Journal of Experimental Psychology: Learning, Memory, and Cognition 2014, Vol. 40, No. 3, 623-644

© 2013 American Psychological Association 0278-7393/14/$12.00 DOI: Í0.1037/a0035154

Activation of Lexical and Semantic Representations Without Intention Along GPC-Sublexical and Orthographic-Lexical Reading Pathways in a Stroop Paradigm Kathryn F. Anton, Layla Gould, and Ron Borowsky University of Saskatchewan Dual route models of reading suggest there are 2 pathways for reading words: an orthographic-lexical pathway, used to read familiar regular words and exception words, and a grapheme-to-phonemeconversion-(GPC)-sublexical pathway, used to read unfamiliar regular words, pseudohomophones (PHs), and nonwords. It is unclear, however, whether PHs activate lexical and semantic representations without intention in the GPC-sublexical pathway to the same extent as words along the orthographic-lexical pathway. The present study explored this by introducing a novel condition, color pseudohomophone associates (CPHAs; e.g., "skigh"), in 3 experiments using the Stroop paradigm. Experiment 1 examined 4 types of stimuli: color words (CWs), color word associates (CWAs), color PHs (CPHs), and color PH associates (CPHAs), in a mixed list context. Significant Stroop effects were found for all 4 types of stimuli. To ensure the robustness of this effect. Experiment 2 was conducted using pure list contexts whereby participants received only word stimuli (e.g., CWs, CWAs) or only PH stimuli (e.g., CPHs, CPHAs). The results replicated those of Experiment 1, suggesting that CPHAs activate lexical and semantic representations without intention in the GPC-sublexical pathway. Experiment 3 added 2 novel conditions: color exception word associates (which can only be pronounced correctly using the orthographic-lexical pathway) to compare the effects obtained with color exception PH associates (which rely on the GPC-sublexical pathway for correct pronunciation). Stroop effects of similar magnitude were found for both types of stimuli, suggesting lexical and semantic representations are accessed without intention in either reading pathway to a similar degree. Implications for models of reading are discussed. Keywords: Stroop, sublexical, lexical, reading, semantic activation

In the popular song "What A Wonderful World," Louis Armstrong sings "I see skies of blue, clouds of white" (Thiele & Weiss, 1967, Track 1), employing the typical association of the word "sky" with the color blue and the word "clouds" with the color white. But when individuals are presented with a word such as "sky" at fixation and are asked to ignore the word and name the color of the font that the word is printed in, do they have difficulty naming the color of the font when it is incongruent with some aspect of the word or the word's meaning, thus suggesting that the word is being read without intention? Research in the reading literature suggests that they do (Augustinova & Ferrand, 2012a, 2012b; Klein, 1964). What remains unclear, however, is whether this finding extends to letter strings that must be read via the

sublexical pathway of reading using grapheme-to-phoneme conversion (GPC, also referred to as phonetic decoding), and whether such effects of reading without intention are similar in magnitude to those for words along the orthographic-lexical pathway.

The Stroop Effect and Unintentional Lexical-Semantic Processing The seminal work of Stroop (1935/1992), and many studies since then (for a review see MacLeod, 1991), showing that individuals are significantly slower and less accurate to name the color of a letter string when it is presented in an incongruent color (e.g., the letter string "blue" printed in red font) has often been taken as evidence that individuals cannot inhibit reading words that are presented at fixation, even though doing so may be detrimental to their performance of the task at hand. This research has also been extended to a lexical and semantic level. Klein (1964) used the Stroop paradigm to determine whether individuals access the meaning of words without intention through to lexical and semantic levels of processing. Klein presented participants with nonsense letterstrings (e.g., "gsxrq"), rare English words (e.g., "abjure"), common English words not associated with a particular color (e.g., "friend"), common English words associated with a particular color that implicate those colors in their meaning (e.g., "sky" is commonly associated with the color blue), color words (CWs) not used as font colors in the task, and CWs used as font colors in the task. The results indicated that the more a word referred to a color

This article was published Online First December 2, 2013. Kathryn F. Anton, Layla Gould, and Ron Borowsky, Department of Psychology, University of Saskatchewan, Saskatoon, Saskatchewan, Canada. Ron Borowsky is the senior author of this article. This research was supported by the Natural Sciences and Engineering Research Council of Canada by Undergraduate Student Research Awards to Kathryn F. Anton, Canada Graduate Scholarships to Layla Gould, and Discovery Grant 183968-2013 to Ron Borowsky. Correspondence concerning this article should be addressed to Ron Borowsky, Department of Psychology, University of Saskatchewan, Saskatoon, SK, S7N 5A5, Canada. E-mail: ron.borowsky @usask.ca 623

ANTON, GOULD, AND BOROWSKY

624

in its meaning, the greater the amount of interference when naming a font color that was incongruent with the word. Words with color connotations in their meaning, such as the CWs not used as font colors in the task and the CWs used as font colors in the task, exhibited greater interference relative to words unrelated to colors, such as the nonsense syllables or rare English words. Importantly, Klein found that words associated with a particular color, color word associates (CWAs), were influenced by pairings with incongruent font colors (e,g,, the letter string "sky" printed in red font) just like regular CWs, albeit with a reduced effect. This suggests that CWAs somehow activate their semantic representations (i,e,, meaning), as the conflict between the color accessed through the word's meaning (e,g,, "sky" would activate the color blue) and the color that the letter string was presented in (e,g,, red) led to an increase in color naming reaction time (RT),

Dual Route Models of Reading and Unintentional Sublexical-Semantic Processing The aforementioned experiments are all similar in that they examine the effects of reading automaticity for familiar words, and thus likely reflect the orthographic-lexical pathway of reading. Some of the more predominant and successful models of reading aloud today are the dual route models of reading (e,g., Reynolds, Besner, & Coltheart, 2011 ; for an alternative dual route model, see Perry, Ziegler, & Zorzi, 2007), According to dual route models

(e,g„ Borowsky et al,, 2013; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001; Cummine et al, 2013; Gould, Cummine, & Borowsky, 2012) there are two ways in which words can be read (see Figure 1, upper two routes). The orthographic-lexical pathway relies on a memory representation of a word's visual features (i,e,, • the sequence of graphemes that comprise a word) and is used to read familiar regular words (e,g,, "tree") and exception words (e,g,, "pint"). In contrast, the GPC-sublexical pathway utilizes GPC and is used to read unfamiliar regular words (e,g,, "bucolic"), pseudohotnophones (PHs; words that sound like actual words but are spelled differently, e,g,, "pensill") and nonwords (NWs; e,g,, "grem"). To determine whether letter strings that rely on the GPCsublexical pathway of reading also demonstrate Stroop effects, Dennis and Newstead (1981) conducted a series of experiments in which they presented participants with a combination of CWs, color PHs (CPHs; e,g,, "bloo"), and neutral PHs that did not have color associations (e,g., "greef ). It was found that both the CWs and CPHs had longer color naming RTs relative to the neutral stimuli, with no difference being found between the CWs and CPHs, The finding that the CPHs exhibited a Stroop effect just as the CWs did suggests that phonological codes are also activated without intention through the GPC-sublexical route of reading and that individuals cannot stop themselves from reading the letter string. As Dennis and Newstead suggested, this is likely due to the

Phonological Output System

Visual Feature Encoding

Figure 1. The two upper routes represent a dual route model of reading (Borowsky et al., 2013; Cummine et al., 2013; Gould et al., 2012; see also Coltheart et al., 2001). There are two pathways in which words can be read: the orthographic-lexical pathway, in which familiar regular words (e.g., "tree") and exception words (e.g., "pint") are read, and a sublexical pathway, in which grapheme-to-phoneme conversion (GPC) or "sounding out" is used to read unfamiliar regular words (e.g., "bucolic"), pseudohomophones (e.g., "skigh"), and nonwords (e.g., "grem"). The lower third route represents how the color processing system could map onto such a model (see also Coltheart, Woolams, Kinoshita, & Perry, 1999). Following basic visual feature encoding, color processing connects to the semantic system and phonological lexical system so that the font color can be named. Given the significant Stroop effects in the present experiments, and particularly in Experiment 3, unintentional reading of the Stroop target would occur via the orthographic-lexical to phonological-lexical/semantic route with exception word color associate targets (e.g., "ocean"—blue), and via the GPC-sublexical to phonological-lexical/semantic route with their corresponding pseudohomophones (e.g., "oshin"—blue), as well as for pseudohomophones of other color associate words (e.g., "kerit"—orange). The remaining stimulus types may rely to varying degrees on both routes, or may not force activation through to the level of semantics (e.g., stimuli that are themselves color names, such as "blue" or "bloo"). The present results support the idea that both reading routes can be activated without intention through to a semantic level, yielding similar sized Stroop effects.

LEXICAL AND SEMANTIC ACTIVATION WITHOUT INTENTION fact that PHs activate an entry in the phonological lexical system corresponding to the actual words that they sound like (e.g., when "bloo" is sounded out, it sounds like "blue"). These results indicate that individuals read letter strings that are presented at fixation without intention, even though doing so harms their performance on identifying the font color of the letter string. Since Dennis and Newstead (1981) first conducted their study, numerous other studies have examined the effects of PHs. Besner and Stolz (1998) conducted two experiments in which they examined whether the processing of phonological codes occurs automatically. Participants were shown CWs, CPHs, and neutral controls, and asked to identify the color of the letter string. Both experiments found Stroop effects for the CPHs, indicating that GPC in the sublexical route of reading is difficult to control, although Besner and Stolz suggested that this finding does not imply that phonological computation itself is automatic as other contexts, such as the single letter cuing and coloring paradigm, have found decreased Stroop effects for CWs and GWAs (Besner, Stolz, & Boutilier, 1997; Küper & Heil, 2012; Manwell, Roberts, & Besner, 2004; we note, however, the limitations in ecological validity that are intrinsic to studies that focus attention on a single letter in a word). Furthermore, Besner and Stolz suggested it might be possible for individuals to control GPC to a certain extent. For example, in his review of the Stroop effect, MacLeod (1991) found that older adults produce significandy larger Stroop effects than younger adults. This is argued to be due to a decline in inhibitory processes, as younger adults may be able to suppress the word information in the Stroop task to a greater extent than older adults. West and Alain (2000) found evidence of this by showing that older adults had decreased activity in brain regions associated with the suppression of information relative to younger adults during the Stroop task. Despite these caveats, Besner and Stolz demonstrated that, at the very least, individuals have difficulty controlling GPC in the sublexical route of reading. Other research examining PHs has focused on whether the contexts in which PHs are presented differentially affects the results (e.g., Borowsky, Owen, & Masson, 2002; Marmurek & Kwantes, 1996; Reynolds & Besner, 2005; Reynolds et al., 2011). As previously mentioned, PHs rely on the GPC-sublexical route of reading, but it is possible that in mixed list contexts PHs are read more lexically (albeit phonological-lexical) than when presented in pure lists because the orthographic-lexical pathway of reading would also be activated due to the presence of actual words. Marmurek and Kwantes (1996) demonstrated that in a mixed list context containing words and PHs, words had shorter naming latencies and fewer errors relative to PHs but that PHs and words had similar frequency effects, indicating that both accessed their lexical representations. This contrasted with their findings from another experiment using pure and mixed lists, which suggested that PHs only access lexical representations in pure lists. Studies by Borowsky et al. (2002), Reynolds and Besner (2005), and Reynolds et al. (2011) have also examined the effects of PHs in different contexts by examining PHs and NWs. These studies found that when PHs and NWs are presented in a mixed list context, PHs have shorter naming latencies than NWs, whereas when PHs are presented before NWs in a pure list context, the PHs have longer naming RTs. Thus, the context in which PHs are presented must be taken into account, as the results of experiments have found differential effects in pure and mixed list contexts.

625

Last, some measure is needed in order to compare the effects found with PHs, which rely exclusively on the GPC-sublexical pathway of reading, to letter strings that rely exclusively on the orthographic-lexical pathway of reading. This can be accomplished through the use of exception words, words with inconsistent orthographic patterns that do not follow normal spelling-tosound correspondences (Cummine et al., 2013; Gould et al, 2012; Monsell, Patterson, Graham, Hughes, & Milroy, 1992). For example, "blood" is considered an exception word because using GPC would most likely result in a pronunciation of "blued" due to the fact that most of its orthographic neighbors (i.e., words that differ from the target by only one letter, e.g., "bloom, brood") are pronounced that way. In contrast, regular words can be pronounced correctly using the orthographic-lexical pathway or the GPC-sublexical pathway of reading because they follow consistent spelling-to-sound correspondences. Any comparisons involving the GPC-sublexical and orthographic-lexical pathways of reading would benefit from using both PHs and exception words, which force readers to utilize the GPC-sublexical and orthographiclexical pathways of reading, respectively. Thus, a study that directly examines exception words and matched PHs may provide additional evidence whether GPC in the sublexical pathway occurs without intention through to the level of lexical and semantic representations.

Overview of the Present Research Overall, the majority of previous research suggests that when CWs and CPHs are presented at fixation, they are read without intention (Dennis & Newstead, 1981; Stroop, 1935/1992). Similarly, when CWAs are presented at fixation they exhibit Stroop effects (Klein, 1964) and thus must activate lexical and semantic representations without intention. To our present knowledge, however, no research has attempted to determine whether color pseudohomophone associates (CPHAs; e.g., "skigh") also activate their lexical and semantic representations without intention. If CPHAs were found to exhibit Stroop effects similar to CPHs, CWAs, and CWs, it would provide further evidence that GPC in the sublexical route of reading is activated without intention and that this activation occurs to the level of lexical and semantic representations. To examine this, we conducted three experiments using the Stroop paradigm. The first experiment tested whether GPC operates without intention through to the level of lexical and semantic representations in a mixed list context consisting of words and PHs. Given the possibility that the results of Experiment 1 could have been influenced by the use of the mixed list context, which may have encouraged lexical activation as a result of the presence of actual word stimuli, a second experiment was conducted to determine if unintentional activation of lexical and semantic representations via the GPC-sublexical route of reading is differentially influenced by the use of pure list contexts. A third experiment was then conducted to control for the possibility that the real words used in the first two experiments may have been read using either the orthographic-lexical or GPC-sublexical pathways of reading. As a result, exception words with color associations and PHs derived from these stimuli were introduced in a third experiment that also included the CWs and CPHs of Experiments 1 and 2, to directly compare unintentional activation of lexical and semantic representations in the two reading pathways.

ANTON, GOULD, AND BOROWSKY

626 Experiment 1

Experiment 1 was designed to extend the results of previous experiments examining the automaticity of word reading (e.g., Dennis & Newstead, 1981; Klein, 1964; Stroop, 1935/1992) and to determine whether the process of GPC operates without intention through to the level of lexical and semantic representations. Four different types of stimuli were presented using the Stroop task: CWs (e.g., "blue"), CWAs (e.g., "sky"), CPHs (e.g., "bloo"), and CPHAs (e.g., "skigh"). There were three main hypotheses for Experiment 1 : (a) To the degree that the CWs and CWAs are read unintentionally, standard Stroop congruency effects should be found for these stimuli; (b) to the degree to which unintentional reading effects occur sublexically through GPC, the CPHs should exhibit significant Stroop congruency effects; and (c) to the degree to which such unintentional reading effects occur sublexically through GPC due to semantic activation, the CPHAs should also show significant Stroop congruency effects.

Method Participants. Participants in Experiment 1 were 24 undergraduate students from the University of Saskatchewan (Saskatoon, Saskatchewan, Canada), who received partial course credit for their participation. All participants were sufficiently fluent in EngUsh at a university level. Informed consent was obtained from all participants, and ethical approval for the study was obtained from the Research Ethics Board at the University of Saskatchewan prior to the commencement of the experiment. Stimuli and design. Participants were presented with four types of stimuli: CWs, CWAs, CPHs, and CPHAs, as shown in the Appendix. The stimuli were presented individually in bolded, lowercase, 32-point Arial font on a black background. The longest stimulus, "perpull," subtended a visual angle 2.864° in height X 10.570° in width. Each stimulus appeared in the following colors: blue (E-Prime's blue), green (E-Prime's green), pink (RGB: 255, 128, 192), red (E-Prime's red), yellow (E-Prime's yellow), orange (RGB: 210, 105, 0), purple (RGB: 159, 0, 159), white (E-Prime's white), or gray (RGB: 141, 141, 141). A stimulus was congruent if the word or PH that the letter string spelled out matched the font color (e.g., the letter string "blue" seen in blue font), whereas a stimulus was incongruent if the word or PH that the letter string spelled out did not match the font color (e.g., the letter string "blue" seen in red font). Each stimulus was randomly presented eight times in a congruent condition, and once in every incongruent condition (e.g., the word "blue" would be seen in blue font eight times, and once in every other font color: green, pink, red, yellow, orange, purple, white, and gray). Therefore, for each type of stimulus there were 72 congruent trials and 72 incongruent trials, for a total of 144 trials per stimulus type. The entire experiment consisted of 576 trials, presented in a single, randomized mixed list context. A 2 X 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate] X Stimulus Type [word, PH]) within-subject design was used in the experiment. Apparatus and procedure. Stimuli were presented on a 14.5in. (35.56-cm) IBM E94 color monitor. E-Pdme software (Version 2.0; Psychology Software Tools) was used to control the presentation of the stimuli, as well as the data collection. Vocal responses made by the participants were collected via a LabTec AM-22

microphone interfaced with the voicekey in the E-Prime serial response box, and RTs were measured to the nearest millisecond. Each participant was tested individually in a quiet room. Participants were presented with on-screen instructions describing the nature of the task. Participants read the following instructions: You will see a fixation cross {+) followed by a string of letters. When you are ready to proceed after each fixation, press the "Go" button to continue. Please name, as quickly as possible, the font color that the letter string is presented in. Do not worry about making occasional errors, in fact this type of research relies on you making at least some errors! Press the "Go" button when you are ready to begin. During this time, participants were also verbally instructed to ignore what the letter string said and to simply name the font color as quickly and accurately as possible. Participants then completed 18 practice trials, which contained nine congruent CW trials to familiarize the participants with the colors used in the experiment, and nine incongruent trials consisting of a mixture of CWs, CWAs, CPHs, and CPHAs so as to familiarize the participants with the four stimulus types used in the experiment. When a participant was ready to begin either the practice trials or the experimental trials, he or she pressed a button labeled "Go" on a manual-press response box, which replaced the on-screen instructions with a fixation cross presented in the center of the screen. The participant then pressed the "Go" button again, which caused the fixation cross to disappear and be immediately replaced by a stimulus in the center of the screen. The stimulus remained on the monitor until the participant verbalized his or her response, and the screen then turned black while the experimenter coded the response on the computer's keyboard as either the correct color, an incorrect color, or as a spoiled trial (e.g., reading the stimulus [responding "lemon" instead of the font color], beginning to vocalize one word but switching to another word half-way through [e.g., "bl- yellow"], or failure of the microphone to detect the verbalized response). After the participant's response was coded, the fixation cross immediately reappeared in the center of the screen to begin the next trial. During the practice trials, the experimenter provided feedback regarding incorrect color responses. No feedback was provided during the experimental trials. Following the conclusion of the experiment, participants were debriefed regarding the nature of the experiment. The entire experiment took approximately 30 min to complete.

Results The RT that it took participants to correctly name the color of the font for each stimulus was measured in milliseconds (ms) for each trial. Incorrect and spoiled trials were removed before conducting the analyses on median RTs (i.e., the median RT of each condition for each participant was the dependent variable for the RT analyses, given that medians are safeguarded from the influence of outliers, unlike means). In addition, the mean error rate of each condition (e.g., CW, CWA, CPH, CPHA) was computed for each participant by subtracting the number of correct responses for each stimulus type from 72 (the total number of congruent trials), dividing that result by 72, and then multiplying by 100. Median RT omnibus analysis of variance (ANOVA). Correct median RT data were analyzed using a 2 X 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate] X Stimulus Type [word, PH]) repeated-measures general linear model (GLM) ANOVA. The average median RT was 637.188 ms for congruent CPHs, 788.958 ms for incongruent

LEXICAL AND SEMANTIC ACTIVATION WITHOUT INTENTION

CPHs, 709.667 ms for congruent CPHAs, and 741.854 ms for incongruent CPHAs. The average median RT was 634.063 ms for congruent CWs, 809.938 ms for incongruent CWs, 714.625 ms for congruent CWAs, and 746.896 ms for incongruent CWAs. The median RT to respond to congruent stimuli (M = 673.885 ms) was faster than the median RT taken to respond to incongruent stimuli (M = 771.911 ms), in that there was a significant main effect of Congruency, F(l, 23) = 198.706, MSE = 2321.198, p < .001. The main effects of both Color Type, F(l, 23) = 3.689, MSE = 1496.503, p = .067, and Stimulus Type, F(l, 23) = 3.686, MSE = 631.474,/? = .067, approached significance. An interaction between Congruency and Color Type was found, F{\, 23) = 127.520, MSE = 1629.570,/) < .001, indicating a larger difference between the congruent and incongruent actual Color Type stimuli (e.g., CWs, CPHs) than between the congruent and incongruent associate Color Type stimuli (e.g., CWAs, CPHAs). The Congruency X Stimulus Type interaction approached significance, F(l, 23) = 3.055, MSE = 574.532, p = .094, whereas there was no evidence of a Color Type X Stimulus Type interaction, F(l, 23) = 0.384, MSE = 482.344, p = .542, or a Congruency X Color Type X Stimulus Type interaction, F(l, 23) = 1.978, MSE = 874.939, p = .173. The repeated-measures 95% confidence interval (CI; Loftus & Masson, 1994) was ±13.636 ms (see Figure 2). Median RT simple effects ANOVAs. The significant Congruency X Color Type interaction found in the median RT omnibus ANOVA was further investigated by evaluating the simple effects separately for the two Stimulus Types (e.g., PHs and words). This analysis was conducted in order to be consistent with the analyses that follow in Experiment 2, in which Stimulus Type was a between-subjects factor. In Experiment 1, this was com-

Congruent

Incongruent Congruency

" PH Associate

-PH Actual

-" — *" Word Associate — — ~ Word Actual

Figure 2. Median reaction time (RT) representing the time taken to correctly name the font color of pseudohomophones (PHs; color PH associates = CPHAs, e.g., "skigh"; actual color PHs = CPHs, e.g., "bloo") and words (color word associates = CWAs, e.g., "sky"; actual color words = CWs, e.g., "blue") as a function of Congruency, Color Type, and Stimulus Type in Experiment 1 (mixed list context). The 95% confidence intervals (CIs) demonstrate that, in each condition, participants were quicker to correctly name the font color on congruent trials than on incongruent trials, with a larger Stroop effect being found for the actual Color Type stimuli relative to the associate Color Type stimuli, regardless of the context in which the stimuli were viewed. The 95% CIs are represented by the error bars attached to each line using Loftus and Masson's (1994) method.

627

pleted with a 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate]) repeated-measures GLM ANOVA for each of the two Stimulus Types. For the PH simple effects ANOVA, there was a significant main effect of Congruency, F(l, 23) = 157.359, MSE = 1290.320, p < .001, indicating that font colors of congruent CPHs and CPHAs were named faster (M = 673.427 ms) than font colors of incongruent CPHs and CPHAs (M = 765.406 ms). A main effect of Color Type was also found, with the font colors of actual Color Type stimuli (e.g., CPHs) being named faster (M = 713.073 ms) than those of associate Color Type stimuli (e.g., CPHAs; M = 725.760 ms), F(l, 23) = 5.505, MSE = 701.784,/7 = .028. Last, there was a Congruency X Color Type interaction, F(l, 23) = 223.028, MSE = 384.710, p < .001, which indicated a greater effect of Congruency on the actual Color Type stimuli (e.g., CPHs) than on the associate Color Type stimuli (e.g., CPHAs), as shown in Figure 2. The repeatedmeasures 95% CI (Loftus & Masson, 1994) was ±11.462 ms. For the word simple effects ANOVA, a significant main effect of Congruency was found, F(l, 23) = 161.920, MSE = 1605.410, p < .001, whereby the font colors of congruent CWs and CWAs were read faster (M = 674.344 ms) than the font colors of incongruent CWs and CWAs (M = 778.417 ms). No evidence was found for a main effect of Color Type, F(l, 23) = 1.442, MSE = 1277.062, p = .242. Last, there was a significant interaction between Congruency and Color Type, F(l, 23) = 58.370, MSE = 2119.799, p < .001, indicating a greater effect of congruency on the CWs than on the CWAs (see Figure 2). The repeated-measures 95% CI (Loftus & Masson, 1994) was ± 16.628 ms. Error rate omnibus ANOVA. The accuracy of participants was analyzed by calculating the percentage of errors in order to determine whether there were any speed-accuracy trade-offs. The mean error rate (%) was analyzed using a 2 X 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate] X Stimulus Type [word, PH]) repeated-measures GLM ANOVA. The mean error rate was 4.109% for congruent CPHs, 17.130% for incongruent CPHs, 7.986% for congruent CPHAs, and 7.465% for incongruent CPHAs. The mean error rate was 3.588% for congruent CWs, 21.528% for incongruent CWs, 6.597% for congruent CWAs, and 8.218% for incongruent CWAs. Results of the omnibus repeated-measures ANOVA revealed a significant main effect of Congruency, F(l, 23) = 63.584, MSE = 48.496, p < .001, whereby participants made less errors in congruent trials (M = 5.570%) than in incongruent trials (M = 13.585%). A significant main effect was also found for Color Type, F(l, 23) = 35.372, MSE = 21.951, /? < .001, indicating that participants made fewer errors when identifying the font colors of the associate Color Type stimuli (M = 7.567%) than when identifying the font colors of the actual Color Type stimuli {M = 11.589%). There was no evidence of a main effect of Stimulus Type, F(l, 23) = 2.465, MSE = 12.783, p = .130. There was a Congruency X Color Type interaction, F(l, 23) = 76.949, MSE = 34.764, p < .001, as well as interactions between Congruency and Stimulus Type, F(l, 23) = 9.040, MSE = 16.542, p = .006, and Color Type and Stimulus Type, F(l, 23) = 5.851, MSE = 10.447, p = .024. In contrast, there was no evidence of a Congruency X Color Type X Stimulus Type interaction, F(l, 23) = 2.374, MSE = 9.750, p = .137. The repeated-measures 95% CI (Loftus & Masson, 1994) was ±1.895% (see Figure 3).

ANTON, GOULD, AND BOROWSKY

628

22.474, p < .001. Last, an interaction was found between Congruency and Color Type, F(l, 23) = 55.285, MSE = 28.904, p < .001, indicating a larger difference in the mean error rate between the congruent and incongruent conditions for the actual Color Type stimuli than between the congruent and incongruent conditions for the associate Color Type stimuli, as can be seen in Figure 3. The repeated-measures 95% CI (Loftus & Masson, 1994) was ±2.253%. There were no significant speed-accuracy trade-offs.

32 30 28 . 26 . 24 22 20 18 16 14 12 10 8 6 4 2 0 Congruent

Incongruent Congruency

-PH Associate

— P H Actual

*" "- — Word Associate — — — Word Actual

Figure 3. Mean error rate (%) representing the percentage of pseudohomophones (PHs; color PH associates = CPHAs, e.g., "skigh"; actual color PHs = CPHs, e.g., "bloo") and words (color word associates = CWAs, e.g., "sky"; actual color words = CWs, e.g., "blue") whose font color was incorrectly identified as a function of Congruency, Color Type, and Stimulus Type in Experiment 1 (mixed list context). The 95% confidence intervals (represented by the error bars) demonstrate that, on average, participants made fewer errors in the congruent condition, with a larger difference being found between the congruent and incongruent trials for the actual Color Type stimuli relative to the associate Color Type stimuli. No significant speed-accuracy trade-offs were found.

Error rate simple effects ANOVAs. The significant effects found in the error rate omnibus ANOVA were further evaluated by separately examining the simple effects for the two Stimulus Types (e.g., PHs and words). This was completed through the use of a 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate]) repeated-measures GLM ANOVA for each of the two Stimulus Types. Eor the PH simple effects ANOVA, there was a main effect of Congruency, F(l, 23) = 38.085, MSE = 24.616, p < .001, indicating that the font colors of congruent CPHs and CPHAs were named with fewer errors (M = 6.047%) than the font colors of incongruent CPHs and CPHAs (M = 12.297%). A significant main effect of Color Type was also found, F(l, 23) = 20.246, MSE = 9.925, p < .001, with the associate Color Type letter strings exhibiting fewer errors (M = 7.726%) than the actual Color Type letter strings (M = 10.619%). The interaction between Congruency and Color Type reached significance, F(l, 23) = 70.483, MSE = 15.610, p < .001, indicating a larger effect of Congruency on the mean enor rate of the CPHs than on the mean error rate of the CPHAs (see Figure 3). The repeated-measures 95% CI (Loftus & Masson, 1994) was ±1.665%. The results of the mean error rate analyses for the PH data indicated there were no significant speed-accuracy trade-offs. In regards to the word simple effects ANOVA, a significant difference in the mean error rate was found between the congruent CWs and CWAs (M = 5.093%) and the incongruent CWs and CWAs (M = 14.873%), indicating a main effect of Congruency, F(l, 23) = 56.791, MSE = 40.422, p < .001. There was also a significant difference in the mean error rate of responses based on Color Type as the font colors of CWAs were named with fewer errors (M = 7.407%) than the font colors of CWs (M = 12.558%), indicating a main effect of Color Type, F(l, 23) = 28.329, MSE =

Discussion The results of Experiment 1 provide evidence of significant Stroop effects for all four types of stimuli. The median RT data revealed significant differences between the congruent and incongruent conditions, with significantly larger differences being observed between the actual Color Type stimuli relative to the associate Color Type stimuli. As the confidence intervals demonstrate, significant differences were found between the congruent and incongruent conditions for the CWs, CWAs, CPHs, and CPHAs. The error rate data similarly revealed significant differences between the congruent and incongruent conditions, and the two Color Type conditions. As the confidence intervals of the error rate data show, significant differences between the congruent and incongruent conditions were found only for the CWs and CPHs. No evidence was found of any significant speed-accuracy tradeoffs in the CWs, CWAs, CPHs, or CPHAs. Taken together, these data support the three hypotheses that were proposed for Experiment 1. Significant Stroop effects were found for the CWs and CWAs, as evidenced by the faster RTs and lower percentage of errors in the congruent trials compared to the incongruent trials. Similarly, a significant Stroop effect was found for the CPHs. Most important, however, was the finding of a significant Stroop effect for the novel condition involving the CPHAs. The finding of a significant Stroop effect for the CPHAs supports the hypothesis that the GPC-sublexical route of reading operates without intention through to the level of lexical and semantic representations and suggests that when individuals read a CPHA such as "skigh," they automatically associate the letter string with aspects related to its meaning (e.g., the color blue). Furthermore, there was a significant interaction between Congruency and Color Type that indicated a larger Congruency effect for actual Color Type stimuli (e.g., CWs, CPHs) than for associate Color Type stimuli (e.g., CWAs, CPHAs), but there was no significant three-way interaction with Stimulus Type (e.g., words, PHs), suggesting that the Congruency by Color Type interaction is similar for words and PHs, both of which were significant in the simple effects ANOVAs. This similarity in Stroop effects across the word and PH stimuli suggests that the degree of reading without intention may be similar across the orthographic-lexical and GPC-sublexical pathways, but we return to this issue in Experiment 3 with a stimulus manipulation that will directly test this new hypothesis. It is important to note, however, that previous research has demonstrated that some lexical and semantic effects involving PHs can change or even reverse as a result of using mixed versus pure lists (Borowsky et al, 2002; Marmurek & Kwantes, 1996; Reynolds & Besner, 2005; Reynolds et al., 2011). The use of pure blocks containing only words or only PHs would increase the generalizability of the present study while also providing evidence

LEXICAL AND SEMANTIC ACTIVATION WITHOUT INTENTION of the robustness of the Stroop effects found in Experiment 1, As a result of these concerns, a second experiment was conducted.

Experiment 2 The results of Experiment 1 provided novel evidence that GPC operates without intention through to the level of lexical and semantic processing, as evidenced by the finding of a significant Stroop effect for the CPHAs, In order to ensure that this finding was not simply due to the single, mixed list context that was used in Experiment 1, as both the orthographic-lexical and GPCsublexical reading pathways were activated and may have allowed the CPHs and CPHAs to be read in a more lexical (albeit phonological-lexical) manner. Experiment 2 was conducted using pure list contexts. Previous research by Marmurek and Kwantes (1996) examined the use of PHs in mixed and pure list contexts. In one experiment, Marmurek and Kwantes presented participants with PHs (e,g,, "brane"), their respective base words (e,g,, "brain"), and control NWs (e,g., "prane") in either pure lists of words and PHs, or mixed lists of words and NWs or mixed lists of PHs and NWs, Shorter naming latencies were found in pure lists relative to mixed lists regardless of whether PHs or words were analyzed, although words were responded to more quickly and named with fewer errors than PHs, Importantly, examination of the frequency effects revealed that lexical access occurred for words in both mixed and pure lists, whereas lexical access for PHs only occurred in pure lists, Marmurek and Kwantes then conducted another experiment in which they presented participants with words and PHs in a mixed list context in order to determine whether lexical access occurs for both words and PHs when being directly compared. Although words had lower error rates and shorter naming latencies relative to PHs, both the words and the PHs exhibited parallel frequency effects. As frequency effects can be taken to represent the time required to activate lexical representations (Marmurek & Kwantes, 1996), the results suggest that words and PHs access their lexical representations in a similar manner when presented in a mixed list. This suggests that the use of pure lists may increase the probability that PHs will be read via the GPC-sublexical pathway of reading in a way that cannot be biased by words in the same set and that PHs appear to access lexical representations regardless of whether they are mixed with words or presented in a pure list context (see also Borowsky et al,, 2002), To determine if any of the results of Experiment ! were due to the mixed list context that was used, a second experiment was conducted in which the stimuli were presented in pure blocks of trials consisting of either all words (e,g,, CWs and CWAs) or all PHs (e,g,, CPHs and CPHAs), In the single, mixed list context of Experiment 1, participants would be expected to use both the orthographic-lexical and GPC-sublexical pathways of reading. In contrast, when the stimuli are presented in pure list contexts, CWs and CWAs should primarily use the orthographic-lexical pathway of reading due to their familiarity as real words (although this does not rule out the use of the GPC-sublexical pathway of reading for regular words; see Experiment 3), whereas CPHs and CPHAs should use the GPC-sublexical route of reading and may be read with less involvement of the orthographic-lexical reading pathway as words are no longer part of the list context. Therefore, the CPHs and CPHAs should not be affected by any external (i,e,, contex-

629

tual) semantic and/or lexical access, and any significant Stroop effects observed for PHs in a pure list context can be attributed to their involvement of their lexical and semantic representations, and not due to any biasing effects of words. Thus, one of the hypotheses for Experiment 2 was that, to the degree that words (e,g„ CWs, CWAs) and PHs (e,g„ CPHs, CPHAs) are read unintentionally via the orthographic-lexical and GPC-sublexical pathways of reading, respectively, standard Stroop Congruency effects should be found for both types of stimuli, replicating Experiment 1, The second hypothesis of Experiment 2 concerned the comparison of results between Experiment 1 and Experiment 2, which was conducted in order to determine whether the results of the experiments differed as a function of the use of mixed and pure list contexts. It was hypothesized that the four types of stimuli would produce significant Stroop effects across both experiments, although the PHs may show smaller Stroop effects in a pure list context compared to a mixed list context because they would no longer benefit from any additional lexical access that may be caused by the presence of word stimuli. Finding significant Stroop effects in both conditions would provide further evidence that individuals are likely to access lexical and semantic representations of letter strings without intention when they are presented at fixation.

Method The same methods were used in Experiment 2 as in Experiment 1, with the following exceptions. Participants. Participants in Experiment 2 were an additional 48 undergraduate students from the University of Saskatchewan, Canada, who received partial course credit for their participation. All were fluent in English at a university level. Stimuli and design. The stimuli that were used in Experiment 1 were used in Experiment 2, and the design of the experiment was the same with the exception of the following changes. Participants viewed either pure blocks of words (CWs and CWAs), or pure blocks of PHs (CPHs and CPHAs), with the order of presentation of the actual Color Type stimuli (CWs or CPHs) and associate Color Type stimuli (CWAs or CPHAs) counterbalanced across participants to counteract any order effects. As in Experiment 1, each type of stimulus was randomly presented eight times in a congruent color condition and eight times in incongruent color conditions, resulting in 72 congruent and 72 incongruent trials for a total of 144 trials per stimulus type. As each participant either viewed both CWs and CWAs together, or CPHs and CPHAs together, each participant viewed a total of 288 trials per experiment, A 2 X 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate] X Stimulus Type [word, PH]) mixedmeasures design was used in the experiment, with within-subject factors of Congruency and Color Type, and a between-subjects factor of Stimulus Type, Apparatus and procedure. The apparatus and procedure were identical to those in Experiment 1, except for the blocked presentation of the stimuli. As in Experiment 1, participants completed 18 practice trials. In contrast to Experiment 1, however, the practice trials corresponded only to the types of stimuli that the participants would be viewing during the experimental trials. Participants then completed the 288 experimental trials corresponding to either pure lists of 144 CWs and 144 CWAs, or 144 CPHs and 144 CPHAs,

630

ANTON, GOULD, AND BOROWSKY

Following completion of the experimental trials, participants were debriefed as to the nature of the study. The entire experiment took approximately 20 min to complete.

Results The same approach to data analyses was employed in Experiment 2 as in Experiment 1, with the exception that Stimulus Type was analyzed as a between-subjects factor. Median RT omnibus ANOVA. Correct median RT data were analyzed using a 2 X 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate] X Stimulus Type [word, PH]) mixed-measures omnibus ANOVA, with Congruency and Color Type as within-subject factors and Stimulus Type as a between-subjects factor. The average median RT was 655.000 ms for congruent CPHs, 787.583 ms for incongruent CPHs, 709.271 ms for congruent CPHAs, and 736.979 ms for incongruent CPHAs. The average median RT was 670.375 ms for congruent CWs, 835.625 ms for incongruent CWs, 702.271 ms for congruent CWAs, and 742.104 ms for incongruent CWAs. There was a significant main effect of Congruency, F(l, 46) = 165.568, MSE = 2418.928, p < .001, as the median RT taken to respond to congruent stimuli was faster (M = 684.229 ms) than the median RT taken to respond to incongruent stimuli (M — 115.513 ms). The main effect of Color Type did not reach significance, F(l, 46) = 2.324, MSE = 4335.619, p = .134, nor did the main effect of Stimulus Type reach significance, F(l, 46) = 0.267, MSE = 42490.402, p = .608. There was no evidence of an interaction between Congruency and Stimulus Type, F(l, 46) = 2.488, MSE = 2418.928, p = .122, whereas the interaction between Color Type and Stimulus Type approached significance, F(l, 46) = 2.950, MSE = 4335.619, p = .093. Similar to Experiment 1, an interaction between Congruency and Color Type was found, F(l, 46) = 92.852, MSE = 1713.502,p < .001, indicating a larger difference between the congruent and incongruent actual Color Type stimuli than between the congruent and incongruent associate Color Type stimuli. There was no evidence of a Stimulus Type X Congruency X Color Type interaction, F(l, 46) = 0.739, MSE = 1713.502, p = .395 (see Figure 4). Median RT simple effects ANOVAs. The significant Congruency X Color Type interaction found in the median RT omnibus ANOVA was further investigated through simple effects ANOVAs for the two Stimulus Types (e.g., PHs and words) and was completed through the use of a 2 X 2 (Congruency [congruent, incongruent] X Color Type [actual, associate]) repeatedmeasures GLM ANOVA for each Stimulus Type. For the PH simple effects ANOVA, there was a main effect of Congruency, F(l, 23) = 131.570, MSE = 1171.695, p < .001, indicating that the font colors of congruent CPHs and CPHAs (M = 682.135 ms) were named faster than the font colors of incongruent CPHs and CPHAs (M = 762.281 ms). There was tio evidence for a main effect of Color Type, F(l, 23) = 0.020, MSE = 4052.351, p = .889. As Figure 4 shows, there was a Congruency X Color Type interaction, F(l, 23) = 57.437, MSE = 1148.952, p < .001, which indicated a greater effect of Congruency on the actual Color Type stimuli than on the associate Color Type stimuli. The repeatedmeasures 95% CI (Loftus & Masson, 1994) was ±18.769 ms. For the word simple effects ANOVA, a significant main effect of Congruency was found, F(l, 23) = 68.834, MSE = 3666.161,p