Unconscious processing dissociated: No evidence for word inversion
e
ffects on visual localization
Ron van de Klundert
University of AmsterdamRecent cognitive neuroscience studies seem to suggest that a great extent of human information processing can occur without awareness. Previous breaking continuous flash suppression (b-CFS) studies suggest unconscious processing of word orientation and an advantage for upright words compared to inverted words (i.e. word inversion effect), since they break interocular suppression more quickly. Unfortunately, this commonly adopted paradigm in unconscious processing research does not dissociate unconscious from conscious processes in an unam-biguous manner. This study re-investigates the word inversion effect while adopting the novel localization-discrimination dissociation paradigm. This novel paradigm incorporates a dis-crimination task to accurately delineate unconscious and conscious processes and their effect on word localization performance. With the use of this paradigm this study failed to replicate commonly reported word inversion effects and thus gives reason to reconsider the unconscious origin of this effect.
Keywords: SDT, Unconscious processing, b-CFS, visual perception, word inversion effect
Introduction
What is the function of conscious awareness? At first thought conscious awareness seems indispensable. Almost every act of volition, perception and cognition seems to be paired with conscious awareness, giving us the intuition that these complex processes are the functions of conscious-ness. However, these are just intuitive claims. Conscious inessentialism argues that in principle any intelligent activity in any cognitive domain could be performed without it be-ing accompanied by conscious awareness (Levine & Flana-gan, 1994). This contrasting view sees conscious awareness unnecessary on a metaphysical level and deems it nothing more than an epiphenomena (see Kirk, 2006). Recent cog-nitive neuroscience studies seem to provide empirical evi-dence for conscious inessentialism by suggesting that many processes do not require, but are inevitably paired with con-scious awareness. However, on closer inspection these stud-ies might not implement the tools necessary to accurately in-fer the true extent of conscious processing and its function-ality. So then how do these studies go about unveiling the functionality of conscious awareness?
A particular promising and long adopted approach is con-trasting that what is possible unconsciously with that what requires consciousness. This approach was coined the "con-trastive analysis" by Baars (1988). A crucial foundation of this approach is to carefully investigate which functions or phenomena can occur without conscious awareness (i.e. through unconscious processing), ultimately mapping out the functionality of conscious awareness, if there is any, through means of exclusion. Consequently, an extensive body of re-search has been dedicated to characterizing the extent of un-conscious perceptual and cognitive processes (Hassin, 2013; Hesselmann & Moors, 2015; Kouider & Dehaene, 2007;
Kouider & Faivre, 2017; Schmidt, 2015; Stein, Awad, Gayet, & Peelen, 2018; Stein, Hebart, & Sterzer, 2011; Van Gaal & Lamme, 2012).
One of the many fathomed topics in unconscious process-ing is those of visual words. Visual processprocess-ing of words is thought to belong to specialized processing regimes as-sociated with high efficiency (McCandliss, Cohen, & De-haene, 2003; Weiner & Zilles, 2016) and brain regions with an expertise in word processing (Cohen et al., 2002). This seems in line with the apparent effortlessness to how words are quickly identified. This might suggest that a great aspect of word processing is managed unconsciously. Indeed, neu-roimaging studies show unconscious word repetition effects in the visual word form area (Dehaene et al., 2001) and even unconscious semantic category specific patterns of activation throughout the brain (Sheikh, Carreiras, & Soto, 2019).
A hugely popular paradigm used to dissect the function-ality of unconscious word processing is breaking continuous flash suppression (b-CFS). This paradigm renders a stimu-lus completely invisible for up to several seconds through strong interocular suppression, nevertheless the stimulus will eventually become visible (or "break through"). A com-mon aim of this paradigm is to contrast the time required to break through of two different stimuli or stimuli dimen-sions to infer differences in unconscious processing. In sim-ilar fashion to the aforementioned studies, a multitude of b-CFS studies suggest that a great aspect of word processing operates at the unconscious level. Pronounced unconscious effects have been reported for familiarity with the writing system (Jiang, Costello, & He, 2007; Rabagliati, Robertson, & Carmel, 2018) to higher level effects of emotional valence of the word (Prioli & Kahan, 2015; Yang & Yeh, 2011). One study even provides evidence for unconscious
process-ing of semantic relationships (Sklar et al., 2012). However, the most established effect is that of word inversion. Sev-eral studies suggest unconscious processing of word orien-tation and an advantage for upright words compared to in-verted words (i.e. rotated by 180 degrees), since upright words break CFS faster than inverted words (Kerr, Hessel-mann, Räling, Wartenburger, & Sterzer, 2017; Yang, Tien, Yang, & Yeh, 2017; Yang & Yeh, 2011, 2014). Since word inversion results in essentially the same stimuli with regards to low-level image properties, it is a cogent and controlled way to study the unconscious effects of meaning, spatial re-lations and familiarity with the orientation.
Unfortunately, the b-CFS paradigm does not implement a conclusive awareness control. Therefore, this paradigm es-sentially fails to distinguish whether differences in reaction times are due to unconscious processes or due to processes that occur after a stimulus reaches awareness, such as de-cision making processes. A proposed solution to rule out post-perceptual processes is adopting non-speeded accuracy-based measures such as localization or detection sensitiv-ity index d’ from signal detection theory, instead of reac-tion time measures. These measures are essentially response criteria-free and thus should rule out post-perceptual pro-cesses. However, even criterion-free sensitivity measures are possibly affected by later perceptual processes such as conscious recognition of the critical stimuli (Hochstein & Ahissar, 2002). If conscious recognition of stimuli affects localization or detection performance and conscious identifi-cation differs between experimental conditions, the paradigm is prone to overestimate detection or localization differences between the conditions (Stein & Peelen, 2020). Additionally, implementing non-speeded accuracy-based b-CFS is a time-consuming and tedious process due to high variability in the desired task parameters between individuals (Gayet & Stein, 2017; Stein et al., 2011).
For studies to effectively rule out conscious processing it is necessary to implement a paradigm that collects an addi-tional awareness control measure. A paradigm that adopts this logic is the classical dissociation paradigm (Erdelyi, 1986). This paradigm, much like the b-CFS paradigm, col-lects a measure that should capture the processing of stimuli and in addition it collects a subjective or objective measure of awareness of these stimuli. If the measure of awareness indi-cates an absence of awareness and processing of the critical stimulus is still present, then this paradigm intuitively seems to show unconscious processing. However, classical disso-ciation studies are problematic in their measures of aware-ness. Measuring subjective awareness in classical dissocia-tion studies is mostly done with scales that have a range of awareness categories (e.g. the Perceptual Awareness Scale). Subjects need to categorize their subjective experience ac-cording to those scales. One serious issue with this catego-rization is that subjects might be inherently conservative and
therefore categorize a slightly liminal experience as having no experience whatsoever. This is also in line with signal detection theory, where "no subjective experience" is equal to an experience that fails to reach a response criterion, but is not necessarily completely unconscious (Schmidt, 2015). An issue like this will most likely result in an overestimation of the extent of unconscious processing. On the other hand, measuring awareness in an objective manner is performed by rendering a stimulus undetectable, completely ruling out the possibility of conscious perception seen in subjective aware-ness measures. However, this does not rule out the possibility of the objective awareness measure (e.g. detectability) being determined by unconscious processes. In this case, objective unawareness of a critical stimulus might result in diminished or no unconscious processing of the critical stimulus, leading to an underestimation of the extent of unconscious process-ing. Additionally, studies that employ the objective approach rarely compare their measure of awareness to their measure of stimulus processing in the same units (Schmidt & Vorberg, 2006). A study by Franz and von Luxburg (2015) emphasizes the importance of comparing both measures in the same units for a meaningful comparison.
This portrays the issues with dissociating conscious and unconscious processes in commonly adopted paradigms designed to test for unconscious visual processing. In this study, we therefore adopt a novel paradigm that clearly dissociates conscious and unconscious processing; the localization-discrimination dissociation paradigm. The logic behind this novel paradigm is straightforward. Indi-viduals perform a non-speeded accuracy-based localization task as well as a discrimination task. For the localization task, individuals have to simply localize a stimulus, while for the discrimination task they have to assess a critical di-mension of said stimulus. If individuals are unable to dis-criminate between the critical stimulus dimension, then all differences in localization performance caused by this critical stimulus dimension must be ascribable to differential uncon-scious processing. In classical dissociation paradigm logic, the discrimination task provides us with an objective measure of awareness, while the localization task provides a measure of processing. This task also allows for longer stimulus pre-sentation times than those in classical dissociation paradigms adopting the objective awareness measure, thus allowing in-dividuals to receive more stimulus energy. This makes it less likely to underestimate the extent of unconscious processing compared to previous designs. Additionally, both localiza-tion and discriminalocaliza-tion responses can be transformed to their respective sensitivity index d’, allowing direct comparison of the awareness and processing measures in the same unit and criterion-free assessment. Unpublished results from Stein and Peelen (2020) validated this novel approach and repli-cated one of the most entrenched unconscious functionali-ties: the face inversion effect. This study shows that upright
faces are localized better than inverted faces, all while partic-ipants were unable to dissociate between the critical stimuli dimensions. The results have further proven that visual back-ward masking is an effective method for rendering critical stimuli dimensions invisible, while concurrently presenting the stimuli above the localization threshold, given that ade-quate presentation time parameters have been found.
Altogether, this gives a striking incentive to revisit other effects that are deemed to function at an unconscious level by older paradigms and see if they still replicate by adopting this novel paradigm.
The primary aim of this study is unveiling whether famil-iarity with lexical stimuli truly influences unconscious pro-cessing. Specifically, this study investigates whether word inversion effects are present when awareness of the stimuli is absent. We hypothesize that upright words have an advantage over inverted words in unconscious processing, due to famil-iarity with the stimuli. The novel localization-discrimination dissociation paradigm will be conducted to test this hypoth-esis. There are three requirements to demonstrate uncon-scious word inversion effects. Firstly, discrimination perfor-mance has to be at chance. Secondly, there should be di ffer-ences between localization performance for the upright and inverted words. Lastly, the difference in localization perfor-mance should be larger than the discrimination perforperfor-mance. In order to meet the first requirement, it is important to find suitable presentation time parameters. For this reason, a to-tal of four different presentation times will be implemented with a visual backward mask procedure with the hopes of one yielding zero discriminability, but above chance localization. For exploratory purpose, word length, frequency and valence were controlled to examine whether these factors would have a play in any potential effects. It is expected that localiza-tion and discriminalocaliza-tion performance increase as presentalocaliza-tion times increase, since more stimulus energy will be available to perform the task. Since a multitude of studies have demon-strated that upright words reach awareness more quickly than inverted words (Kerr et al., 2017; Yang et al., 2017; Yang & Yeh, 2011, 2014), suggesting an unconscious processing ad-vantage for upright words, it is expected that upright words are localized better than inverted words in general. More im-portantly, it is expected that upright words are localized bet-ter than inverted words, even when participants are unable to discriminate between this critical stimulus dimension. How-ever, given the issues of older paradigms with accurately dis-sociating unconscious processing, it may as well be possible that previously reported word inversion effects prove to be absent when awareness is absent.
Methods& Materials Subjects.
A total of 34 subjects (5 males; 31 right-handed; mean age of 19.61, SD= 3.12) participated in this experiment. All
subjects had either normal or corrected-to-normal vision and had given a written informed consent before starting the ex-periment. All subjects were compensated with research cred-its. This study was approved by the ethical committee of the University of Amsterdam.
Stimuli.
For this study, a total of 40 Dutch words (appendix A), all monosyllabic or disyllabic nouns, were selected from the SUBTLEX-NL database (Keuleers, Brysbaert, & New, 2010). SUBTLEX-NL is a library of Dutch word frequencies based subtitles. Words were selected to fit into two of four categories; based on frequency a word was either frequent or infrequent and based on valence a word was either negative or neutral. They were selected in such a manner that each category would contain 10 words each. Additionally, words were chosen to differ in length to some extent within cate-gories (median length= 5.5 letters), but not differ between categories. Negative words were words that directly refer to or are associated with negative emotions, disease, negative events, swearing or death. Neutral words were words that do not inherently refer to negative emotions, disease, negative events, swearing or death. Words were handpicked to either be frequent (M = 10249.95 per million) or infrequent (M = 856.15 per million). Importantly, words were carefully se-lected so that the word frequency category was the only fac-tor significantly affecting word frequency values (p < .001), while not being significantly affected by word valence cat-egory (p = 0.997) or an interaction between the two (p = 0.999). Words were presented with the font ‘Arial’ with font size 13 in dark gray (RGB: 78 78 78) in either an upright or inverted configuration (i.e. rotated by 180 degrees) on a lighter gray square (RGB: 128 128 128) of 240 by 240 pix-els. Words were presented either 35 pixels above or below the center of fixation. Visual backward masks consisted of random matrices of 2 pixel rectangles in different gray lev-els and were generated individually for each participant and trial.
Experimental paradigm.
A total of 34 participants completed the localization-discrimination dissociation task. The task was presented with use of MATLAB version 9.5. Natrick, Massachusetts: The Mathworks Inc., 2018 using the Psychophysics Toolbox ex-tension (Brainard, 1997; Pelli, 1997; Kleiner et al, 2007). The experiment was shown on a 24 inch (16:9) ASUS TFT-LCD monitor (1920×1080 pixel, 120 Hz) and with a refresh rate of 120 Hz. An experimental session consisted of a total of 960 trials. 40 unique words were presented in two orien-tations, at two locations and for four different presentation times. Due to a coding error half of the words were shown twice as often as intended. This, however, has not affected the balancing of the factors orientation, presentation time,
va-Figure 1. Localization-discrimination dissociation paradigm. A trial started by presenting a fixation cross in the middle of the screen for 1s, after which a blank screen was shown for 0.5s to indicate the start of stimulus presentation. Then one of the 40 words was randomly selected and presented as either an upright or inverted and either above or below the center of fixation. Stimulus presentation time variable and lasted for either 16.7, 25, 33.3 or 41.7ms. Stimulus presentation was directly followed by three backward masks presented for 100ms each. After this, participants had to indicate whether a word was upright or inverted for the discrimination task and whether the word was presented above or below fixation for the localization task. The order of these task was counter-balanced between participants. The sensitivity index d’ was derived from both tasks.
lence and frequency. The experiment was split into six blocks of 160 trials with forced breaks lasting at least 10s between the blocks. A trial started by presenting a fixation cross in the middle of the screen for 1s, after which a blank screen was shown for 0.5s to indicate the start of stimulus presentation. Then one of the 40 words was randomly selected and pre-sented as either an upright or inverted and either above or be-low the center of fixation. Stimulus presentation time lasted for either 16.7, 25, 33.3 or 41.7ms. These presentation times correspond to 2, 3, 4 and 5 synchronized frames of the 120 Hz screen. Stimulus presentation was directly followed by three backward masks presented for 100ms each. After this, participants had to indicate whether a word was upright or in-verted for the discrimination task and whether the word was presented above or below fixation for the localization task by pressing corresponding keys. The localization and discrimi-nation tasks were non-speeded and the order in which these tasks had to be completed was switched between each partic-ipant to control for order effects (figure 1). Additionally, at the start of each session, each participant had to complete 10 practice trials to gain familiarity with the task and the stimuli signal. The practice trials were followed by training blocks of 10 trials each and lasted until the participant achieved 80% correct in at least one of the training blocks.
Statistical analysis behavior.
Analysis was conducted with use of JASP version 0.12.2 (JASP Team, 2020). For each participant, both localization and discrimination responses for each presentation time were transformed to the sensitivity index d’ from signal-detection theory. Localization hits were chosen to be "up" responses in trials where the words were presented above fixation, while localization false alarms were "up" responses where the words were presented below fixation. Discrimination hits
were chosen to be "upright" responses in upright word trials, while discrimination false alarms were "upright" responses in inverted word trials. Hit or false alarm rates that are equal to 0 or 1 were recalculated to equal 1/(2n) and 1-1/(2n) respectively, with n referring to the total amount of trials used to calculate the initial value (Macmillan & Creelman, 2005). Discrimination and localization d’ were then calcu-lated by subtracting the respective z-transformed false alarm rate from the z-transformed hit rate. To account for the fact that the localization is a two-alternative force-choice (2AFC) task, which is considered less difficult than a yes-no signal detection paradigm, the localization d’ was divided by the square root of two (Macmillan & Creelman, 2005). To assess main and interaction effects of presentation time and word orientation on localization sensitivity, a two-way repeated measures ANOVA was conducted. Additionally, to assess the effect of presentation time on discrimination sensitivity, a separate repeated measures ANOVA was conducted. One of the aims of this study is to test whether a critical presen-tation time exists that results in zero discriminability, while concurrently resulting in above zero localization sensitivity. This is critical to ultimately test for unconscious inversion ef-fects (i.e. differences in localization d’ between upright and inverted words, while at zero discriminability). To achieve this, discrimination and localization d’ were tested for devia-tions higher than zero via one-tailed one-sided t-tests for each presentation time separately. This was done in order to iden-tify the longest presentation time that yields zero discrim-inability, but still yields above zero localization. Identifying the longest presentation time that satisfies these criteria sures presenting participants with the strongest stimulus en-ergy. Finally, to test for unconscious inversion effects a post-hoc comparison was conducted between localization d’ for
Figure 2. Results main analysis. The mean sensitivity index d’ for the discrimination and localization task (n= 34) for each of the presentation times. Localization is plotted separately for both word orientations (i.e. upright & inverted). Data shows an overall main effect of presentation time, but no main effect of orientation. These were not qualified by an interaction effect of presentation time and word orientation. Error bars represent 95% confidence interval.
upright and inverted words, specifically for this aforemen-tioned critical presentation time. To satisfy multiple statisti-cal philosophies but, most importantly, provide evidence for the null hypothesis, all tests were conducted with inferential statistics and their Bayesian counterpart where possible. All inferential statistics were conducted with α= 0.05 and using the Holm-Bonferroni method to correct for multiple testing (Holm, 1979). Greenhouse-Geisser estimates of sphericity were applied to correct degrees of freedom if data departed from sphericity (Abdi, 2010). Bayesian analysis was per-formed with the default priors in JASP. Bayes Factors for main and interaction effects were calculated by comparing models that include the effect to all matched models without the effect, as suggested by Sebastiaan Mathôt (2017).
Exploratory analyses.
To further extend this research paper, the main analysis was complemented by three exploratory analysis. With these exploratory analyses, the effects of word length, word fre-quency and word valence on localization performance were tested. Each exploratory analysis was conducted separately and with the use of the same data as the main analysis. Furthermore, percentage correct was used for localization performance instead of sensitivity d’, since further direct comparisons with discrimination performance was unneces-sary for the exploratory analyses. For the word length, fre-quency and valence analyses a three-factor repeated mea-sures ANOVA was conducted separately to test main and in-teraction effects with word orientation and presentation time on localization performance. For each of the analyses a
post-hoc comparison of the exploratory factor was conducted at the critical presentation time that yielded zero discriminabil-ity, but above zero localization sensitivity.
Results
Main analysis.
To assess main and interaction effects of presentation time and word orientation on localization sensitivity a two-way repeated measures ANOVA was conducted. Additionally, to assess the effect of presentation time on discrimination sensitivity a separate repeated measures ANOVA was con-ducted. The data showed a main effect of presentation time, indicating an increase in localization sensitivity as presenta-tion time increased [F(3, 99) = 229.82, p < .001, η2p = 0.874,
BFincl = 9.060e+95]. The data further indicated that there
was no main effect of word orientation on localization sensi-tivity [F(1, 33)= 0.838, p = 0.367, η2p= 0.025 BFincl= 0.143].
These findings were not qualified by a significant interaction effect of presentation time and word orientation on localiza-tion sensitivity [F(3, 99)= 1.903, p = 0.134, η2p= 0.055, BFincl
= 0.063]. Furthermore the data also showed that discrimina-tion sensitivity increased as presentadiscrimina-tion increased [F(3, 99)=
70.672, p< .001, η2
p= 0.682, BFincl= 1.257e+23]. All of
these results are shown in figure 2.
Although no main effect of orientation or interaction ef-fect of orientation and presentation time was present, a post-hoc analysis was necessary to converge towards the main question of this study. This post-hoc analysis focused on the longest presentation time that resulted in zero discrim-inability while concurrently resulting in above-chance local-ization. This was done to ensure a maximum bottom-up sig-nal strength of the stimuli when dissociating localization and discrimination. To find this respective presentation time both discrimination and localization sensitivity were tested for a deviation higher than zero sensitivity for each of the pre-sentation times (one-tailed tests). The longest prepre-sentation time that yielded zero discriminability was the second short-est of 25ms (M= 0.047, SD = 0.209, p = 0.902, d = -0.226, BF0+ = 11.639). While at zero discriminability of stimulus orientation, localization sensitivity was significantly higher than zero for both upright and inverted words (Upright: M= 0.609, SD= 0.442, p < .001, d = 1.379, BF+0 = 8.582e+6; inverted: M= 0.578, SD = 0.385, p < .001, d = 1.502, BF+0 = 5.371e+7). These results indicate that the second shortest presentation time is the longest presentation time that results in zero discriminability, while also resulting in above-chance localization. To test the main hypothesis of word inversion effects at zero discriminability a post-hoc comparison of lo-calization sensitivity between upright and inverted words was conducted for the second shortest presentation time specifi-cally. There was no indication of difference in localization sensitivity between upright (M= 0.609, SD = 0.442) and in-verted words (M= 0.578, SD = 0.385, p = 1.00, d = -0.115)
at the second shortest presentation time (figure 3).
Figure 3. Post-hoc comparison. Post hoc-comparison of localiza-tion d’ for upright and inverted words for the critical presentalocaliza-tion time of 25ms. Error bars represent 95% confidence intervals. There was no significant difference in localization sensitivity between up-right and inverted words. This was tested by means of a t-test, cor-rected by Holm-Bonferroni at α= 0.05.
Exploratory results Word length.
All stimuli were categorized as short (<6 letters) or long (>5 letters) words, which assured an equal distribution of tri-als and thus statistical power between the groups. A three factor repeated measures ANOVA indicated that across all presentation times long words were localized significantly better than short words [F(1, 33) = 9.453, p = 0.004, η2p =
0.223, BFincl= 1.801]. This effect of word length on
local-ization accuracy did not interact significantly with the orien-tation of the words [F(1, 33)= 0.707, p = 0.407, η2p = 0.021,
BFincl= 0.151] or the presentation times [F(1, 33)= 1.648, p
= 0.183, η2
p= 0.048, BFincl= 0.050], nor was it qualified by
a significant interaction of both orientation and presentation time [F(2.362, 77.930)= 0.360, p = 0.734, η2p = 0.011, BFincl=
0.025]. Post-hoc analysis showed that this word length effect was present at the second shortest presentation time, where long words (M = 0.665, SD = 0.122) were localized better than short words (M = 0.639, SD = 0.105, p = 0.01, d = 0.538). These results are shown in figure 4.
Figure 4. Word length results. The mean percentage correct for the localization task (n = 34) for each of the presentation times. Localization accuracy is plotted separately for long (>5 letters) and short words (<6 letters). Data shows that across all presentation times long words are localized significantly better than short words (p= 0.004). Post-hoc analysis shows that this word length effect is significant at the second shortest presentation time (p= 0.01). Word length was controlled for word frequency and word valence.
Word frequency.
Word frequency alongside orientation and presentation time effects were assessed by conducting a three factor re-peated measures ANOVA. The data showed a significant main effect of frequency on localization performance, but this was not supporter by Bayesian statistics [F(1, 33)= 4.782,
p = 0.036, η2p = 0.127, BFincl = 0.461]. This main effect
of frequency on localization accuracy was not qualified by a significant interaction with presentation time [F(1.810, 59.719)
= 1.781, p = 0.156, η2
p = 0.051, BFincl= 0.056] or
orienta-tion [F(3, 99)= 0.303, p = 0.586, η2p= 0.009, BFincl= 0.137].
Moreover, there was no significant interaction of word fre-quency, orientation and presentation time on localization ac-curacy [F(3, 99) = 0.626, p = 0.600, η2p = 0.019, BFincl =
0.065]. Post-hoc analysis did not indicate a significant di ffer-ence in localization accuracy between frequent (M= 0.650, SD= 0.116) and infrequent words (M = 0.654, SD = 0.107, p= 1.00, d = 0.060) at the second shortest presentation time (figure 5).
World valence.
A three factor repeated measures ANOVA was conducted to assess main and interaction effects of word valence, ori-entation and presori-entation time. Data showed no significant main effect of word valence on localization accuracy [F(1, 33)
= 0.342, p = 0.563, η2
p= 0.010, BFincl= 0.101] and no
signif-icant interaction effect of word valence and presentation time on localization accuracy [F(2.474, 81.640)= 0.443, p = 0.686, η2p
= 0.013, BFincl= 0.022]. Data further indicated a significant
interaction of word valence and orientation, but was not
sup-Figure 5. Word frequency results. The mean percentage cor-rect for the localization task (n = 34) for each of the presenta-tion times. Localizapresenta-tion accuracy is plotted separately for frequent (M = 10249.95 per million) or infrequent (M = 856.15 per mil-lion) words. Frequency was based on the SUBTLEX-NL database (Keuleers, Brysbaert, & New, 2010). Frequency was balanced be-tween the length and valence of the words.
ported by Bayesian statistics [F(1, 33)= 4.179, p = 0.049, η2p=
0.112, BFincl= 0.282]. However, post-hoc analysis did show
that none of the pairwise comparisons of these two factors were significant. This was not qualified by a significant in-teraction of word valence, presentation time and orientation on localization accuracy [F(2.155, 71.120)= 0.829, p = 0.448, η2p
= 0.025, BFincl= 0.042]. Post-hoc analysis did not indicate
a significant difference between negative (M = 0.664, SD = 0.120) and neutral words (M= 0.647, SD = 0.109, p = 1.00, d= 0.080) at the second shortest presentation time (figure 6).
Figure 6. Word valence results. The mean percentage correct for the localization task (n = 34) for each of the presentation times. Localization accuracy is plotted separately for negative and neutral words. Negative words were words that directly refer to or are as-sociated with negative emotions, disease, negative events, swearing or death. Neutral words were words that do not inherently refer to negative emotions, disease, negative events, swearing or death.
Discussion
In this study, the effects of familiarity with lexical stim-uli on unconscious processing were re-investigated. Specif-ically, we revisited word inversion effects to investigate if they are truly present when awareness of this critical di-mension is absent. For this purpose, the novel localization-discrimination dissociation paradigm from Stein and Pee-len (2020) was adopted, as this paradigm enables relating differences in localization of stimuli to differences in un-conscious processing in a clear and unambiguous manner. The current study found that at a stimuli presentation time of 25ms, individuals were still able to localize words with above chance performance, but were unable to discriminate between the critical dimension of the words (i.e. whether the word was upright or inverted), indicating unawareness of the critical stimulus dimension. At this presentation time, there was no indication of a difference in localization performance between upright and inverted words. Moreover, there was no indication of such a difference at any of the investigated presentation times, even those which allowed above chance discrimination of word orientation. Thus, in contrast with previous studies that suggested word inversion effects due to differential unconscious processing (Kerr et al., 2017; Yang et al., 2017; Yang & Yeh, 2011, 2014), this study did not demonstrate unconscious nor conscious inversion effects of words.
These contrasting results could be caused by the lack of statistical power in the current study, resulting in a false neg-ative for inversion effects. This could especially be the case since the face inversion effect seen in a study of Stein and Peelen (2020) is of medium magnitude in effect size terms and considering that inversion effects for words are reported to be substantially smaller than those of faces (Albonico, Fu-rubacke, Barton, & Oruc, 2018). Similarly, this lack of sta-tistical power might have led to a false demonstration of ab-sence of awareness, a particularly concerning issue in uncon-scious research (Vadillo, Linssen, Orgaz, Parsons & Shanks, 2020). Another concern might be that the paradigm was in-susceptible to detect effects of stimuli dimension on localiza-tion performance. However, the exploratory analysis shows that longer words were localized better than short words. This suggests that the paradigm was at least susceptible to detect some effect of amount of bottom-up stimulus energy on localization performance.
Moreover, discrepancies with previous studies could be explained with reasons other than the lack of statistical power or paradigm validity. This study adopted the novel localization-discrimination dissociation paradigm, which has the advantage of accurately delineating unconscious from conscious processes and with that, accurately relating per-formance differences to differences in unconscious process-ing. On the other hand, previous studies that reported unconscious word inversion effects all adopted a
reaction-time based b-CFS paradigm. As stated in the introduction, since there is no conclusive awareness control measure, this paradigm fails to delineate unconscious processes from post-perceptual processes, such as decision making processes. Word inversion effects reported in previous b-CFS studies could therefore be related to differential post-perceptual pro-cesses, such as different response criteria for upright words. In other words, the shorter reaction times for upright words compared to inverted words as seen in the b-CFS studies might not reflect shorter suppression times, but rather faster decision times. Visual word processing is thought to be sub-ject to highly efficient processing regimes for which experi-ence is skewed to upright orientations, possibly resulting in an orientation dependency (Barton, Hanif, Eklinder Björn-ström, & Hills, 2014; BjörnBjörn-ström, Hills, Hanif, & Barton, 2014; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001). Then especially when there is high perceptual uncertainty, evidence extraction might be particularly efficient for upright words, leading to the response threshold being reached more quickly. A study by Albonico, Furubacke, Barton & Oruc (2018) indeed showed that word inversion reduced process-ing efficiency in the presence of external noise. Alternatively, one could argue that people hold a less rigorous decision cri-terion for upright words compared to inverted words. This bias in processing or decision criterion could be the underly-ing mechanism that mediate words inversion effects seen in previous studies. On the other hand, here we adopted a novel paradigm that includes a conclusive control to effectively rule out awareness in conjunction of criterion-free measures that rule out the influence of decision criterion. Therefore, this study convincingly demonstrates an absence of unconscious and conscious word inversion effects.
However, the current results do not invalidate previously found word inversion effects nor reject the existence of such word inversion effects. They could demonstrate that the ef-fects may be less prevalent than previously claimed. Addi-tionally, the emergence of word inversion effects could be de-pendent on the methods used to render stimuli less visible or invisible. This study implemented visual backward masking, a method know to perturbate recurrent processes (Fahren-fort, Scholte, & Lamme, 2007; V. A F Lamme, Zipser, & Spekreijse, 2002; Rajaei, Mohsenzadeh, Ebrahimpour, & Khaligh-Razavi, 2019), while relatively preserving the early initial feedforward processes (Ghodrati, Farzmahdi, Rajaei, Ebrahimpour, & Khaligh-Razavi, 2014; Lamme & Roelf-sema, 2000). On the other hand, b-CFS makes use of strong interocular suppression and may actually perturbate the early initial feedforward processes (Moors, Hesselmann, Wage-mans, & van Ee, 2017; Sterzer, Stein, Ludwig, Rothkirch, & Hesselmann, 2014; Yuval-Greenberg & Heeger, 2013). This could mean that the preserved potential of conscious and unconscious computations depends on the methodology of the paradigm. As a result, differential processing of
up-right and inverted words may only become apparent in cer-tain paradigms, such as in b-CFS.
This also introduces one of the notable limitations of this study. Namely, if inversion effects rely on recurrence, it could be that a backward masking procedure diminished any of those effect. Although this seems unlikely, as Stein and Peelen (2020) have demonstrated face inversion effects with the same procedure, it has been suggested that inversion ef-fects for words and faces rely on fundamentally different pro-cessing properties (Albonico et al., 2018). In addition, re-currence is a proposed underlying component of conscious awareness (Dehaene & Naccache, 2001). If the underlying processes that result in word inversion effects indeed rely on recurrence, chances are that they are inevitably linked to con-scious perception of the stimuli. In order to go beyond spec-ulation, it is necessary that future neuroimaging studies un-veil the processing dynamics of upright and inverted words and those that govern orientation discrimination. King and Dehaene (2014) propose elegant ways of using multivariate pattern analysis (MVPA) applied to EEG or MEG data to gain insights into differences in temporal processing archi-tecture between conditions. Adopting methods, like using MVPA on neural data, is also part of the brain-based frame-work for unconscious processing proposed by Soto, Sheikh and Rosenthal (2019). They argue that this approach, in ad-dition to psychophysics, is critical to unveil unconscious pro-cesses that follow conditions that produce weak or no effects on behavior. This approach could prove especially fruitful for unconscious word processing, as previous studies that are in line with this framework demonstrate differential uncon-scious processing for conditions that did not produce signif-icant behavioral effects (Axelrod, Bar, Rees, & Yovel, 2015; Sheikh et al., 2019). Another serious limitation of the current study is the overall low discriminability, even at the longest presentation time. This might indicate that participants did not get familiar with the different stimuli signals, making the task particularly difficult. This high difficulty might have lead to a loss of motivation, which might have prevented word inversion effects from emerging. A potential solution is implementing a delay between the word stimuli and on-set of the mask, or stimulus onon-set asynchrony (SOA), which are known to reduce the effectiveness of backward masking as the SOA increases (Francis, 2000; Spencer & Shuntich, 1970). This would decrease task difficulty and potentially increase participants motivation.
This study further emphasizes the importance of deriv-ing criterion-free measures by implementderiv-ing non-speeded accuracy-based tasks to rule out the influence of post-perceptual decision criteria. Seeing replication studies of previously published work that report word inversion effects, while they implement the non-speeded accuracy based ap-proach, would be insightful. The results of these studies would allow a better comparison of results with the current
study and provide an approximation on the extent of post-perceptual influences. Importantly, future studies that adopt the novel localization-discrimination paradigm are essential to get a more accurate benchmark of the extent of uncon-scious processes, but also to re-investigate whether effects previously thought to reflect unconscious processes are gen-uinely unconscious.
To conclude, this study investigated whether word inver-sion effects are present when there is absence of awareness of the respective word orientation and thus, whether this ef-fect can be related to differential unconscious processing. By adopting the novel localization-discrimination dissocia-tion paradigm this study unveiled that, in contrast to previ-ous studies, word inversion effects are absent in both uncon-scious and conuncon-scious conditions. These findings provide a small, yet important contribution toward understanding un-conscious visual perception of words. Moreover, they pro-vide reason to reconsider the previously suggested uncon-scious origin of the word inversion effect. Several alternative, equally valid, possible origins have been discussed. A closer examination of these alternatives through future studies is necessary to gain a more profound understanding of uncon-scious word processing and its functionality. Importantly, for future studies, adopting the novel localization-discrimination dissociation paradigm should be considered, as it holds po-tential to broaden our current understanding of unconscious and conscious functionality.
References
Abdi, H. (2010). The greenhouse-geisser
cor-rection. In Encyclopedia of Research Design.
https://doi.org/10.1007/BF02289823
Albonico, A., Furubacke, A., Barton, J. J. S., & Oruc, I. (2018). Perceptual efficiency and the inversion effect for faces, words and houses. Vision Research. https://doi.org/10.1016/j.visres.2018.10.008
Axelrod, V., Bar, M., Rees, G., & Yovel, G. (2015). Neu-ral Correlates of Subliminal Language Processing. CerebNeu-ral Cortex. https://doi.org/10.1093/cercor/bhu022
Baars, B. J. (1988). A cognitive theory of consciousness. In A Cognitive Theory of Consciousness.
Barton, J. J. S., Hanif, H. M., Eklinder Björn-ström, L., & Hills, C. (2014). The word-length ef-fect in reading: A review. Cognitive Neuropsychology. https://doi.org/10.1080/02643294.2014.895314
Björnström, L. E., Hills, C., Hanif, H., & Barton, J. J. S. (2014). Visual word expertise: A study of inversion and the word-length effect, with perceptual transforms. Perception. https://doi.org/10.1068/p7698
Brainard, D. H. (1997) The Psychophysics Toolbox, Spa-tial Vision 10:433-436.
Cohen, L., Lehéricy, S., Chochon, F., Lemer, C., Rivaud, S., & Dehaene, S. (2002). Language-specific tuning of visual cortex? Functional properties of the Visual Word Form Area. Brain. https://doi.org/10.1093/brain/awf094
Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review. https://doi.org/10.1037/0033-295X.108.1.204
Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic
evidence and a workspace framework. Cognition.
https://doi.org/10.1016/S0010-0277(00)00123-2
Dehaene, S., Naccache, L., Cohen, L., Bihan, D. Le, Man-gin, J. F., Poline, J. B., & Rivière, D. (2001). Cerebral mech-anisms of word masking and unconscious repetition priming. Nature Neuroscience. https://doi.org/10.1038/89551
Erdelyi, M. H. (1986). Experimental
indeter-minacies in the dissociation paradigm of
sublim-inal perception. Behavioral and Brain Sciences. https://doi.org/10.1017/S0140525X00021348
Fahrenfort, J. J., Scholte, H. S., & Lamme, V. A. F. (2007). Masking disrupts reentrant processing in hu-man visual cortex. Journal of Cognitive Neuroscience. https://doi.org/10.1162/jocn.2007.19.9.1488
Francis, G. (2000). Quantitative theories of
metacontrast masking. Psychological Review.
https://doi.org/10.1037/0033-295X.107.4.768
Gayet, S., & Stein, T. (2017). Between-subject variabil-ity in the breaking continuous flash suppression paradigm: Potential causes, consequences, and solutions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2017.00437
Ghodrati, M., Farzmahdi, A., Rajaei, K., Ebrahimpour, R., & Khaligh-Razavi, S.-M. (2014). Feedforward object-vision models only tolerate small image variations com-pared to human. Frontiers in Computational Neuroscience. https://doi.org/10.3389/fncom.2014.00074
Hassin, R. R. (2013). Yes It Can: On the Functional Abil-ities of the Human Unconscious. Perspectives on Psycholog-ical Science. https://doi.org/10.1177/1745691612460684
Hesselmann, G., & Moors, P. (2015). Definitely maybe: Can unconscious processes perform the same func-tions as conscious processes? Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2015.00584
Hochstein, S., & Ahissar, M. (2002). View from the top: hierarchies and reverse hierarchies in the visual system. Neu-ron.
Holm, S. (1979). A Simple Sequentially Rejective Mul-tiple Test Procedure Author(s): Scandinavian Journal of Statistics. https://doi.org/10.2307/4615733
JASP Team. (2019). JASP. [Computer Software]. Jiang, Y., Costello, P., & He, S. (2007). Processing of in-visible stimuli: Advantage of upright faces and recognizable words in overcoming interocular suppression. Psychological Science. https://doi.org/10.1111/j.1467-9280.2007.01902.x
Kerr, J. A., Hesselmann, G., Räling, R., Warten-burger, I., & Sterzer, P. (2017). Choice of analy-sis pathway dramatically affects statistical outcomes in breaking continuous flash suppression. Scientific Reports. https://doi.org/10.1038/s41598-017-03396-3
Keuleers, E., Brysbaert, M., & New, B. (2010). SUBTLEX-NL: A new measure for Dutch word frequency based on film subtitles. Behavior Research Methods. https://doi.org/10.3758/BRM.42.3.643
King, J. R., & Dehaene, S. (2014). Characteriz-ing the dynamics of mental representations: The tempo-ral genetempo-ralization method. Trends in Cognitive Sciences. https://doi.org/10.1016/j.tics.2014.01.002
Kirk, R. (2006). Zombies and
Con-sciousness. In Zombies and Consciousness.
https://doi.org/10.1093/0199285489.001.0001
Kleiner M, Brainard D, Pelli D, 2007, “What’s new in Psychtoolbox-3?” Perception 36 ECVP Abstract Supple-ment.
Kouider, S., & Dehaene, S. (2007). Levels of processing during non-conscious perception: A critical review of visual masking. Philosophical Transactions of the Royal Society B: Biological Sciences. https://doi.org/10.1098/rstb.2007.2093
Kouider, S., & Faivre, N. (2017). Conscious and Uncon-scious Perception. In The Blackwell Companion to Con-sciousness. https://doi.org/10.1002/9781119132363.ch39
Lamme, V. A F, Zipser, K., & Spekreijse, H.
(2002). Masking interrupts figure-ground
sig-nals in V1. Journal of Cognitive Neuroscience.
https://doi.org/10.1162/089892902320474490
Lamme, Victor A.F., & Roelfsema, P. R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neurosciences. https://doi.org/10.1016/S0166-2236(00)01657-X
Levine, J., & Flanagan, O. (1994).
Conscious-ness Reconsidered. The Philosophical Review.
https://doi.org/10.2307/2185744
Macmillan, N. A., & Creelman, C. D. (2005). Detection theory: A user’s guide (2nd ed.). Lawrence Erlbaum Asso-ciates Publishers.
Mathôt, S. (2017, May 15). Bayes like a Baws: Interpret-ing Bayesian Repeated Measures in JASP. Retrieved from https: //www.cogsci.nl/blog/interpreting-bayesian-repeated-measures-in-jasp
McCandliss, B. D., Cohen, L., & Dehaene, S. (2003). The visual word form area: Expertise for reading in
the fusiform gyrus. Trends in Cognitive Sciences. https://doi.org/10.1016/S1364-6613(03)00134-7
Moors, P., Hesselmann, G., Wagemans, J., & van Ee, R. (2017). Continuous Flash Suppression: Stimulus Fraction-ation rather than IntegrFraction-ation. Trends in Cognitive Sciences. https://doi.org/10.1016/j.tics.2017.06.005
Pelli, D. G. (1997) The VideoToolbox software for visual psychophysics: Transforming numbers into movies, Spatial Vision 10:437-442.
Prioli, S. C., & Kahan, T. A. (2015). Identifying words that emerge into consciousness: Effects of word valence and unconscious previewing. Consciousness and Cognition. https://doi.org/10.1016/j.concog.2015.04.005
Rabagliati, H., Robertson, A., & Carmel, D. (2018). The importance of awareness for understanding lan-guage. Journal of Experimental Psychology: General. https://doi.org/10.1037/xge0000348
Rajaei, K., Mohsenzadeh, Y., Ebrahimpour, R., & Khaligh-Razavi, S.-M. (2019). Beyond core object recog-nition: Recurrent processes account for object recog-nition under occlusion. PLOS Computational Biology. https://doi.org/10.1371/journal.pcbi.1007001
Schmidt, T. (2015). Invisible stimuli, implicit thresh-olds: Why invisibility judgments cannot be inter-preted in isolation. Advances in Cognitive Psychology. https://doi.org/10.5709/acp-0169-3
Sheikh, U. A., Carreiras, M., & Soto, D. (2019). Decoding the meaning of unconsciously processed
words using fMRI-based MVPA. NeuroImage.
https://doi.org/10.1016/j.neuroimage.2019.02.010
Sklar, A. Y., Levy, N., Goldstein, A., Mandel, R., Maril, A., & Hassin, R. R. (2012). Reading and doing arithmetic nonconsciously. Proceedings of the National Academy of Sciences of the United States of America. https://doi.org/10.1073/pnas.1211645109
Soto, D., Sheikh, U. A., & Rosenthal, C.
R. (2019). A Novel Framework for
Uncon-scious Processing. Trends in Cognitive Sciences. https://doi.org/10.1016/j.tics.2019.03.002
Spencer, T. J., & Shuntich, R. (1970). Evidence for an interruption theory of backward masking. Journal of Experi-mental Psychology. https://doi.org/10.1037/h0029510
Stein, T., Awad, D., Gayet, S., & Peelen, M. V. (2018). Unconscious processing of Facial Domi-nance: The role of low-level factors in access to aware-ness. Journal of Experimental Psychology: General. https://doi.org/10.1037/xge0000521
Stein, T., Hebart, M. N., & Sterzer, P. (2011). Breaking continuous flash suppression: A new mea-sure of unconscious processing during interocular
suppression? Frontiers in Human Neuroscience.
https://doi.org/10.3389/fnhum.2011.00167
Stein, T. & Peelen, M. V. (2020) Dissociating conscious and unconscious influences on visual detection. Unpublished manuscript.
Sterzer, P., Stein, T., Ludwig, K., Rothkirch, M., & Hes-selmann, G. (2014). Neural processing of visual information under interocular suppression: A critical review. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2014.00453
Sussman, B. L., Reddigari, S., & Newman, S. D. (2018). The impact of inverted text on visual word processing: An fMRI study. Brain and Cognition. https://doi.org/10.1016/j.bandc.2018.02.004
Vadillo, M. A., Linssen, D., Orgaz, C., Parsons, S., & Shanks, D. R. (2020). Unconscious or underpowered? Prob-abilistic cuing of visual attention. Journal of Experimental Psychology: General. https://doi.org/10.1037/xge0000632
Van Gaal, S., & Lamme, V. A. F. (2012). Uncon-scious high-level information processing: Implication for neurobiological theories of consciousness. Neuroscientist. https://doi.org/10.1177/1073858411404079
Weiner, K. S., & Zilles, K. (2016).
The anatomical and functional specialization
of the fusiform gyrus. Neuropsychologia.
https://doi.org/10.1016/j.neuropsychologia.2015.06.033 Yang, Y. H., Tien, Y. H., Yang, P. L., & Yeh, S. L. (2017). Role of consciousness in temporal integration of semantic information. Cognitive, Affective and Behavioral Neuro-science. https://doi.org/10.3758/s13415-017-0525-9
Yang, Y. H., & Yeh, S. L. (2011). Accessing the meaning of invisible words. Consciousness and Cognition. https://doi.org/10.1016/j.concog.2010.07.005
Yang, Y. H., & Yeh, S. L. (2014). Unmasking the di-choptic mask by sound: Spatial congruency matters. Experi-mental Brain Research. https: //doi.org/10.1007/s00221-014-3820-5
Yuval-Greenberg, S., & Heeger, D. J. (2013). Con-tinuous flash suppression modulates cortical activ-ity in early visual cortex. Journal of Neuroscience. https://doi.org/10.1523/JNEUROSCI.4612-12.2013
