shared by prime and target remains an important issue for future research. Ongoing research in our laboratory has further attempted to answer this question by using superset primes, that is primes that include all of the target’s letters in the correct order, plus some irrelevant letters (Van Assche & Grainger, 2006). Another very interesting extension of the present work would be to test for similar types of orthographic priming using parafoveal priming techniques (see Rayner, 1998, for a review). Since parafoveal priming is thought to reflect the earliest phases of printed word perception operating before the eyes are actually fixating the target word, we would expect to observe priming effects very much in line with those observed in the present study. Another methodology that has recently proved sensitive to early orthographic processing involves recording eventrelated brain potentials (ERPs). In recent work combining ERP recordings with masked priming methodology, Misra and Holcomb (2003) found priming effects arising as early as 200 ms posttarget onset (see also Holcomb & Grainger, 2006). It will therefore be interesting to see whether ERPs are sensitive to the type of priming manipulated in the present study, and at what time during target processing such influences appear. Converging evidence of this kind will be critical for defining the type of phenomena that computational models of orthographic processing must be able to accommodate. References Besner, D., Coltheart, M., & Davelaar, E. (1984). Basic processes in reading: Computation of abstract letter identities. Canadian Journal of Experimental Psychology, 38, 126 –134. Briihl, D., & Inhoff, A. W. (1995). Integrating information across fixations during reading: The use of orthographic bodies and of exterior letters. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 55– 67. Caramazza, A., & Hillis, A. (1990). Levels of representation, coordinate frames, and unilateral neglect. Cognitive Neuropsychology, 7, 391– 445. Carreiras, M., Ferrand, L., Grainger, J., & Perea, M. (2005). Sequential effects of phonological priming in visual word recognition. Psychological Science, 16, 585–589. Cohen, J., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope: An interactive graphic system for designing and controlling experiments in the psychology laboratory using Macintosh computers. Behavior Research Methods, Instruments and Computers, 25, 257–271. Coltheart, M., Curtis, B., Atkins, P., & Haller, M. (1993). Models of reading aloud: Dual-route and parallel-distributed-processing approaches. Psychological Review, 100, 589 – 608. Coltheart, M., Rastle, K., Perry, C., & Ziegler, J. (2001). DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review, 108, 204 –256. Davis, C. J. (1999). The Self-Organising Lexical Acquisition and Recognition (SOLAR) model of visual word recognition. Unpublished doctoral dissertation, University of New South Wales, Australia. Davis, C. J., & Bowers, J. S. (2004). What do letter migration errors reveal about letter position coding in visual word recognition? Journal of Experimental Psychology: Human Perception and Performance, 30, 923–941. Dehaene, S., Cohen, L., Sigman, M., & Vinckier, F. (2005). The neural code for written words: A proposal. Trends in Cognitive Sciences, 9, 335–341. Diependaele, K., Sandra, D., & Grainger, J. (2005). Masked cross-modal morphological priming: Unravelling morpho-orthographic and morphosemantic influences in early word recognition. Language and Cognitive Processes, 20, 75–114. Dunnett, C. W. (1955). A multiple comparison procedure for comparing several treatments with a control. Journal of the American Statistical Association, 50, 1096 –1121. Evett, L. J., & Humphreys, G. W. (1981). The use of abstract graphemic information in lexical access. Quarterly Journal of Experimental Psychology, 33A, 325–350. Ferrand, L., & Grainger, J. (1992). Phonology and orthography in visual word recognition: Evidence from masked nonword priming. Quarterly Journal of Experimental Psychology, 45A, 353–372. Ferrand, L., & Grainger, J. (1993). The time-course of phonological and orthographic code activation in the early phases of visual word recognition. Bulletin of the Psychonomic Society, 31, 119 –122. Ferrand, L., & Grainger, J. (1994). Effects of orthography are independent of phonology in masked form priming. Quarterly Journal of Experimental Psychology, 47A, 365–382. Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 680 – 698. Forster, K. I., Davis, C., Schoknecht, C., & Carter, R. (1987). Masked priming with graphemically related forms: Repetition or partial activation? Quarterly Journal of Experimental Psychology, 39A, 211–251. RELATIVE-POSITION PRIMING Forster, K. I., & Forster, J. (2003). DMDX: A Windows display program with millisecond accuracy. Behavioral Research Methods: Instruments & Computers, 35, 116 –124. Frost, R., Ahissar, M., Gotesman, R., & Tayeb, S. (2003). Are phonological effects fragile? The effect of luminance and exposure duration on form priming and phonological priming. Journal of Memory and Language, 48, 346 –378. Grainger, J., Diependaele, K. Spinelli, E. Ferrand, L., & Farioli, F. (2003). Masked repetition and phonological priming within and across modalities, Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 1256 –1269. Grainger, J., & Jacobs, A. M. (1993). Masked partial-word priming in visual word recognition: Effects of positional letter frequency. Journal of Experimental Psychology: Human Perception and Performance, 19, 951–964. Grainger, J., & Jacobs, A. M. (1999). Temporal integration of information in orthographic priming. Visual Cognition, 6, 461– 492. Grainger, J., & Van Heuven, W. (2003). Modeling letter position coding in printed word perception. In P. Bonin (Ed.), The mental lexicon (pp. 1–24). New York: Nova Science Publishers. Grainger, J., & Whitney, C. (2004). Does the huamn mnid raed wrods as a wlohe? Trends in Cognitive Sciences, 8, 58 –59. Grossberg, S. (1978). A theory of human memory: Self-organization and performance of sensory-motor codes, maps, and plans. In R. Rosen & F. Snell (Eds.), Progress in theoretical biology (pp. 233–374). New York: Academic Press. Holcomb, P. J., & Grainger, J. (2006). On the time-course of visual word recognition: An ERP investigation using masked repetition priming. Journal of Cognitive Neuroscience, in press. Humphreys, G. W., Evett, L. J., & Quinlan, P. T. (1990). Orthographic processing in visual word identification. Cognitive Psychology, 22, 517–560. Imbs, P. (1971). Etudes statistiques sur le vocabulaire franc¸ais: Dictionnaire des fre´quences. Vocabulaire litte´raire des XIXe et XXe sie`cles. Paris: Librairie Marcel Didier. Inhoff, A. W. (1989). Lexical access during eye fixations in reading: Are word access codes used to integrate lexical information across interword fixations? Journal of Memory and Language, 28, 444 – 461. Inhoff, A. W., Radach, R., Eiter, B. M., & Juhasz, B. (2003). Distinct subsystems for the parafoveal processing of spatial and linguistic information during eye fixations in reading. Quarterly Journal of Experimental Psychology, 56A, 803– 827. Jacobs, A. M., Rey, A., Ziegler, J. C., & Grainger, J. (1998). MROM-p: An interactive activation, multiple read-out model of orthographic and phonological processes in visual word recognition. In J. Grainger & A. M. Jacobs (Eds.), Localist connectionist approaches to human cognition (pp. 147–188). Mawah, NJ: Erlbaum. Jordan, T. R., Thomas, S. M., Patching, G. R., & Scott-Brown, K. C. (2003). Assessing the importance of letter pairs in initial, exterior, and interior positions in reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 883– 893. 881 McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Pt. 1. An account of basic findings. Psychological Review, 88, 375– 407. Misra, M., & Holcomb, P. J. (2003). The electrophysiology of word-level masked repetition priming. Psychophysiology, 40, 115–130. Mozer, M. (1987). Early parallel processing in reading: A connectionist approach. In M. Coltheart (Ed.), Attention and performance XII (pp. 83–104). Hillsdale, NJ: Erlbaum. New, B., Pallier, C., Ferrand, L., & Matos, R. (2001). Une base de donne´es lexicales du franc¸ais contemporain sur internet: LEXIQUE. L’Anne´e Psychologique, 101, 447– 462. Perea, M., Go´mez, P., & Ratcliff, R. (2003). A model of encoding letter positions: The overlap model. Proceedings of the XIIth Congress of the European Society for Cognitive Psychology, Granada, Spain. Perea, M., & Lupker, S. J. (2003). Transposed-letter confusability effects in masked form priming. In S. Kinoshita and S. J. Lupker (Eds.), Masked priming: State of the art (pp. 97–120). Hove, UK: Psychology Press. Perea, M., & Lupker, S. J. (2004). Can Caniso activate Casino? Transposed-letter similarity effects with nonadjacent letter positions. Journal of Memory and Language, 51, 231–246. Peressotti, F., & Grainger, J. (1999). The role of letter identity and letter position in orthographic priming. Perception & Psychophysics, 61, 691– 706. Plaut, D., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56 –115. Pollatsek, A., & Well, A. (1995). On the use of counterbalanced designs in cognitive research: A suggestion for a better and more powerful analysis. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 785–794. Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124, 372– 422. Rayner, K., McConkie, G. W., & Zola, D. (1980). Integrating information across eye movements. Cognitive Psychology, 12, 206 –226. Schoonbaert, S., & Grainger, J. (2004). Letter position coding in printed word perception: Effects of repeated and transposed letters. Language and Cognitive Processes, 19, 333–367. Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word recognition and naming. Psychological Review, 96, 523–568. Van Assche, E., & Grainger, J. (2006). A study of relative-position priming with superset primes. Journal of Experimental Psychology: Learning, Memory, and Cognition, in press. Whitney, C. (2001). How the brain codes the order of letters in a printed word: The SERIOL model and selective literature review. Psychonomic Bulletin & Review, 8, 221–243. Ziegler, J., Ferrand, L., Jacobs, A. M., Rey, A., & Grainger, J. (2000). Visual and phonological codes in letter and word recognition: Evidence from incremental priming. Quarterly Journal of Experimental Psychology, 53A, 671– 692. (Appendix follows) GRAINGER ET AL. 882 Appendix The appendix describes the method of calculating match values between incoming orthographic information (prime) and a whole-word orthographic representation (target) for an unconstrained open-bigram model (UOB), a constrained open-bigram model (COB, Grainger & van Heuven, 2003), and the overlap open-bigram model (OOB) discussed in the present work. The match values generated by these models are compared with the net priming effects obtained in the present study and the corresponding match values generated by the SOLAR model (see footnote 3) and the SERIOL modelA1. The results of these calculations and the correlations with priming effects are given in Table A1 for three Experiments (Experiments 1, 5, and 6) where phonological prime-target overlap are hypothesized not to have influenced differences in priming effects across conditions. (Experiment 4 is not included because of the different dependent variable used in that experiment.) Table A2 provides the match calculations and correlations with net priming effects in Experiments 2 and 3 where priming effects were shown to be influenced by prime-target phonological overlap. Unconstrained Open-Bigram Model (UOB) To calculate match values for this model we first determine the openbigrams that are activated by the prime and target string. Open-bigrams are activated from an input string by adjacent and non-adjacent letter pairs in the correct order. For example, the input string 12345 activates the openbigrams 12, 13, 14, 15, 23, 24, 25, 34, 35, and 45. Each open-bigram (OB) receives an activation value of 1.0. Equation (1) is then used to calculate the match value of the prime and target string. In this formula the sum of products of activations of matching OBs of primes and targets is divided by the sum of products of activations from all target OBs. match value ⫽ ⌺OB共 prime兲*OB共target兲 ⌺OB共target兲 2 Example: Target ⫽ 1234567 Open-bigrams: 12, 13, 14, 15, 16, 17, 23, 24, 25, 26, 27 34, 35, 36, 37, 45, 46, 47, 56, 57, 67 Prime ⫽ 15437 Open-bigrams: 15, 14, 13, 17, 54, 53, 57, 43, 47, 37 The open-bigrams 15, 14, 13, 17, 57, 47, and 37 are activated by the prime and target string, thus the match value is 0.33. match value ⫽ 7 1⫹1⫹1⫹1⫹1⫹1⫹1 ⫽ ⫽ 0.33 21 21 Constrained Open-Bigram Model (COB) This model is similar to the previous model except that it includes a constraint on the maximum distance (in number of letters) separating the constituent letters of open-bigrams in the input string. Following Grainger and van Heuven (2003) this is set at two letters (e.g., the string 12345 only activates the open-bigrams 12, 13, 14, 23, 24, 25, 34, 35, and 45). Each A1 The match values for the SERIOL model and the equations used to calculate these are given in the following unpublished manuscript: Whitney, C., Supporting the Serial in the SERIOL model (available upon request from the author). Table A1 Match Values Generated by the Unconstrained Open-Bigram Model (UOB), the Constrained Open-Bigram Model (COB), the Overlap Open-Bigram Model (OOB), the SERIOL Model (Whitney, 2001; Footnote A1), and the SOLAR Model (Davis, 1999; Footnote 3) for the Priming Conditions Tested in Experiments 1, 5 (Unmasked), and 6 (33 ms Prime Duration) Prime Priming effect UOB COB OOB 1a 1-345-7 13-4-57 13457 20* 22* 25* 0.48 0.48 0.48 0.47 0.47 0.47 0.47 0.47 0.47 0.57 0.57 0.57 0.64 0.64 0.64 1b 1-345-7 1-543-7 7-345-1 23* ⫺2 7 0.48 0.33 0.14 0.47 0.27 0.20 0.47 0.11 0.30 0.57 0.27 0.23 0.64 0.45 0.39 5 (7-letter) 1234 4567 1357 36* 30* 31* 0.29 0.29 0.29 0.40 0.40 0.20 0.48 0.48 0.23 0.48 0.45 0.38 0.58 0.55 0.40 5 (9-letter) 1234 6789 1469 26* 17* 23* 0.17 0.17 0.17 0.29 0.29 0.14 0.35 0.35 0.08 0.36 0.35 0.18 0.46 0.44 0.23 6 1357 1537 7351 20* 8 7 0.29 0.24 0.05 0.40 0.13 0.07 0.48 0.10 0.08 0.38 0.19 0.02 0.58 0.33 0.24 0.28 0.48 0.61⫹ 0.59⫹ 0.43 Experiment Correlation with net priming effect SERIOL SOLAR Note. ⫹ Correlation is significant at the 0.05 level (2-tailed); * significantly different from the ddddd condition, p ⬍ .05 by participant using Dunnett’s test; parameters of the SOLAR model: sigma ⫽ 3.0, w_IMR ⫽ 1.4, w_FMR ⫽ 1.2; ‘-’ characters were removed from input strings in the match calculations. RELATIVE-POSITION PRIMING 883 Table A2 Match Values Generated by the unconstrained Open-Bigram Model (UOB), the Constrained Open-Bigram Model (COB), the Overlap Open-Bigram Model (OOB), the SERIOL Model (Whitney, 2001; Footnote A1), and the SOLAR Model (Davis, 1999; Footnote 3) for the Priming Conditions Tested in Experiments 2 and 3 Prime Priming effect 2 (7-letter) 12345 34567 13457 45* 37* 29* 0.48 0.48 0.28 0.60 0.60 0.47 0.65 0.65 0.47 0.61 0.60 0.57 0.71 0.68 0.64 2 (9-letter) 12345 56789 14569 30* 28* 12 0.28 0.28 0.28 0.43 0.43 0.24 0.48 0.48 0.25 0.46 0.46 0.32 0.56 0.54 0.38 3 (7-letter) 1234 4567 1357 23* 12 8 0.29 0.29 0.29 0.40 0.40 0.20 0.48 0.48 0.23 0.48 0.45 0.38 0.58 0.55 0.40 3 (9-letter) 1234 6789 1469 23* 20* 8 0.17 0.17 0.17 0.29 0.29 0.14 0.35 0.35 0.08 0.36 0.35 0.18 0.46 0.44 0.23 0.68⫹ 0.90⫹⫹ 0.87⫹⫹ 0.82⫹ 0.85⫹⫹ Experiment Correlation with net priming effect UOB COB OOB SERIOL SOLAR Note. ⫹ Correlation is significant at the 0.05 level (2-tailed); ⫹⫹ Correlation is significant at the 0.01 level (2-tailed); * significantly different from the ddddd condition, p ⬍ .05 by participant using Dunnett’s test; parameters of the SOLAR model: sigma ⫽ 3.0, w_IMR ⫽ 1.4, w_FMR ⫽ 1.2. open-bigram receives an activation value of 1.0, and Equation (1) is used to calculate the match value of the prime and target string. Example: Target ⫽ 1234567 Open-bigrams: 12, 13, 14, 23, 24, 25, 34, 35, 36, 45, 46, 47, 56, 57, 67 Prime ⫽ 15437 Open-bigrams; 15, 14, 13, 54, 53, 57, 43, 47, 37 Only the open-bigrams 13, 14, 47, and 57 are activated by the prime and target string, thus the match value is 0.27. match value ⫽ 4 1⫹1⫹1⫹1 ⫽ ⫽ 0.27 15 15 Overlap Open-Bigram Model (OOB) Contrary to the two previous versions of open-bigram coding, the overlap open-bigram model attempts to code for contiguous letter sequences (bigrams) only. However, the noisy coding of letter position implies that non-contiguous letter sequences (open-bigrams) will also be coded. Noisy position coding is introduced as overlapping receptive fields for each letter detector in the alphabetic array (see Figure 2). The receptive fields place a normal distribution of activation values on top of each letter position (e.g., 0.607, 1.0, 0.607 for the three-letter receptive field adopted in the present simulations), such that for a given letter detector in the alphabetic array, the letter present at that location receives full activation (1.0) and the neighboring letters (immediately left and right of that location) receive partial activation (0.607). Thus in the example Table A3 the letter detector that is aligned with the letter B signals the presence of B at that location with 100% likelihood, but also signals the possible presence of A and L at that same location with a probability of 0.607. Contiguous bigrams are then coded by examining the activity of different letter identities at each location in the alphabetic array. Bigram activation is equal to the product of the activations of its component letters. When identical bigrams are constructed, only the one with the highest activation remains. Note that both contiguous and non-contiguous bigrams are formed as well as transposed bigrams, as illustrated in the following example. (Appendix continues) GRAINGER ET AL. 884 Table A3 Activation of Letters in the Alphabetic Letter Array for Each Receptive Field First letter Receptive field ... . .T .TA TAB ABL BLE LE. E. . ... Second letter Activation . . . T A B L E . 0.607 0.607 0.607 0.607 0.607 0.607 0.607 0.607 0.607 Third letter Activation . . T A B L E . . 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Activation . T A B L E . . . 0.607 0.607 0.607 0.607 0.607 0.607 0.607 0.607 0.607 Note. Moving from top to bottom in the above example (i.e., along the alphabetic array from left to right), ordered pairs of adjacent letters are formed, leading to the activation of the following bigrams (activation values between brackets): TA (1.0), AB (1.0), BL (1.0), LE (1.0), TB (0.607), AL (0.607), BE (0.607), TL (0.135), AE (0.135), AT (0.135), BA (0.135), LB (0.135), EL (0.135). These bigrams are then used to calculate the match value between prime and target stimuli using the match value Equation (1) described above. Note that bigrams formed of repeated letters (e.g., TT, AA, etc.) have been ignored here since they do not affect the match calculation. Received January 17, 2005 Revision received January 18, 2006 Accepted January 22, 2006 䡲