Repeated Masked Category Primes Interfere with ... - Semantic Scholar

Journal of Experimental Psychology: Learning, Memory, and Cognition in press

Copyright 2005 by the American Psychological Association

Repeated Masked Category Primes Interfere with Related Exemplars: New Evidence for Negative Semantic Priming Dirk Wentura and Christian Frings Saarland University, Saarbrücken, Germany

In four experiments, we found evidence for negatively signed masked semantic priming effects (with category names as primes and exemplars as targets) using a new technique of presenting the masked primes. By rapidly interchanging prime and mask during the SOA, the total prime exposure was increased to a level comparable to that of a typical visible prime condition without increasing the number of participants having an awareness of the prime. The negative effect was only observed for low dominance exemplars, but not for high dominance exemplars. It was found using lexical decision (Experiments 1 and 2), using lexical decision with a response-window procedure (Experiment 3), and using the pronunciation task (Experiment 4). The results are discussed with regard to different theories on semantic priming.

There is a long-standing debate in cognitive psychology about the automatic processing of subliminally presented stimuli (see, e.g., Holender, 1986; Kihlstrom, 1987; Klinger, Burton, & Pitts, 2000). Typically, by using the priming paradigm it is explored whether processing of a target stimulus is influenced by a related prime that is presented very briefly and usually overwritten by a mask stimulus. Stimuli are considered as “subliminal” if direct categorization or recognition of these stimuli is at chance level. In recent research, studies using masked primes clearly split into two clusters – response priming and semantic priming (Klinger et al., 2000; see also Wentura, 2000, for this distinction). In response priming tasks the focal manipulation of the prime is confounded with the response that is needed for the correct classification of the target. For example, target words that are either positively or negatively connoted have to be categorized with regard to their evaluative meaning. Connotation of prime words is either congruent to the target (and therefore to the correct response) or incongruent. For this type of priming task there is little doubt that replicable congruency effects with subliminally presented stimuli can be found (Damian, 2000; Draine & Greenwald, 1998; Greenwald, Draine, & Abrams, 1996; Klinger et al, 2000; see also DeHaene et al., 1998; Naccache & Dehaene, 2001; Vorberg, Mattler, Heinecke, Schmidt, & Schwarzbach, 2003). An obvious explanation for these kinds of priming effects is the assumption that the prime prepares either for the correct or for the wrong response needed for the target (Klinger et al., 2000; see also Klauer, Roßnagel, & Musch, 1997; Wentura, 1999, 2000).

Dirk Wentura and Christian Frings, Department of Psychology, Saarland University, Saarbrücken, Germany. Correspondence concerning this article should be addressed to Dirk Wentura, Department of Psychology, Saarland University, Building 1, P.O. Box 15 11 50, 66041 Saarbrücken, Germany, email: [email protected].

Masked Semantic Priming Remains a Challenging Issue The second cluster of priming studies are variants of the traditional semantic priming paradigm (see Neely, 1991, Lucas, 2000, Hutchison, 2003), which typically use the lexical decision task or the pronunciation task. In contrast to response priming, the prime manipulation is neutral with regard to the response needed for the target. For example, a category exemplar (e.g., apple) is preceded by the related category name (fruit) or an unrelated one (e.g., bird). For brief but supraliminally presented primes, it is a robust phenomenon that categorization of the target as a legal word is facilitated by a related, compared to an unrelated, prime. Note, that both the related and the unrelated primes are words. Facilitation cannot be explained by a simple process of response preparation triggered by the prime event. It must be related to processes within the memory system, a feature that reveals the semantic priming paradigm as one of the most important “windows to the mind” (see also Hutchison, 2003). Whereas there is evidence for semantic priming at very short stimulus onset asynchronies (e.g., the prime is overwritten by the target after 50 ms exposure; see Perea & Gotor, 1997; Perea & Rosa, 2002), for subliminal semantic priming, however, evidence is not as clear-cut as for response priming. In recent experiments, Bodner and Masson (2003) found evidence for masked semantic priming while Klinger and colleagues (2000), however, contrasted response and semantic priming effects and found only evidence for the former but not the latter. In this research, procedural details were comparable to supraliminal priming experiments (except prime duration and the masking itself, of course). Discussion of subliminal semantic priming becomes more complicated, however, if we take into account a second tradition of subliminal semantic priming research. This research is characterized by at least two features that set it apart from supraliminal priming research.

2

WENTURA AND FRINGS

First, the priming task is preceded by threshold setting procedures. For example, in a first phase of the experiment for each participant an objective threshold is determined that is defined by chance performance in guessing whether a masked word or a mask blank screen had been shown (see, e.g., Balota, 1983; Fowler, Wolford, Slade, & Tassinary, 1981; Marcel, 1983). As alternative procedures, repetition or semantic judgments are used that require participants to choose which of two words was shown as the masked item or which of two words was semantically related to the masked stimulus. It is interesting to learn that different procedures moderated the sign of the effect that was found in a subsequent priming task. Most noteworthy, if participants had repeatedly chosen in the first phase one of two target words which was more semantically similar to the masked word, a negatively signed semantic priming effect was obtained (Dagenbach, Carr, & Wilhelmsen, 1989; Carr & Dagenbach, 1990; Kahan, 2000). The dependence of the effect on preceding procedures that force participants to focus on the prime event suggests that those effects are of strategic nature. Second, the stimulus-onset asynchrony (SOA) of prime and target was unusually long. For example, Carr and Dagenbach (1990) used a SOA of about 1200 ms, Kahan (2000) used a SOA of at least 566 ms. (In contrast, Klinger et al., 2000, used a SOA of 67 ms.) For supraliminal priming research it is known that effects in studies using long SOAs might reflect prospective strategies (i.e., generation of expectancies; see Neely, 1977, 1991). That is not to say that any subliminal effect found with long SOAs must be due to the use of prospective strategies. Nevertheless, research by Neely (1977) showed that strategies do not distort automatic priming effects gathered with short SOAs but those gathered with long SOAs. To summarize the findings with regard to masked semantic priming, even after many years of intensive research there is considerable debate about the existence of this effect. We do not know whether null results are due to a clear absence of activation processes with masked primes or whether they result from analyzing heterogeneous samples with some participants showing positive effects (due, e.g., to spreading activation) and some showing negative effects (due to some not yet understood mechanisms or, perhaps, because of the use of strategies). We do not yet know whether some specific preconditions have to be met to produce masked semantic priming effects. And we do not yet know whether masked semantic priming effects – if there are any – are of strategic character. Given this, it is interesting to note that one procedural parameter of interest was typically not manipulated in previous studies.

Solving a Confound Which is Almost Always Overlooked If results after using clearly visible primes are compared with those after using masked primes, a simple confound is often overlooked. Usually, masked primes are presented for, e.g., 14 to 50 ms before they are overwritten by a random letter of strings. Conversely, in unmasked priming studies typically the prime is presented for, e.g., 100 to 200 ms. (Even if the duration is shorter, the luminescence of the screen lengthens the functional duration.) Thus, typically the total noticeable duration of exposure of the prime is less in masked priming studies compared to supraliminal experiments. Therefore, a failure to find subliminal semantic priming effects might be due

to the shorter duration of exposure and not to a lack of prime awareness. Would it be possible to lengthen prime duration without presenting it supraliminally? We propose the following technique: By rapidly interchanging prime and mask during the stimulus onset asynchrony, we can lengthen the total prime duration for the masked prime condition such that it is comparable to a supraliminal prime condition. The subjective impression of this presentation is a brief flicker such that participants claim to have seen nothing but the masking pattern. Of course, a subsequent direct test has to show for whom this kind of masking establishes an objective threshold (Cheesman & Merikle, 1985; see also Kahan, 2000). This procedure might allow for a better estimate as to whether subliminal primes produce semantic priming effects. In the following Experiments, we investigated priming by category names (i.e., fruit as the related prime for apple). We chose a SOA of 286 ms (i.e., 20 refresh cycles of the computer screen). For the masked presentation mode, prime and mask alternated for 20 refresh cycles. Thus, the overall presentation duration of the prime was 143 ms. In Experiment 1, we contrast the masked condition with a visible priming condition, using a prime duration of 143 ms (10 cycles) that is followed by a blank screen of another 143 ms. The hypothesis is that a priming effect will occur which will be clearly deviant from zero in the repeated-masked condition for participants that operate at the chance level in a subsequent direct test. Given previous research (see above), we are open to the possibilities of obtaining either a positively signed or a negatively signed effect. For the visible priming condition, we expect a positive effect in accordance with supraliminally semantic priming research with short SOAs (see Neely, 1991).

Experiment 1 Method Participants. Participants were 51 students (40 women; 11 men) with a median age of 21 years; all were native speakers of German. Design. Presentation mode (visible vs. repeated-masked) was manipulated between subjects. Two factors were varied within subjects. First, there were three prime conditions (related, unrelated, neutral). Second, in accordance with former studies (e.g., Neely, Keefe & Ross, 1989) using category primes, we used high and low dominance exemplars as targets. Material. For each of four categories (birds, insects, flowers, and fruits) three high dominant and three low dominant exemplars were selected (see Appendix). High dominant exemplars had a mean association frequency of 67.1 % (SD = 10.7 %; range 55 % to 86.5 %), whereas low dominant exemplars had a mean association frequency of 6.2 % (SD = 2.87 %; range 2.5 % to 11.5 %; Mannhaupt, 1983). Mean length was 5.2 (SD = 0.8; range 4 to 7) for the high dominant exemplars and 5.4 (SD = 0.5; range 5 and 6) for the low dominant exemplars. Comparable to other studies (e.g., Neely et al., 1989), dominance was somewhat confounded with word frequency. Specifically, median frequency counts for the set of high-dominance and low-dominance exemplars were 29 and 4, respectively (according to the German database of CELEX, Nijmegen, Netherlands; six million entries). Pronounceable non-words were created by changing one letter of each target word, such that it was necessary for participants to pay attention to the meaning of the targets (see Klinger et al., 2000). Procedure. Participants were seated in front of a standard personal computer. Instructions were given on the CRT screen. Participants were told that words belonging to the categories birds, insects, flowers, and fruits would be presented on the screen, which were either correctly written or misspelled. They were asked to quickly categorize each word

REPEATED MASKED CATEGORY PRIMES

3

Table 1 Mean Response Times (in ms) as a Function of Prime Condition, Category Dominance of Target Exemplars, and Awareness Status of Participants (Errors in % in Parentheses; Priming Effects (in ms; Standard errors in Parentheses) for Response Times (Experiment 1) Low-d’a Cong. Incong. Neutral

High-d’b Cong. Incong. Neutral

663 (6.0) 590 (2.8)

630 (5.1) 584 (1.4)

631 (8.3) 590 (3.2)

624 (5.2) 601 (5.2)

654 (8.3) 607 (2.1)

644 (7.3) 588 (5.2)

Visible Prime Low Dominance

-

-

-

High Dominance

-

-

-

611 (8.0) 575 (4.0)

652 (7.0) 585 (2.3)

604 (4.0) 579 (2.3)

Mean RTs and Error Rates Repeated Mask Low Dominance High Dominance

Priming Effects Repeated Mask Low Dominance -33 * (12) 30 (14) High Dominance -6 (9) 6 (15) Visible Prime Low Dominance 41 * (9) High Dominance 11 (10) Note: Priming scores are calculated by subtracting mean response times for related priming from mean response times for unrelated priming. (Slight inconsistencies between the upper and lower part of the table are due to rounding.) a Sub-sample of participants with low d’ in the direct test. b Sub-sample of participants with high d’ in the direct test. * p < .02

with regard to orthography by pressing either a yes-key with the right index finger for correctly written words or the no-key with the left index finger for misspelled words. A trial began with a fixation stimulus (a ‘+’-sign) appearing in the center of the screen for 500 ms. It was followed by the prime presentation. The prime words were written in capital letters (in standard MS-DOS Text font): FRUCHT (fruit), INSEKT (insect), BLUME (flower), and VOGEL (bird). The related prime was always the category name that corresponded to the target. The unrelated prime was always INSEKT (insect) for fruit exemplars, FRUCHT (fruit) for insect exemplars, VOGEL (bird) for flower exemplars, and BLUME (flower) for bird exemplars. The neutral prime consisted of five random consonant letters. For the visible prime condition, the fixation stimulus was overwritten by the prime, which remained on the screen for 143 ms, and was followed by a blank screen for another 143 ms. Participants of this sample were informed that for a very short duration a first stimulus (i.e., the prime) would appear in each trial to which they should not react. For the repeated masked prime condition, a forward mask was presented for one refresh cycle of the video screen (i.e., 14 ms). It was a randomly generated string of eight consonant capital letters (e.g., LMSDZKHW), that was immediately overwritten by the prime. Random letters were added to the left and to the right of the prime to form a string of eight letters. The prime and a second mask consisting of eight random letters were alternately presented (every refresh cycle) for a total of 20 refresh cycles. Thus, the total prime duration was 143 ms as well. Then, the target stimulus appeared that remained on the screen until a response was given. Latency of response was recorded to the nearest millisecond

(Haussmann, 1992). In the case of an error, an error message appeared on the screen for 500 ms. The blank screen interval between the response and the beginning of the next trial was 1000 ms. At the beginning, participants worked through 24 practice trials (i.e., a randomly selected half of the targets were presented once) in order to become familiarized with the task. The main part of the experiment was started by another 24 practice trials (i.e., the other half of the targets were presented a first time) followed by three blocks of 48 trials each. Within a block, each target word was presented once in one of the three priming conditions. The sequence of priming conditions for a given target was determined by a Latin-square design (i.e., it was made up by three subsamples of participants, three subsets of targets, and three sequences of priming conditions). A rest break was provided after every 24 trials. For the masked prime condition, a direct measure of prime categorization was administered subsequently to the priming task. Thirty-two more trials (eight trials for each prime word) were presented with a row of question marks instead of a target word. Participants were instructed to categorize the prime. If the prime was either the word flower or the word bird, the word flower was presented to the left of the question marks and the word bird to the right. Accordingly, if the prime was either the word insect or fruit, insect was presented to the left, fruit to the right of the question marks. Participants were instructed to categorize by pressing the corresponding key. Before the direct measure, participants practiced this task with 10 trials.

4

WENTURA AND FRINGS

Results Unless otherwise noted, all effects referred to as statistically significant throughout the text are associated with p values less than .05, two-tailed.1 Direct effects. For the repeated masked prime group (n = 26), the signal detection sensitivity for the masked primes was d’ = .61 (SD = .80). Inspection of the distribution reveals that the sample clearly separates into two sub-samples. (1) Eight participants showed a significant contingency between prime category (i.e., flower/insect vs. bird/fruit; see above) and response category (i.e., left key vs. right key), all χ2 > 4.56, p < .05 (d’ was in the range of 0.99 to 2.68); (2) the remainder of the sample (n = 18) had a d’ in the range of –0.64 to 0.65 (M = 0.17, SD = 0.40; all individual χ2 < 2.04, p > .14); hence, it can be assumed that for those participants the masking of primes was successful. We refer to the two sub-samples as low-d’ and high-d’ participants (see Klinger & Greenwald, 1995). Priming effects. Mean RTs were derived from correct responses only. The average error rate across participants was 4.71 % and 5.47 % for words and non-words, respectively. RTs that were 1.5 interquartile ranges above the third quartile with respect to the individual distribution (see Tukey, 1977) or were above 1500 ms were discarded as well (3.16 % and 3.84 % of all word and non-word decisions, respectively). The mean RTs and error rates for word targets are shown in Table 1.2 To simplify the exposition, priming indices were calculated as the difference between unrelated and related priming. Priming differences were subjected to a 3 (presentation: masked-low-d’ vs. masked-high-d’ vs. visible) x 2 (dominance: low vs. high) analysis of variance. There was a significant main effect of presentation, F(2,48) = 15.43, MSe = 1461, and a significant interaction of presentation and dominance, F(2,48) = 3.39, MSe = 2747. For the visible condition, responses to related primetarget pairs were significantly faster than to unrelated pairs, ∆ = 26 ms (d = 0.87), F(1,24) = 18.99, MSe = 1762. This priming effect was significantly moderated by dominance in the analysis using participants’ means, F(1,24) = 4.50, MSe = 2592 (F[1,22] = 3.00, MSe = 1577, p < .10, using targets’ means). Whereas the priming effect for low dominance exemplars was significant, M = 41 ms (d = 0.92), t(24) = 4.59, the effect for high dominance exemplars was not, M = 11 ms (d = 0.24), t(24) = 1.09, ns. For the masked low-d’ sample, responses to related primetarget pairs were significantly slower, ∆ = -20 ms (d = 0.73), compared to unrelated pairs, F(1,17) = 9.69, MSe = 1434. This priming effect was not significantly moderated by dominance, F(1,17) = 2.31, MSe = 2917, p = .15. However, simple effects show an asymmetry as well: Whereas the effect for low dominance targets was significantly below zero, M = -33 ms (d = 0.64), t(17) = -2.71, the effect for high dominance exemplars was not, M = -6 ms (d = 0.15), t(17) = -0.63, ns. The sub-sample of high-d’ participants within the masked prime sample (n = 8) had a significant positive priming effect (by using participants’ means) of 18 ms (d = 1.15), F(1,7) =

1

Unless otherwise noted, all priming effects significant in the withinsubjects analyses reported in this article were also significant when the error term was computed across the 24 word targets. 2 Mean RTs and mean error rates for non-word trials for all experiments are presented in Appendix B.

10.58, MSe = 498 (F[1, 23] = 2.99, MSe = 2357, p < .10, using targets’ means), that was not moderated by dominance, F(1,7) < 1. Again, simple effects show an asymmetry: Whereas the effect for low dominance targets was significantly above zero, M = 30 ms (d = 0.77), t(7) = 2.17, p < .05 (one-tailed), the effect for high dominance exemplars was not, M = 6 ms (d = 0.14), t(17) = 0.41, ns. As expected for short SOA-research (see Neely, 1977), for the repeated-mask condition the RT means of the neutral condition are most similar to those of the unrelated condition (i.e., effects seem to be due to processes instigated by a related prime). In this regard, it is somewhat unexpected that the mean RTs of the visible presentation mode for the neutral condition are as low as that for the related one. However, the use of a random letter string as a neutral condition seems to be well suited for the masked condition (because it resembles the masks). For the visible presentation mode, however, it might have unknown properties that render it somewhat dubious as a neutral baseline (see, e.g., Jonides & Mack, 1984; Neely, 1991).3 For the error data, there were no significant effects in the 3 (presentation) x 2 (dominance) analysis of variance of priming differences, all Fs < 2.84, ns.

Discussion Our data show a clear dissociation of priming effects between the repeated mask condition (low-d’ participants) and the visible presentation mode. For a visible presentation of primes, as expected, a congruency effect was observed: Following the related category prime, lexical decisions to category exemplars were faster compared to the unrelated prime condition. For the repeated mask presentation and participants who responded at chance level in a direct test (low-d’ participants), the effect, however, was reversed. Now, a category prime increases response latencies to a related exemplar target. However, the effect seems to be restricted to low dominance exemplars. To replicate this pattern and to uncover any advantage of the new technique, in Experiment 2 we test it in comparison to the more typical presentation mode of displaying the prime, while maintaining, however, the SOA of 286 ms constant.

Experiment 2 Method Participants. Participants were 48 students (30 women; 18 men) with a median age of 21 years; all were native speakers of German. Five participants were replaced because of an error rate for words above 15 %. Design, Materials, and Procedure. Design, Materials, and Procedure were essentially the same as in Experiment 1 with the following exception. The visible prime condition of Experiment 1 was replaced by a standard masked condition. In this condition, the prime was presented

3 For example, comparing the difference in RTs following word primes minus neutral primes for word targets (∆ = -14 ms; see Table 1) and non-word targets (see Appendix B; ∆ = 20 ms) reveals an interaction pattern for the visible presentation mode (i.e., neutral primes seem to facilitate word responses and inhibit non-word responses) of which, however, an interpretation is beyond the scope of the article.

5


Table 2 Mean Response Times (in ms) as a Function of Prime Condition, Category Dominance of Target Exemplars, and Awareness Status of Participants (Errors in % in Parentheses; Priming Effects (in ms; Standard errors in Parentheses) for Response Times (Experiment 2)

Mean RTs and Error Rates Repeated Mask Low Dominance High Dominance Standard Masked Prime Low Dominance High Dominance

Low-d’a Cong. Incong. Neutral


687 (6.2) 625 (2.6)

652 (8.3) 627 (2.1)

679 (6.3) 634 (1.0)

660 (6.2) 642 (2.1)

668 (3.1) 631 (2.1)

663 (3.1) 625 (1.0)

673 (5.2) 629 (1.6)

654 (8.9) 621 (1.6)

674 (2.6) 630 (1.6)

623 (5.2) 571 (1.0)

631 (8.3) 596 (3.1)

647 (5.2) 579 (2.1)

Priming Effects Repeated Mask Low Dominance -35 * (8) 8 (14) High Dominance 3 (8) -11 (22) Standard Masked Prime Low Dominance -19 (13) 8 (10) High Dominance -8 (12) 25 (19) Note: Priming scores are calculated by subtracting mean response times for related priming from mean response times for unrelated priming. (Slight inconsistencies between the upper and lower part of the table are due to rounding.) a Sub-sample of participants with low d’ in the direct test. b Sub-sample of participants with high d’ in the direct test. * p < .001 for two refresh cycles (i.e., 28 ms); it was overwritten by the second mask, which was presented for one refresh cycle (i.e., 14 ms). In order to make the SOA equivalent to the repeated-masked prime condition, a blank screen was presented for 17 cycles (i.e., 243 ms).

Results Direct effects. We found an overall signal detection sensitivity for the masked primes of d’ = .49 (SD = .69) for the repeated masked prime sample and of d’ = .50 (SD = .62) for the standard masked sample. The difference was not significant, t(46) = 0.07. We used almost the same criteria for splitting the samples as in Experiment 1. (1) Eight participants of the repeated masked sample showed a significant (or nearly significant) contingency between prime category and response, all χ2 > 3.46, p < .06 (d’ was in the range of 0.89 to 1.56); (2) the remainder of the sample (n = 16) had a d’ in the range of – 0.81 to 0.81 (M = 0.15, SD = 0.59; all individual χ2 < 3.14, p > .07). (3) Eight participants of the standard masked sample showed a significant contingency between prime category and response, all χ2 > 4.50, p < .05 (d’ was in the range of 0.98 to 1.77); (4) the remainder of the sample (n = 16) had a d’ in the range of –0.81 to 0.83 (M = 0.17, SD = 0.46; all individual χ2 < 3.24, p > .07).

Priming effects. RTs were treated as in Experiment 1. (Error rates were 3.85 % and 5.38 % for words and non-words, respectively; outlier rates were 3.88 % and 5.56 % for words and non-words, respectively). The mean RTs and error rates for word targets as well as mean priming differences are shown in Table 2. Data were subjected to a 2 (presentation: repeated masked vs. standard masked) x 2 (awareness: low-d’ vs. high-d’ ) x 2 (dominance: low vs. high) analysis of variance with priming differences as the dependent variable. There was a significant main effect of awareness, F(1,44) = 4.40, MSe = 2484; the triple interaction just missed the conventional criterion of significance in the participants’ analysis, F(1,44) = 3.72, p =.06, MSe = 1457 (F[1, 22] < 1.38, MSe = 3027, ns, in the analysis using targets’ means), all other Fs < 2.37. For the repeated mask sample, there was a significant interaction of awareness and dominance, F(1,22) = 4.65, MSe = 1884, all other Fs< 1.83. For the low-d’ sample, there was a significant priming effect (in the analysis using participants’ means), F(1,15) = 7.73, MSe = 1111 (F[1, 22] = 2.59, MSe = 1873, p = .12, in the analysis using targets’ means) that was, however, significantly moderated by dominance, F(1,15) = 11.24, MSe = 1025. The effect for low dominance targets was significantly below zero, M = -35 ms (d = 1.06), t(15) = -4.24, whereas the effect for high dominance exemplars was not, M = 3 ms (d = 0.08), t(15) = 0.32, ns. Thus, Experiment 1 was

6

WENTURA AND FRINGS

perfectly replicated. This time, the moderation by dominance clearly shows up.4 Moreover, to account for the confound of dominance and frequency (see Materials of Experiment 1), we simultaneously regressed the priming differences for the 24 targets on dominance (binary coded) and frequency. There was again an effect of dominance, t(21) = 2.37, p < .05, but none of frequency, t(21) = -0.63, p > .50. For the high-d’ sample there was neither a priming effect (i.e., no overall priming collapsed over dominance) nor a dominance effect, all Fs < 1. For the standard mask sample, there were no significant effects in the 2 (awareness) x 2 (dominance) analysis of priming differences, all F(1,22) < 2.68. Even if we restrict our analyses to the low-d’ sample, there was neither a significant overall priming effect, F(1,15) = 1.47, MSe = 4001, ns, nor a moderation of priming by dominance, F(1,15) < 1. Neither the effect for low dominance targets was significantly below zero, M = -19 ms (d = 0.37), t(15) = -1.50, ns, nor the effect for high dominance exemplars, M = -8 ms (d = 0.16), t(15) = -0.64, ns. For the error data, there were no significant effects in the 2 (presentation) x 2 (awareness) x 2 (dominance) analysis of variance of overall priming differences, all Fs < 2.55, ns.

Discussion First of all, Experiment 2 shows that the pattern of focal interest – i.e., priming effects for low-d’ participants – was clearly replicated. Again, for low dominance exemplars there was a substantial negative priming effect that was completely missing for high dominant exemplars. Secondly, this result cannot be easily attributed to different frequencies of low and high dominant targets, as the regression analyses of priming differences for targets has shown. Thirdly, we can conclude that there might be indeed some advantage to the new technique because a standard masking technique would not have succeeded in showing any effect (given the sample size). Of course, we have to concede that the standard masking procedure yielded a pattern of priming effects for the low-d’ sample that was numerically comparable to the repeated prime condition. Additionally, the differentiation between the results for the two techniques was only marginal with regard to statistical inference. However, relying on the standard method would not have resulted in a significant priming effect, whereas the effect for the repeated-mask condition found in Experiment 1 was easily replicated. For further experiments it can be stated that the negative effect for low dominance exemplars in the standard-mask sample – if we concede its existence – of d = 0.37 can be detected with probability 1-β = .95 (α = .05) if one uses a sample size of N = 97 (see Erdfelder, Faul, & Buchner,

4 It might be seen as somewhat disturbing that the unrelated and neutral priming condition do not resemble each other with regard to mean RTs for the condition of focal interest (i.e., the low dominance exemplars in the low-d’ sample). However, there was one participant with an outlying mean RT for neutral primes (i.e., the difference between mean RTs for neutral primes minus mean RTs for category primes was extreme). Ignoring the data of this participant does not alter the priming effect of focal interest (i.e., the difference in mean RTs between the unrelated and the related condition for low dominance targets; M = -37 ms, SE = 9 ms) whereas the priming difference ‘neutral minus related’ was now significantly below zero as well, M = -16 ms (SE = 9 ms), t(14) = -1.84, p < .05 (one-tailed).

1996). The corresponding effect in the repeated mask condition of d = 1.06 can be detected with 1-β = .95 (α = .05) if one uses a sample of N = 14. To corroborate further the finding of a negative priming effect, we should replicate the result by adopting a technique that was recently used successfully by others to substantiate the claim of subliminal response priming. Greenwald and colleagues (1996; Draine & Greenwald, 1998) introduced a response window procedure that forces participants to respond within a narrow time frame after target presentation thereby making percentage of errors the dominant dependent variable. This procedure controls for speed and accuracy problems by forcing all response latencies to be relatively similar, thereby concentrating the influence of the primes on accuracy and (perhaps) increasing the size of priming. With regard to priming effects, it might be relevant that the targets are not yet fully processed before a response is given: For response priming, it can be argued that in generating a response the weight of the task-relevant target attribute is relatively decreased compared to the corresponding prime attribute (Wentura & Rothermund, 2003). For semantic priming, it can be argued that a related prime is of more help to fully activate a degraded target (e.g., Becker & Killion, 1977; Stolz & Neely, 1995). Insofar as responding to a not yet fully processed target mimics responding to degraded targets, we might obtain a more pronounced effect (see Klinger et al., 2000). Additionally, using this procedure is important for at least two reasons. First, Klinger et al. (2000) found no evidence for semantic priming using it. Thus, if one accepts the response window procedure as a new standard, we have to test whether our negative effect will be found with it. Second, negative semantic priming effects that were found with preceding threshold setting procedures and long SOAs were explained by theories that heavily rest on participants’ usage of strategies to retrieve the meaning of the prime (Dagenbach et al., 1989, Carr & Dagenbach, 1990; Kahan, 2000). We will further elaborate on this topic in the General Discussion. At this point it is sufficient to note that the use of time-consuming strategies is rather unlikely, given the response window procedure.

Experiment 3 Method Participants. Participants were 31 students (20 women; 11 men) with a median age of 22 years; all were native speakers of German. Two participants were replaced because their rate of responses that fell into the response window was below 25 %. Design, Materials, and Procedure. Design, materials, and procedure were essentially the same as in the repeated mask prime conditions of Experiment 1 and 2 with the following exception: Target presentation and instructions were adapted to the response window technique as introduced by Draine and Greenwald (1998). That is, the target word was replaced after 400 ms by an exclamation mark that remained on the screen for 450 ms. Onset of the exclamation mark defined the beginning of the response window. The window width was 150 ms. If the response fell within the range defined by the response window (i.e., from 400 ms to 550 ms after onset of the target), the exclamation mark changed colors (from white to red), giving feedback to the participant that the response was in time. If it did not change colors, the participant knew that the response had been too slow. If the response was given too quickly, the target was erased immediately and the exclamation mark


7

Table 3 Mean Error Rates (in %) as a Function of Prime Condition and Category Dominance of Target Exemplars (Mean RTs in ms in parentheses) and Priming Effects for Error Rates (Standard Errors in Parentheses; Experiment 3)

Mean RTs Low Dom. High Dom.

Low-d’a Cong. Incong. Neutral


27.0 (517) 11.8 (475)

22.9 (496) 10.7 (479)

18.4 (524) 11.3 (486)

23.0 (509) 13.7 (485)

13.7 (503) 10.1 (486)

18.5 (494) 12.2 (477)

Priming Effects Low Dom. -8.6 * (3.8) -9.2 * (3.8) High Dom. -0.5 (2.9) -0.6 (3.4) Note: Priming scores are calculated by subtracting mean response times for related priming from mean response times for unrelated priming. (Slight inconsistencies between the upper and lower part of the table are due to rounding.) a Sub-sample of participants with low d’ in the direct test. b Sub-sample of participants with high d’ in the direct test. * p < .04

never appeared. Participants were instructed to press the correct key within the response window. After every 24 trials, a summarized feedback was given, indicating the percentage of correct trials, median RT, percentage of trials with an RT falling within the response window, and percentage of trials with a response that was too fast. The following message was added: “Your goal should be to maximize the rate of responses within the ‘time window.’ Simultaneously, the rate of correct responses should be about 70 % to 80 %.” During the experiment, the response window was adjusted contingent on the participant’s accuracy. That is, if the error rate of a 20-trial block was above 45 % and the RT median was 100 ms or more above the center of the response window, the onset of the window was increased by 33 ms. If, however, the error rate was below 20 % and the RT median was below the center of the response window plus 100 ms, the onset of the window was decreased by 33 ms.

Results Direct effects. We found again that the sample split into two sub-samples: (1) Fourteen participants (high-d’ participants) showed a significant contingency between prime category and response, all χ2 > 4.56, p < .04 (d’ was in the range of 0.99 to 2.04). (2) The remainder of the sample (low-d’ participants; n = 17) had a d’ in the range of –0.48 to 0.89 (M = 0.41, SD = 0.34; all individual χ2 < 3.46, p > .05). Priming effects. The average error rate across participants was 16.17 % and 30.24 % for words and non-words, respectively. The error rates for word targets are shown in Table 3. Priming indices were subjected to a 2 (awareness: low-d’ vs. high-d’) x 2 (dominance: low vs. high) analysis of variance. There was a main effect of dominance, F(1,29) = 4.86, MSe = 220.9. In contrast to Experiment 1, the status of awareness does not alter the results, neither as a main effect nor in interaction with dominance, all Fs < 1. The overall effect for low dominant targets was M = -8.9 % (d = 0.60), t(30) = -3.36, and for high dominant targets, M = -0.5 % (d = 0.04), t(30) = -0.25, ns. Testing the related condition against the neutral condition for

the low dominance exemplars decreased the effect to -4.2 % (d = 0.35), t(30) = -1.91, p < .05 (one-tailed).5 As in Experiment 2, we tested whether the dominance effect – especially the one for low-d’ participants – might be better attributed to target frequency. However, simultaneously regressing the priming differences for targets on dominance and frequency does not alter the effect of dominance, t(21) = 1.90, p < .05 (one-tailed; t[21] = -0.76, p > .46 for frequency) if targets’ means are solely based on the data of the low-d’ participants. (Using solely the data of high-d’ participants to calculate the targets’ means, leaves only frequency as a significant predictor, t[21] = 2.21, p < .05; t[21] = -0.03, p > .97 for dominance). Unsurprisingly for a response-window experiment, there were no significant effects in the 2 (awareness) x 2 (dominance) analysis of variance of overall RT priming differences, all Fs < 1.

Discussion Experiment 3 clearly replicates the finding of a negatively signed semantic priming effect for a repeated masked presentation condition given response window instructions. Again, the effect was moderated by dominance (even if one controls for frequency) with only low dominant targets contributing to the effect. There is one divergence between the results of Experiment 1 and 2 and the present experiment: Now, the effect was not moderated by prime awareness. It showed up

5 Again, we might attribute this to a bias in the neutral condition. Research on response priming (e.g., Draine & Greenwald, 1998; see also above, Introduction) shows that masked primes are categorized with regard to the task-relevant feature, thereby facilitating a congruent response and impeding an incongruent one. Thus, if word primes nonspecifically facilitate a word response (and impede a non-word response, respectively) compared to the neutral condition, the neutral baseline would be somewhat biased.

8

WENTURA AND FRINGS

Table 4 Mean Response Times (in ms) as a Function of Prime Condition, Category Dominance of Target Exemplars, and Awareness Status of Participants; Priming Effects (in ms; Standard Errors in Parentheses; Experiment 4) Low-d’a Cong. Incong. Neutral Mean RTs Low Dom. High Dom.

501 479

475 499


487 487

449 429

428 423

438 424

Priming Effects Low Dom. -26 * (9) -21 (8) High Dom. 20 * (6) -6 (5) Note: Priming scores are calculated by subtracting mean response times for related priming from mean response times for unrelated priming. (Slight inconsistencies between the upper and lower part of the table are due to rounding.) a Sub-sample of participants with low d’ in the direct test. b Sub-sample of participants with high d’ in the direct test. * p < .02

for low-d’ as well as high-d’ participants. We will return to this issue in the General Discussion. As we have argued in the introduction, priming effects in our design cannot be explained by a simple response priming process because both related and unrelated priming conditions comprised the presentation of a word: Either both of them trigger a tendency to respond with the word key (and therefore speed up a correct word decision) or neither of them.6 We even ruled out the retrospective matching process that was hypothesized by Neely and colleagues (1989) for designs that confounded wordiness of target with prime-target relatedness (due to the fact that typically primes are not related to the words that were the basis for the following non-word target). Here, all non-words were orthographically incorrect versions of the word targets and they were preceded by the related or unrelated category prime as well. Thus, a retrospective process of the kind “If I have the feeling of a semantic match, it must be a word!” is not plausible. However, a further means of determining whether priming effects are response-based is to use the pronunciation task instead of a binary decision task (see Neely, 1991, Hutchison, 2003, for a comparison of lexical decision and pronunciation results with regard to semantic priming). Therefore, in Experiment 4, we used a pronunciation task with the same word stimuli as in the experiments before.

6 Responses triggered by the prime might even be inhibited, thereby yielding a negative priming effect that superficially mimics our effect (Eimer & Schlaghecken, 1998; Klapp & Hinkley, 2002). In our experiments, either both the related word prime’s response code and the unrelated one’s are inhibited or both are not. This does not explain our priming effects.

Experiment 4 Method Participants. Participants were 16 students (15 women; 1 men) with a median age of 22 years; all were native speakers of German. Design, Materials, and Procedure. Design, materials, and procedure were essentially the same as for the repeated mask prime condition in Experiment 1 with the following exceptions. First, we used a voice-key apparatus and presentation software created by our technical staff (using Delphi and DirectX). Presentation parameters (i.e., black background, white letters, type font) were chosen to closely mimic the presentation of Experiment 1 to 3. Second, no non-word targets were presented and participants had to pronounce the target instead of categorizing it; latency of response was recorded. The experimenter coded faulty responses (i.e., all cases in which the voice-key was triggered by events other than the onset of pronouncing the target). Third, each trial was initiated by a press of a key by the participant; there was no break during the priming task.

Results Direct effects. Using the same procedure as in Experiment 1, we again split the sample into two sub-samples: (1) Four participants (high-d’ participants) showed a significant contingency between prime category and response, all χ2 > 4.04, p < .05 (d’ was in the range of 0.94 to 2.21). (2) The remainder of the sample (low-d’ participants; n = 12) had a d’ in the range of -0.21 to 0.83 (M = 0.50, SD = 0.35; all individual χ2 < 3.24, p > .07). Priming effects. Mean RTs were derived from correct responses only. The average rate of faulty triggering of the voice-key across participants was 0.87 %. RTs that were 1.5 interquartile ranges above the third quartile with respect to the individual distribution (see Tukey, 1977) or were above 1500 ms (below 150 ms) were discarded as well (2.43 %). The mean

REPEATED MASKED CATEGORY PRIMES RTs, error rates, and priming effects are shown in Table 4. Data were subjected to a 2 (awareness: masked-low-d’ vs. maskedhigh-d’) x 2 (dominance: low vs. high) analysis of variance with priming differences as the dependent variable. There was a significant main effect of dominance, F(1,14) = 8.36, MSe = 661, but not of awareness, F(1,14) = 1.42, ns; the interaction of dominance and awareness was not significant, F (1,14) = 2.17, MSe = 661, which however might be a problem of low power because the high-d’ sample consists of only four participants. For the low-d’ sample, the effect of dominance was significant as well, F(1,11) = 15.93, MSe = 791. Simple effects show an asymmetry that was even clearer than in the previous experiments: Whereas the effect for low dominance targets was significantly below zero, M = -26 ms (d = 0.87), t(11) = -3.01, the effect for high dominance exemplars was significantly above zero, M = 20 ms (d = 0.93), t(11) = 3.22. Testing the related condition against the neutral condition for the low dominance exemplars yielded a significant effect of -14 ms (d = 0.70), t(11) = -2.43. This was not the case for the high dominance exemplars, M = 7ms (d = 0.37), t(11) = 1.27, ns. As before, we tested whether the dominance effect might be better attributed to target frequency. Simultaneously regressing the priming differences for targets on dominance and frequency, however, does not substantially alter the effect of dominance, t(21) = 1.84, p < .05 (one-tailed; t[21] = 1.42, p > .16 for frequency). For the sub-sample of high-d’ participants (n = 4) we did not find a dominance effect, F(1,3) = 2.38, ns; the overall priming effect (M = -12 ms; d = 1.40) just missed the conventional criterion of significance in the analysis by participants, F(1,3) = 7.89, MSe = 188, p = .07 (but F[1, 22] = 2.53, p = .13, using targets’ means).

Discussion The results of Experiment 4 are clear. We found a pattern of priming effects which is highly comparable (with one notable exception) to that of Experiment 1 to 3 (repeated-mask condition). The negative priming effect for low dominant exemplars replicated under conditions that preclude responsebased effects. Again, we found a significant interaction of priming with dominance (even if one controls for frequency). However, whereas in Experiment 1 to 3 the effect for high dominant exemplars was reduced to a non-significant level, now even a positive priming effect emerged. For high-d’ participants, we found a pattern that is numerically comparable to the one we found throughout the Experiments 1 to 3 for the low-d’ samples. In this regard, the present experiment is comparable to Experiment 3 as no clear dissociation between the two samples was found.

General Discussion We have presented new evidence on semantic priming at subliminal (low-d’ participants) and marginally perceptible (high-d’ participants) levels. The most important results are the following: Negative semantic priming for low dominant exemplars. For low dominant exemplars of semantic categories, we consistently found a negatively signed priming effect if the category label was used as the prime and participants operated at a chance

9

level with regard to prime categorization. This effect would probably not have been detected if a standard masking procedure would have been used (see Experiment 2). However, a robust effect was achieved by using a new technique of repeatedly alternating a prime word with a random letter mask. This might be due to the fact that this technique allows for an increase in the length of prime exposure to a level comparable to that in a standard supraliminal priming experiment, while simultaneously preventing awareness of the prime for the majority of participants. Dissociation of effects for low vs. high dominant exemplars. In all four experiments using category labels as primes we found a clear dissociation of priming for low vs. high dominant exemplars. For the lexical decision experiments, the effect for high dominant exemplars reduced to clearly non-significant levels. We have to concede that two caveats remain: First, there is a confound of dominance and frequency in our materials (see Materials of Experiment 1). For visible priming it is known that effects in the lexical decision task are higher for low frequency words compared to high frequency words (see Becker, 1979). Second, we used only a small set of targets that were repeated several times. If we assume that priming effects may dissipate with repeated exposure to targets, small effects for high frequent words may turn into null effects. Thus, we must be aware that what we interpret as an effect of dominance might in fact be an effect of frequency. However, there are two counter arguments: First, for the pronunciation task, there was a significant positive effect for high-dominance targets. Of course, we should await further studies with regard to this remarkable detail, but it already helps in the present context to refute the simple explanation of the dissociation that for some unknown reason the high dominant exemplars are not susceptible to priming. Second, the regression analyses of the priming differences for the 24 targets showed that dominance remained to be a significant predictor even if we control for frequency. These results make it rather implausible to attribute the effects for dominance to differences in frequency (with the notable exception of the effect for high-d’ participants in Experiment 2). No clear-cut pattern for participants with high d’. For participants with marginal perception of the primes – as indicated by a high d’ – the pattern of effects is more complicated: In the experiments using a standard lexical decision task (Experiment 1 and 2), the best guess at the present moment would be the prediction of a small positive priming effect for low dominant exemplars and a null effect for high dominant exemplars. In the pronunciation task (Experiment 4) and in the response-window version of the lexical decision task (Experiment 3), however, we found a pattern comparable to the one seen for low-d’ prime participants. Interpreted with some caution, these findings suggest that the process that is responsible for the pattern of subliminal priming effects is also working if primes are marginally perceptible. Under some not yet specified circumstances, however, a second counteracting process might take place that is dependent on (marginal) perceptibility of primes. What differentiates Experiments 1 and 2 from Experiments 3 and 4? First and most obviously, latencies are shorter in the two latter experiments. Second, one might speculate about the absence of backward checking strategies (see, e.g., Neely et al., 1989): They cannot be applied to the pronunciation task and a response window procedure might prevent such strategies. This leads to the next point.

10

WENTURA AND FRINGS

The priming effects are not located at a response-level. We designed our experiments in the tradition of semantic priming, not in the tradition of response priming (see Introduction). Thus, our effects cannot be located at the level of response facilitation and/or response interference. The choice of our nonword materials even makes a strategy such as that proposed by Neely and colleagues (1989) unlikely as a potential explanation. In our lexical decision experiments, for words and for nonwords, the relatedness proportion was .33 because our nonwords were orthographically incorrect versions of the target words. Thus, backward checking for relatedness does not help to arrive at a faster “word”-decision. Most importantly, we found a largely comparable pattern in Experiment 4 using the pronunciation task. Since the pronunciation task has the reputation of being a more valid indicator for memory-located processes, this result is extremely important. So, how can we account for our results in terms of current theories of semantic priming? A coarse-grained taxonomy of accounts for semantic priming is between accounts that focus on prospective influences of the prime (i.e., processing the prime alters the state of memory, which is revealed by the ease of processing the target) and those that focus on retrospective memory retrieval processes that are caused by processing the target (see also Whittlesea & Jacoby, 1990).

Prospective accounts of semantic priming The most dominant explanation of semantic priming is the spreading activation account both in its traditional form (e.g., Anderson & Pirolli, 1984; Collins & Loftus, 1975) as well as in its distributed network version (e.g., Masson, 1995). This theory can hardly be applied to our results, because spreading activation is inherently bound to positive priming effects. A better candidate is the center-surround inhibition theory (CSI) of Dagenbach and colleagues (1989; see also Carr & Dagenbach, 1990; Dagenbach, Carr, & Barnhardt, 1990; Barnhardt, Glisky, Polster, & Elam, 1996; Stolz & Besner, 1997; Neely, VerWys, & Kahan, 1998) which was developed to account for negatively signed masked priming effects that were observed following a semantic threshold setting task. CSI theory proposes an active inhibition mechanism that is used only under certain conditions of low code activation that arises when stimulus input is severely limited as, for example, in masked priming. It is argued that the process causes a ‘pop out’ effect by inhibiting closely related concepts (the semantic “surround”) of a weakly activated code to make this code (the “center”) ‘easier to find’. However, center-surround inhibition is seen as a strategic process that is only instigated if retrieval processes are limited by the input. We have no indication that our participants were engaged in trying to retrieve the meaning of the prime. Of course, the possibility remains that centersurround inhibition is a mechanism which is less dependent on strategic retrieval processes, and which needs less time to evolve than previously assumed (see also Barnhardt et al., 1996). Another problem is that to apply CSI theory, we have to consider the high dominant exemplars to be the “center” of the category whereas the low dominant exemplars are the “surround.” There is some arbitrariness in this assumption because one might alternatively argue that high dominant exemplars are the “surround” of the category center, which is the category name itself. To summarize, prospective accounts like the CSI theory can in principle account for our data,

although only by adding some problematic ad hoc assumptions that have to await further tests. ´

Retrospective retrieval accounts of semantic priming To begin with, retrospective prime clarification (RPC) theory (Kahan, 2000) assumes that participants actively try to identify the prime by backward using the target information if they know that there is sometimes a semantic relationship between prime and target. Thus, there is a match/mismatch process between prime and target display. If prime and target are unrelated, this process might arrive quickly with the decision ‘no match.’ In the case of related prime-target pairs (i.e., a partial match of prime and target), however, it is more difficult to separate the memory traces of prime and target. The decision is therefore delayed. The theory provides an elegant solution to the problem (seen commonly in former studies) that negatively signed priming effects were only obtained if the threshold setting task that precedes the priming task asks for semantic similarity judgments of masked primes and targets. In RPC theory it is assumed that participants carry over this backward strategy to the priming task although identification of the prime is no longer needed (and is of no help to arrive at fast decisions). Again, to apply this theory to our results we have to make some ad hoc modifications that are not entirely satisfying. First, to apply RPC theory, we have to assume that repeatedly presenting the prime-mask sequence makes participants curious about the prime event, thus perhaps initializing the RPC process. Second, at least for Experiment 3 the use of retrospective clarification is quite unlikely (although not impossible) because participants were forced to respond very quickly, before the target was fully processed. Third, Kahan (2000) found negative priming effects for strongly associated pairs as well as for repetition pairs. Thus, an a priori hypothesis for our experiments would have been to assume negative effects especially for high-dominant exemplars. However, adapting an argument provided by Kahan to account for results of Barnhardt, Glisky, Polster, and Elam (1996), we could speculate that possibly if low-d’ participants are shown a low dominant target, they will make a long partial match between the related prime and the target, slowing responses to the target compared to the unrelated prime condition. If the target is a high dominant exemplar, low-d’ participants will quickly make a complete match between the related prime and target before responding to the target. In this case, responses are not slower compared to the unrelated condition. To summarize, RPC can account for our data, although only by adding some ad hoc assumptions. Whittlesea & Jacoby (1990; Hughes & Whittlesea, 2003) as well as Bodner & Masson (1997, 2001, 2003) proposed similar accounts (hereinafter: episodic retrieval accounts) that are based on the idea that prime events create a form of episodic resource that can be recruited to assist with target processing: The compound of the not-yet-fully identified target with a related prime is a stronger cue to memory than the target by itself. This process will be instigated especially if the target is perceptually degraded (Whittlesea & Jacoby, 1990) and if prime information is valid (i.e., if the relatedness proportion is, e.g., 80 %; Bodner & Masson, 2003). For the discussion of our results, we can add a third account, compound-cue theory (Ratcliff & McKoon, 1989), which proposes that during binary decision tasks prime and target combine to form a compound cue to memory. The familiarity of the compound is used as a basis for the decision:

REPEATED MASKED CATEGORY PRIMES For example, in a lexical decision task, if a compound is highly familiar, a quick “word” response can be made; if it is highly unfamiliar, a quick “non-word” response can be made; moderate familiarity, however, will result in time-consuming additional processing). Typically, a compound comprised of a related prime-target pair is assumed to have higher familiarity than one based on an unrelated pair. Thereby a related prime helps to identify the target (according to the episodic retrieval accounts) or helps to quickly decide that the target is a word (according to compound-cue theory). However, one might suppose that the compound of a prime category with a related low-dominance exemplar represents poorer familiarity compared to the unrelated condition. That is, although low dominance exemplars are nominal members of the prime’s category, by definition they are not very good ones. Thus, the negative semantic priming observed in our experiments could be a direct consequence of the atypicality of exemplars, which is, for example, characteristic known to lead to slower responding in the categorization literature (Rosch, 1978). On the basis of this assumption, compound-cue theory can account for our lexical decision results because Ratcliff and McKoon (1988) hold that a target word always combines with its prime (i.e., creation of a compound is not seen as dependent on being of help in at least a non-negligible number of trials). However, compound-cue theory is not well suited to account for the pronunciation data, because familiarity of the compound is only seen as the basis for binary decisions. Conversely, the episodic retrieval accounts can be easily adapted to the pronunciation data because the related prime is of help at least for high-dominant targets. The lexical decision data pose a more severe problem because it is not clear why primes should be retrospectively processed if they are not of help in any condition. However, it might be that due to the repeated exposure to targets, facilitation effects dissipate. That is, subjectively it might be of help to process the category prime in order to identify a high-dominant target. The RT measure, however, is no longer sensitive for the facilitation because of a floor effect in response speed. In conclusion, we have shown a remarkable new phenomenon in the field of semantic priming. By using the technique of repeatedly alternating a prime word with a random letter mask, we found a robust negatively signed masked (i.e. “subliminal”) priming effect for low dominance exemplars of semantic categories, which seems to reflect automatic processes. Although a variety of theories can in principle be applied to explain this effect, all of them are associated with ad hoc assumptions in doing so. Thus, our demonstration of subliminal semantic priming using a novel technique presents some challenges to accounts of semantic priming.

References Abrams, R. L., Klinger, M. R., & Greenwald, A. G. (2002). Subliminal words activate semantic categories (not automated motor responses). Psychonomic Bulletin and Review, 9, 100-106. Anderson, J. R., & Pirolli, P. L. (1984). Spread of activation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 791-798. Balota, D. A. (1983). Automatic semantic activation and episodic memory encoding. Journal of Verbal Learning and Verbal Behavior, 22, 88-104. Barnhardt, T. M., Glisky, E. L., Polster, M. R., & Elam, L. (1996).

11

Inhibition of associates and activation of synonyms in the rare-word paradigm: Further evidence for a center-surround mechanism. Memory and Cognition, 24, 60-69. Becker, C. A. (1979). Semantic context and word frequency effects in visual word recognition. Journal of Experimental Psychology: Human Perception and Performance, 5, 252-259. Becker, C. A., & Killion, T. H. (1977). Interaction of visual and cognitive effects in word recognition. Journal of Experimental Psychology: Human Perception and Performance, 3, 389-401. Bodner, G. E., & Masson, M. E. J. (1997). Masked repetition priming of words and nonwords: Evidence for a nonlexical basis for priming. Journal of Memory & Language, 37, 268-293. Bodner, G. E., & Masson, M. E. J. (2001). Prime validity affects masked repetition priming: Evidence for an episodic resource account of priming. Journal of Memory & Language, 45, 616-647. Bodner, G. E., & Masson, M. E. J. (2003). Beyond spreading activation: An influence of relatedness proportion on masked semantic priming. Psychonomic Bulletin & Review, 10, 645-652. Carr, T. H., & Dagenbach, D. (1990). Semantic priming and repetition priming from masked words: Evidence for a center-surround attentional mechanism in perceptual recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 341-350. Cheesman, J., & Merikle, P. M. (1985). Word recognition and consciousness. In D. Besner, T. G. Waller, & G. E. MacKinnon (Eds.), Reading research: Advances in theory and practice (Vol. 5, pp. 311-352). New York: Academic Press. Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82, 407-428. Dagenbach, D., Carr, T. H., & Barnhardt, T. M. (1990). Inhibitory semantic priming of lexical decisions due to failure to retrieve weakly activated codes. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 328-340. Dagenbach, D., Carr, T. H., & Wilhelmsen, A. (1989). Task-induced strategies and near-threshold priming: conscious influences on unconscious perception. Journal of Memory and Language, 28, 412-443. Damian, M. F. (2000). Semantic negative priming in picture categorization and naming. Cognition, 76, B45-B55. Dehaene, S., Naccache, L., Le Clec'H, G., Koechlin, E., Mueller, M., Dehaene-Lambertz, G., van de Moortele, P.-F., & Le Bihan, D. (1998). Imaging unconscious semantic priming. Nature, 395, 597600. Draine, S. C., & Greenwald, A. G. (1998). Replicable unconscious semantic priming. Journal of Experimental Psychology: General, 127, 286-303. Eimer, M., & Schlaghecken, F. (1998). Effects of masked stimuli on motor activation: Behavioral and electrophysiological evidence. Journal of Experimental Psychology: Human Perception and Performance, 24, 1737-1747. Erdfelder, E., Faul, F. & Buchner, A. (1996). GPOWER: A general power analysis program. Behavior Research Methods, Instruments, and Computers, 28, 1-11. Fowler, C. A., Wolford, G., Slade, R., & Tassinary, L. (1981). Lexical access with and without awareness. Journal of Experimental Psychology: General, 110, 341-362. Greenwald, A. G., Draine, S. C., & Abrams, R. L. (1996). Three cognitive markers of unconscious semantic activation. Science, 273, 1699-1702. Haussmann, R. E. (1992). Tachistoscopic presentation and millisecond timing on the IBM PC/XT/AT and PS/2: A Turbo Pascal unit to provide general-purpose routines for CGA, Hercules, EGA, and VGA monitors. Behavior Research Methods, Instruments and Computers, 24, 303-310 Holender, D. (1986). Semantic activation without conscious identification in dichotic listening, parafoveal vision, and visual masking: A survey and appraisal. Behavioral and Brain Sciences, 9, 1-66. Hughes, A. D., & Whittlesea, B, W. (2003). Long-term semantic transfer: An overlapping-operations account. Memory & Cognition,

12

WENTURA AND FRINGS

31, 401-411. Hutchison, K. A. (2003). Is semantic priming due to association strength or feature overlap? Psychonomic Bulletin and Review, 10, 785-813. Jonides, J. & Mack, R. (1984). On the cost and benefit of cost and benefit. Psychological Bulletin, 96, 29-44. Kahan, T. A. (2000). Negative priming from masked words: Retrospective prime clarification or center-surround inhibition? Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 1392-1410. Kihlstrom, J. F. (1987). The cognitive unconscious. Science, 237, 14451452. Klapp, S. T., & Hinkley, L. B. (2002). The negative compatibility effect: Unconscious inhibition influences reaction time and response selection. Journal of Experimental Psychology: General, 131, 255-269. Klauer, K. C., Roßnagel, C., & Musch, J. (1997). List context effects in evaluative priming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 246-255. Klinger, M. R., Burton, P. C., & Pitts, G. S. (2000). Mechanisms of unconscious priming: I. Response competition, not spreading activation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 441-455. Klinger, M. R., & Greenwald, A. G. (1995). Unconscious Priming of Association Judgments. Journal of Experimental Psychology: Learning, Memory and Cognition, 21, 369-581. Kouider, S., & Dupoux, E. (2001). A functional disconnection between spoken and visual word recognition: evidence from unconscious priming. Cognition, 82, B35-B49. Lucas, M. (2000). Semantic priming without association: A metaanalytic review. Psychonomic Bulletin and Review, 7, 618-630. Marcel, A. T. (1983b). Conscious and unconscious perception: An approach to the relations between phenomenal experience and perceptual process. Cognitive Psychology, 15, 238-300. Marcel, A. T. (1983a). Conscious and unconscious perception: Experiments on visual masking and word recognition. Cognitive Psychology, 15, 197-237. Masson, M. E. J. (1995). A distributed memory model of semantic priming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 3-23. Maxwell, S. E., Delaney, H. D. (1990). Designing experiments and analyzing data: A model comparison perspective. Belmont, CA: Wadsworth/Thomson Learning. McKoon, G., & Ratcliff, R. (1988). Contextually relevant aspects of meaning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 331-343. Naccache, L., & Dehaene, S. (2001). Unconscious semantic priming extends to novel unseen stimuli. Cognition, 80, 215-229. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limitedcapacity attention. Journal of Experimental Psychology: General, 106, 226-254. Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. Besner & G. W. Humphreys (Eds.), Basic processes in reading. Visual word recognition (pp. 264-336). Hillsdale, NJ: Erlbaum. Neely, J. H., Keefe, D. E., & Ross, K. L. (1989). Semantic priming in the lexical decision task: Roles of prospective prime-generated expectancies and retrospective semantic matching. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 1003-1019. Neely, J. H., Verwys, C. A., & Kahan, T. A. (1998). Reading "glasses" will prime "vision", but reading a pair of "glasses" will not. Memory and Cognition, 26, 34-39. Perea, M., & Gotor, A. (1997). Associative and semantic priming effects occur at very short stimulus-onset asynchronies in lexical decision and naming. Cognition, 62, 223-240. Perea, M., & Rosa, E. (2002). The effects of associative and semantic priming in the lexical decision task. Psychological Research, 66, 180-194. Pollack, I. (1970). A non-parametric procedure for evaluation of true

and false positives. Behavior Research Methods and Instrumentation, 2, 155-156. Pollack, I., & Norman, D. A. (1964). A non-parametric analysis of recognition experiments. Psychonomic Science, 1, 125-126. Rosch, E. (1978). Principles of categorization. In E. Rosch & B. B. Lloyd (Eds.), Cognition and categorization (pp. 27-48). Hillsdale, NJ: Erlbaum. Stolz, J. A., & Besner, D. (1997). Visual word recognition: Effort after meaning but not (necessarily) meaning after effort. Journal of Experimental Psychology: Human Perception and Performance, 23, 1314-1322. Stolz, J. A., & Neely, J. H. (1995). When target degradation does and does not enhance semantic context effects in word recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 595-611. Tukey, J. W. (1977). Exploratory data analysis (pp. 1-56). Reading, MA: Addison-Wesley. Vorberg, D., Mattler, U., Heinecke, A., Schmidt, T. & Schwarzbach, J. (2003). Different time courses for visual perception and action priming. Proceedings of the National Academy of Sciences, 100, 6275-6280. Wentura, D. (1999). Activation and inhibition of affective information: Evidence for negative priming in the evaluation task. Cognition and Emotion, 13, 65-91. Wentura, D. (2000). Dissociative affective and associative priming effects in the lexical decision task: Yes vs. no responses to word targets reveal evaluative judgment tendencies. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 456-469. Wentura, D., & Rothermund, K. (2003). The "meddling-in" of affective information: A general model of automatic evaluation effects. In J. Musch & K. C. Klauer (Eds.), The psychology of evaluation: Affective processes in cognition and emotion (pp. 51-86). Mahwah, NJ: Lawrence Erlbaum. Whittlesea, B. W., & Jacoby, L. L. (1990). Interaction of prime repetition with visual degradation: Is priming a retrieval phenomenon? Journal of Memory & Language, 29, 546-565.


13

Appendix A Categories (primes) and category exemplars (targets) of the Experiments Category

Category exemplars High dominant Low dominant

Insect (INSEKT)

Bee (BIENE) Fly (FLIEGE) Midge (MÜCKE)

Moth (MOTTE) Bed-bug (WANZE) Cricket (GRILLE)

Flower (BLUME)

Rose (ROSE) Tulip (TULPE) Carnation (NELKE)

Dahlia (DAHLIE) Crocus (KROKUS) Lily (LILIE)

Fruit (FRUCHT)

Apple (APFEL) Pear (BIRNE) Banana (BANANE)

Date (DATTEL) Mango (MANGO) Fig (FEIGE)

Bird (VOGEL)

Blackbird (AMSEL) Swan (SCHWAN) Thrush (DROSSEL) Pheasant (FASAN) Starling (STAR) Daw (DOHLE) Note. In Experiment 4, exemplars were used as primes instead of category names; additionally midge (MÜCKE) was replaced by wasp (WESPE) because of the diacritical points.

Appendix B Mean Response Times (in ms) of Non-Word Trials as a Function of Prime Condition, Category Dominance of Target Exemplars, and Awareness Status of Participants (Errors in % in Parentheses, Experiment 1a, 1b, 2, and 4) Low-d’a Cong. Incong. Neutral


688 (4.6) 685 (5.6)

693 (8.3) 692 (6.9)

691 (7.9) 686 (3.7)

669 (5.2) 646 (6.2)

659 (6.2) 656 (5.2)

667 (4.2) 655 (7.3)

712 (6.3) 733 (2.6)

735 (5.2) 724 (6.3)

726 (9.4) 717 (5.2)

770 (8.3) 741 (7.3)

758 (1.0) 716 (3.1)

748 (5.2) 739 (3.1)

581 (27.0) 581 (31.9)

580 (27.5) 595 (37.5)

578 (34.1) 588 (32.8)

570 (26.2) 567 (35.1)

558 (25.6) 589 (28.6)

571 (26.8) 580 (28.6)

735 (5.2) 714 (3.1)

728 (4.7) 719 (2.1)

717 (7.3) 713 (5.2)

658 (11.5) 644 (5.2)

656 (6.2) 650 (5.2)

673 (8.3) 649 (4.2)

Visible Prime Experiment 1 Low Dom.

-

-

-

High Dom.

-

-

-

683 (5.0) 654 (4.3)

671 (5.7) 665 (3.7)

697 (7.7) 680 (3.0)

Repeated Masked Prime Experiment 1 Low Dom. High Dom. Experiment 2 Low Dom. High Dom. Experiment 3 Low Dom. High Dom. Standard Masked Prime Experiment 2 Low Dom. High Dom.

a Sub-sample of participants with low d’ in the direct test (see text for further explanation). b Sub-sample of participants with high d’ in the direct test.

14

WENTURA AND FRINGS