Semantic processing of unconsciously presented pictures
Simon Reichwein Student number: 11343656
Abstract
A priming experiment using backwards masking was conducted to study if invisible presented pictures are processed on a semantic level. Target words of either animals or objects were preceded by an invisible prime picture of either an animal or an object. The experiment consisted of either congruent trials (target word and prime picture belonging to the same category) or incongruent trials (target word and prime picture belonging to a different category). During experiment 1 the prime was presented for 13ms and there was no priming effect obtained in the masked condition. The experiment was repeated in experiment 2 with a prime presentation time of 17ms. In experiment 2 there was again no priming effect found of the prime pictures on the word targets. Thus, both experiments failed to demonstrate
unconscious semantic processing. The results from this study support the view that unconscious processing is rather limited.
Semantic processing of unconsciously presented pictures
Studying the mechanisms underlying consciousness has proven to be a challenge. An interesting method is to investigate if unconsciously presented stimuli influence behavior. This kind of research can lead us to better understanding in which wayunconsciously presented stimuli influence behavior and decision making. It is debated to what extent
cognitive functions can take place unconsciously (Cleeremans, 2001). Whereas some authors claim that unconscious processes are able to perform the same functions as conscious
processes can (Hassin, 2013), others are more skeptical about the limits of unconscious processing (Hesselman & Moors, 2015).
Much research concerning unconscious processing is done utilizing masking, which is a widely used technique to render stimuli invisible. Here a stimulus is presented for a short amount of time and is preceded and followed by a mask, rendering the stimulus invisible to the perceiver. Studying how an invisible presented stimulus influences the processing speed of a successive visible stimulus can give insight in the way the brain processes invisible or unconscious stimuli. Simple feature-based priming has been demonstrated by Singhal, Breitmeyer and Garcia (2009). These researchers used metacontrast masking to demonstrate that participants were faster with identifying the color of a target when the color of the prime and target were congruent, a so called congruency effect. They used the same design to demonstrate the priming of shapes as well. Furthermore, Schmidt et al. (2010), demonstrated that the processing of brightness of light can take place without awareness. Though
processing of simple features outside of awareness has been demonstrated, there has been no true consensus on the extent of unconscious processing of more complex features such as the semantic processing of stimuli.
One of the first studies describing unconscious semantic processing was conducted by Deheane et al. (1998). They discovered that participants categorized target numbers faster when unconsciously presented words had similar semantic meaning. Semantic priming has been a subject of interest to study for years but to rule out non-semantic effects of
unconscious processing some requirements need to be met.
First, studies utilizing word-word priming tasks, have shown that showing only part of the target word as a prime was enough to elicit a congruency effect (Abrams, 2000) even if it was just one repeated letter from an earlier used target (Greenwald & Abrams, 2002). Another study conducted by Kouider & Dupoux (2004) discovered that a congruency effect also arose when primes consisted of pseudo words created from earlier used targets. By using, for example, both word primes and word targets it is likely that a congruency effect is mainly due to overlap of word parts and not because the meaning of the prime was processed.
Second, when primes are used as targets another non-semantic processing mechanism may be reflected (Damian, 2001). They described this as an automatic stimulus-response effect which entails that primes elicit an automatic motor response based on previous trials. More specific, if a consciously perceived target elicited a certain response, that target can elicit the same response when later presented as a prime.
Third, action triggers can be formed when there is a limited selection of primes or when participants create expectations based on instructions or expected stimuli (Kiesel, Kunde, & Hoffmann,2007). These researchers argue that participants automatically match external stimuli to response actions. When such an external stimulus, or prime in this case, is presented, participants automatically initiate the appropriate response without processing the actual meaning of the prime.
To prevent the formation of this stimulus-response effect or action triggers, the primes should come from categories with many items and primes shouldn’t be used as targets. To ensure that the primes are truly presented unconscious, an objective awareness check should be conducted. This should be a task that is almost identical to the main experiment except that instead of the target, the prime is the imperative stimulus (Schmidt, Haberkamp, Schmidt 2011).
Two studies were done which found congruity effects induced by subliminally presented primes (Dell’Acqua & Grainger,1999; Van den Bussche, Notebaert, & Reynvoet, 2009). Both studies accounted for all these non-semantic ways of unconscious processing by having primes and targets in different formats, from almost infinite categories (animals or objects) and an objective awareness check.
The study by Van den Busche et al. (2009) met all the requirements for genuine unconscious semantic processing. In this study they rendered prime pictures invisible using backward masking and instructed participants to categorize target words as fast and accurate as they could. The prime pictures were either animals or non-animals (objects) and the target words consisted of the Dutch words for the prime pictures. They found that participants categorized target words faster when preceded by picture primes of the same category in comparison to trials were the prime and target belonged to a different category. By presenting the primes as pictures and the targets as words, there was no visual overlap between primes and targets. Targets were not used as primes to prevent a stimulus-response effect from arising. The primes came from a large, almost infinite category, so no action-triggers are expected to be formed. On top of that Van den Busche et al. (2009) tested that the primes were truly presented unconscious utilizing an objective awareness check.
When Stein, Utz and Van Opstal (2018) performed the same research as Van den Busche et al. (2009) did, they failed to obtain congruity effects induced by unconsciously
presented primes. A Stimulus Onset Asynchrony (SOA) of 290ms was used, which is the time from the onset of the prime till the onset of the target. However, studies have shown that there is a rapid decline of priming effects when a SOA of longer than 100ms is used (Greenwald, Draine, and Abrams, 1996). In the study described here, the same design, stimuli and procedure will be used as Stein, Utz and Van Opstal did. However, a shorter SOA will be used, because other studies did find subliminal priming effects with a shorter SOA(100ms) (Pohl, Kiesel, Kunde & Hoffman, 2010). In the second experiment of this study a longer prime visibility time will be used which is expected to increase priming effects (Holcomb et al., 2005). In the experiments described here, it is expected that participants will respond faster with categorizing target words when these target words are preceded by a picture prime of the same semantic category in comparison to the trials in which target words are preceded by a picture prime of a different semantic category.
Method experiment 1 Participants
Participants signed up via a website (lab.uva.nl) and consisted mainly of students from the University of Amsterdam. They received either a small sum of money or student credit for their participation. A total of 77 participants were tested and 12 were excluded. Of the
participants that were excluded, 2 met the exclusion criteria for reaction time and 10 met the exclusion criteria for accuracy. The analyses was divided in a whole sample (N=65),
including people who scored above chance on the awareness check, and a reduced sample (N=60) of people who scored below chance on the awareness check. The mean age of the whole sample group was 24.3 years and consisted of 13 males and 52 females. For the
participants were unaware of the aim of the study, were native Dutch speakers with normal eyesight.
Exclusion criteria
Participants were excluded when their median reaction time was longer than 700ms or when they had an error rate of above 25%, in either the masked or the unmasked condition. 700ms and 25% were based on 5 SDs above the mean median reaction time and mean error rates for incongruent non-animal trials in the Van den Busche et al. (2009) study. The
analyses was divided in a ‘whole’ and a ‘reduced’ sample. The whole sample group included the participants who scored above chance on the awareness check, whereas in the reduced sample, these participants were excluded. However,it should be noted that post-hoc exclusion of participants can be problematic, because if you select a specific subgroup based on the awareness check, due to the effect of regression to the mean, the chance of false positive results increases (Shanks, 2016). Above-chance prime discrimination was tested (one-tailed). A binominal test was used to determine prime discrimination significance. The awareness check consisted of a 100 trials in which the chance level was 50%.Participants having a correct response value above 58% (p<.05) in the awareness check were excluded for the reduced sample analysis.
Design
There were three conditions used, the masked condition, an unmasked condition and an awareness check. In the masked condition a picture prime was presented unconsciously by backward masking. Every trial started with the presentation of a fixation cross for 400ms which was then followed by a 4x13ms noise pattern (figuring as mask). Then a picture prime was presented for 13ms, from the stimulus set used by Van den Busche et al.(2009), followed by a 27ms blank screen. Then another maskof 4x13ms noise pattern was presented. Lastly a target word was presented and participants were asked to indicate with the left and right arrow
keys whether the word belonged to the category ‘animal’ or the category ‘non-animal’. The trials were one second apart. The assignments of the arrow keys was randomized between subjects. In the unmasked condition the design was exactly the same except that the masks were removed. Both the masked and the unmasked condition consisted of 200 trials. During experiment 1 the order of the conditions was either ‘masked’ – ‘awareness check’-‘unmasked’ or ‘unmasked’ – ‘masked’- ‘awareness check’ and this was counterbalanced between subjects.
Materials and Stimuli
The experiment was shown on an 19-inch Iiyama Vision Master Pro 510 (A201HT) CRT monitor with a resolution of 1024x768 pixels. The monitor’s refresh rate was set to 75Hz so the picture primes could be presented for 13.3ms. The participants sat comfortably behind the computer in a dimly lit room. The Psychtoolbox functions (Brainard, 1997) in MATLAB were used to write the code for the experiment. The picture primes and target words were identical to the primes and targets used by Busche et al. (2009). The primes consisted of line drawings of 25 animals and 25 non-animals from the ‘‘Snodgrass and Vanderwart-like’’ objects (Busche et al. 2009; Rossion & Pourtois, 2004). The pictures of the drawings had a size of 144x144 pixels. The distance from which the participants viewed the stimuli, together with the stimuli size, corresponded to a visual angle of 5.2°. The target words were presented in a black Arial fond with capital letters and a size of 20 points on a white background. The targets were the Dutch words of the primes and were between three and six letters. There were four variable trials. Congruent trials with either an animal target word and an animal prime or a non-animal target preceded by a non-animal prime. In
addition, there were incongruent trials were animal target words were preceded by non-animal primes and vice versa. Target words and picture primes were never identical (Picture of a chicken followed by the word “chicken”).
Objective Awareness check
The objective awareness check was designed the almost same as the masked condition. However, during the awareness check participants were made aware of the fact that a prime was presented. During the awareness check participants were instructed to categorize the primes instead of the targets and instead of a word target, a string of X’s was presented. Participants were instructed that in this task, speed was not important, only accuracy, and they had to guess if necessary. The awareness check consisted of 100 trials and every prime picture was presented twice.
Pre-analyses processing
Trials in which participants responded incorrectly were excluded. After this was done, the median reaction times (RTs) were calculated for each participant for congruent and incongruent trials separately for both the masked as well as for the unmasked condition.
Results experiment 1 Prime discrimination
Prime discrimination for the whole sample (M=.50, SD=.05) did not differ from chance, t(64) = -.607, p = .273 (one-tailed). This means the primes were objectively invisible for the whole sample. Five participants scored above chance on the prime discrimination task. A separate analyses was also done without these participants, the reduced sample. For the Figure 1. Example of a congruent trial during experiment 1
reduced sample, prime discrimination (M=.49, SD=.03) was again not significantly different from chance, t(59) = -.645, p = .255 (one-tailed) so the primes were objectively invisible as well for the reduced sample.
Priming effects for the whole sample
To explore any effect of either the factor target category or congruency, an ANOVA was performed. Participants responded faster to target words of the animal category
(M=515ms, SD= 54.5ms) compared to the non-animal category (M=531ms, SD= 57.3ms), this was significant F(1,64) = 23.55, p < .001, ηp2 = .27. However, there was no significant
effect of congruency and there was also no interaction effect F<1.
Priming effects for the reduced sample
For the reduced sample another ANOVA with the factors “target category” and “congruency” was performed. Participants responded faster to word targets of the animal category (M=517ms, SD= 55.7) compared to word targets of the object category (M=533ms, SD= 58.5), this effect was again significant F(1,59) = 21.24, p < .001, ηp2 = .27. The factor
congruency, as well as an interaction effect were not significant F<1.
The same analysis was done for the unmasked condition. In the unmasked condition all factors were significant reflecting faster response times for animal targets and a significant effect of prime-target congruency (all p<.001). While unconscious presented primes elicited no congruency effects, visible primes convincingly did.
The accuracies for the reduced sample were also analyzed. First the accuracies were converted to z-scores and then a repeated measures ANOVA was performed. A significant effect of target category was found, F(1,59) = 26.69, p < .05, ηp2 = .14 reflecting higher
accuracies for object targets (M=.97%, SD=.03%) compared to animal targets (M=.95%, SD=.05%). There was no significant effect of prime-target congruency on the accuracies and there was also no interaction effect (both F<1).
Figure 2. Results for the reduced sample from experiment 1 with a prime presentation time of 13ms. The results of the masked condition is plotted on the left, with separate columns for animal targets and object targets. On the right are the results shown from the condition with the unmasked prime.
Experiment 2: Replication of experiment 1 using longer prime presentation time
During experiment 1 no priming effects were found in the masked condition. It is possible that with the design that was used the stimulus strength of the prime was too low to elicit a priming effect. Therefore the experiment was repeated with some slight adjustments. Experiment 2 was the same as experiment 1, except that the stimulus strength of the prime was increased by presenting it for 17ms this time. Because of the set-up each mask was also presented for 4x17ms.
Participants
In the second experiment, 49 participants were tested and the analyses was again divided in a whole group and a reduced group (as in experiment 1). The exclusion criteria for experiment 2 were the same as for experiment 1. Two participants were excluded, because they met the exclusion criteria for accuracy and two others were excluded, because they met the exclusion criteria for reaction time. The average age of the whole group (N=45) was 19.1 years and consisted of 17 males and 28 females. For the reduced sample (N=30), the mean
age was 19.3 and consisted of 12 males and 18 females. All the participants were recruited from the University of Amsterdam and had to take part to fulfill part of a course.
Design, Materials and Stimuli
The experiment design and materials of experiment 2 were almost identical to experiment 1 except that during the second experiment a different monitor was used, which was a 24-inch BenQ Lcd monitor (XL2420-T).The corresponding visual angle with which the participants viewed the stimuli was 3.63º. The monitor’s refresh rate was set to 60Hz so
the prime pictures, as well as the noise pictures, could be presented for 17ms. The order of the conditions was ‘masked’ – ‘awareness check’ – ‘unmasked’. The stimuli used during
experiment 2 were exactly identical to the stimuli used in experiment 1.
Results
Priming discrimination
Prime discrimination for the whole sample (M=.55, SD=.07) did differ from chance, t(44) = 5.027, p < .001(one-tailed), Cohen’s dz = .75 This means the primes were not
objectively invisible for the whole sample. After removing participants which scored above chance on the prime discrimination task, a further analyses was done with the reduced sample. Prime discrimination for the reduced sample (M=.51, SD=.04) did not differ from chance t(29) = .141, p = .07 (one-tailed).
Priming effects for the whole sample
An ANOVA was performed for the whole sample with the factors target category and prime-target congruency. There was a significant effect of target category with participants responding faster to animal targets (M=546ms, SD=52ms) compared to object targets (M=578ms, SD=60ms), F(1,44) = 57.13, p < .001, ηp2 = .565. The factor congruency was,
however, not significant F(1,44) = 3.034, p = .084, ηp2 = .065 and there was also no interaction
Priming effects for the reduced sample
The same analysis was done for the reduced sample. This analysis showed a
significant effect for target category F(1,29) = 31.28, p < .001, ηp2 = .519 with faster responses
to animal targets (M=551ms ,SD=51.7ms) compared to non-animal targets (M=580ms ,SD=64.4ms). No effect was found for the factor congruency F(1,29) = 3.163, p = .086, ηp2 = .
098. There was also no interaction effect found between the two factors F(1,29) = 3.084, p = . 090, ηp2 = 3.084. Analyses of the results of the unmasked condition for the reduced sample
showed significant effects of both target category F(1,29) = 8.615, p = .006, ηp2 = .229 as well
as for prime-target congruency F(1,29) = 79.132, p < .001, ηp2 = .732.
The accuracies for the reduced sample were also analyzed. The accuracies were again first converted to z-scores to normalize the data and then a repeated measures ANOVA was performed. No significant effect of either target category or prime-target congruency on the accuracies was found (both p>.05).
The results of experiment 2 are comparable to the results of experiment 1. Whereas unconscious presented primes do not elicit a congruency effect, consciously presented primes do.
Figure 3. Results for the reduced sample from experiment 2 with a prime presentation time of 17ms. The results of the masked condition is plotted on the left, with separate columns for animal targets and object targets. On the right are the results shown from the condition with the unmasked prime.
General Discussion
The experiments described here tested for unconscious semantic priming utilizing backwards masking. The study by Van den Busche et al. (2009) was replicated, except the study described here used a shorter SOA, because it met all the requirements for genuine unconscious semantic priming. In experiment 1, the same prime duration was used as Van den Busche et al. (2009) did. However, no semantic priming effect was found in the masked condition. In the condition with visible primes on the other hand, a strong congruency effect was found. Though both Dell’Acqua & Grainger (1999) and Van den Bussche et al. (2009) found a priming effect of picture primes on word targets, this study failed to do so. This study does not invalidate the results of the previous studies. However, considering enough statistical power (N=65), it can be concluded that the congruency effect found by the two previous mentioned studies is probably weaker than expected, if existent at all.
In the second experiment a longer prime duration, and a similar SOA, was used to discover if this would yield any priming effects. The results of experiment 2 were mostly comparable with the results of experiment 1. Only in the trials with animal targets, a
significant difference was found (p<.05) between the congruent trials and incongruent trials in the masked condition, based on comparison of the means with a t-test. However, there was no congruency effect based on the ANOVA analysis and also no interaction effect (both p>.05). Therefore this difference was probably an artefact.
In both experiment the only factor leading to faster response times was the factor target category. Participants responded faster in trials with animal targets compared to object targets. This could be, because the task instructions stated that the participants had to
categorize the target to belong to the animal category or the non-animal category. This way it is probable that participants had a better representation of the category animals and were therefore faster in responding in the trials with animal targets. In experiment 2, just as in
experiment 1, no effect of prime-target congruency was found on either the reaction times or the accuracies. Therefore, unconscious semantic priming was again not demonstrated.
In both experiment 1 and in experiment 2 there was no congruency effect found which means that a false positive effect, as was mentioned in the exclusion criteria of experiment 1, was absent. However, post-hoc exclusion of participants can be problematic, especially if a significant number is excluded, (Shanks, 2016) like in experiment 2. This is, because if you select the participants that scored at or below chance on the awareness check, this
automatically leads to regression to the mean of the other variable that you are testing, in this case, reaction time. It would be best to prevent post-hoc exclusion of participants all together. For example, prime visibility can be determined for each participant separately so they are just unable to categorize the primes. This can be done by manipulating the prime presentation time. However, this is quite time consuming. Another way to prevent post-hoc exclusion is to ensure that the prime is invisible for the whole sample so there is no need to exclude
participants that were able to categorize the primes. However, there is a fine line between a prime being invisible enough not to be seen and the prime being too invisible to have enough stimulus strength to elicit a priming effect.
Studies have shown that a congruency effect was dependent on semantic relatedness of the prime and target (Van den Bussche, Smets, Sasanguie & Reynvoet, 2012; Ortells, Marí-Beffa, & Plaza-Ayllón, 2013). These studies have shown reliable unconscious semantic priming, but only for strongly related prime-target pairs. They found priming effects for strongly related pairs like CAT-DOG but not for FLY-DOG. The design of this study made it impossible to do an analysis on the measure of semantic relatedness, but future studies should take this into account.
In conclusion, this study failed to demonstrate unconscious semantic processing of invisible pictures and therefore contributes to the view of Hesselman & Moors (2015), that unconscious processing has its limits.
Literature
Abrams, R. L., & Greenwald, A. G. (2000). Parts outweigh the whole (word) in unconscious analysis of meaning. Psychological Science, 11, 118–124.
Almeida, J., Mahon, B. Z., Nakayama, K., & Caramazza, A. (2008). Unconscious processing dissociates along categorical lines. Proceedings of the National Academy of
Sciences, 105(39), 15214–15218. doi:10.1073/pnas.0805867105
Brainard, D. H. (1997). The psychophysics toolbox. Spatial Vision, 10, 433-436.
Cleeremans, A. (2001). Conscious and Unconscious Processes in Cognition. International Encyclopedia of the Social & Behavioral Sciences, 2584–2589. doi:10.1016/b0-08-043076-7/03560-9
Damian, M. F. (2001). Congruity effects evoked by subliminally presented primes:
Automaticity rather than semantic processing. Journal of Experimental Psychology: Human Perception and Performance, 27, 154–165.
Dehaene, S., Naccache, L., Le Clec’H, G., Koechlin, E., Mueller, M., Dehaene-Lambertz, G., Van de Mortele, PF., Le Bihan, D. (1998). Imaging unconscious semantic priming. Nature, 395(6702), 597–600. doi:10.1038/26967
Dell’Acqua, R., & Grainger, J.(1999). Unconscious semantic priming from pictures. Cognition, 73, 1-15
Greenwald, A. G., Draine, S. C., & Abrams, R. L. (1996). Three Cognitive Markers of Unconscious Semantic Activation. Science, 273(5282), 1699–1702.
doi:10.1126/science.273.5282.1699
Greenwald, A., Abrams, R., (2002) Visual masking reveals two qualitatively different levels of unconscious cognition. Paper presented at the 43rd Annual Meeting of the
Hassin, R. R. (2013). Yes it can: On the functional abilities of the human unconscious. Perspectives on Psychological Science, 8, 195-207
Hesselmann, G., & Moors, P. (2015). Definitely maybe: Can unconscious processes
perform the same functions as conscious processes? Frontiers in Psychology, 6, 584 Holcomb, P. J., Reder, L., Misra, M., & Grainger, J. (2005). The effects of prime visibility on
ERP measures of masked priming. Cognitive Brain Research, 24, 155–172.
Kiesel, A., Kunde, W., & Hoffmann, J. (2007). Mechanisms of subliminal response priming. Advances in Cognitive Psychology, 3(1), 307–315. doi:10.2478/v10053-008-0032-1 Kouider, S., & Dupoux, E. (2004). Partial Awareness Creates the “Illusion” of Subliminal
Semantic Priming. Psychological Science, 15(2), 75–81. doi:10.1111/j.0963-7214.2004.01502001.x
Ortells, J. J., Marí-Beffa, P., & Plaza-Ayllón, V. (2013). Unconscious congruency priming from unpracticed words is modulated by prime-target semantic relatedness. Journal of Experimental Psychology: Learning, Memory and Cognition, 39, 394-413.
Pohl, C., Kiesel, A., Kunde, W., & Hoffmann, J. (2010). Early and late selection in
unconscious information processing. Journal of Experimental Psychology: Human Perception and Performance, 36, 268-285.
Ro, T., Singhal, N. S., Breitmeyer, B. G., & Garcia, J. O. (2009). Unconscious processing of color and form in metacontrast masking. Perception & Psychophysics, 71(1), 95–103. doi:10.3758/app.71.1.95
Rossion, B., & Pourtois, G. (2004). Revisiting Snodgrass and Vanderwart’s object set: The role of surface detail in basic level object recognition. Perception, 33, 217–236. Schmidt F, Haberkamp A, Schmidt T. (2011). Dos and don'ts in response priming research.
Shanks, D. R. (2017). Regressive research: The pitfalls of post hoc data selection in the study of unconscious mental processes. Psychonomic Bulletin & Review, 24, 752–775. Ro, T., Singhal, N. S., Breitmeyer, B. G., & Garcia, J. O. (2009). Unconscious processing of
color and form in metacontrast masking. Perception & Psychophysics, 71(1), 95–103. doi:10.3758/app.71.1.95
Van den Bussche, E., Notebaert, K., & Reynvoet, B. (2009). Masked primes can be
genuinely semantically processed: A picture prime study. Experimental Psychology, 56, 295–300.
Van den Bussche, E., Smets, K., Sasanguie, D. and Reynvoet, B., 2012. The power of unconscious semantic processing: The effect of semantic relatedness between prime and target on subliminal priming. Psychologica Belgica, 52(1), pp.59–70. DOI:
