Contextual influence on emotion and race processing

C

ONTEXTUAL INFLUENCE ON

EMOTION AND RACE PROCESSING

Tessa C. van Lobbrecht

Bachelor Research Psychobiology Dr. Marte Otten (supervisor) Department of Brain & Cognition

University of Amsterdam Submitted: 22 january 2021

A

BSTRACT

In this article, we elaborate on two conducted experiments that were aimed at researching the influence of context on early perceptional processes beyond conscious awareness. Based on the assumption that invisible faces can be partially processed early on in high-level visual networks, we used visible primes as contextual factors and indirectly measured unconscious processing with backward masked face localization tasks. Experiment 1 (n=23) used valanced sentence priming and expressional targets (e.g., happy and angry faces) to test whether emotional congruency effects (in reaction time and accuracy) could be observed but lacked statistical power and failed to provide significant results. Experiment 2 (n=58) was based on racial bias (namely, Black people associated with guns) and used images of tools and weapons as priming combined with neutral Black and White face-targets, with congruent conditions now described by Black-gun or White-tool trials. Again, no congruency effects were found, whereas we did find longer presentation positively influenced speed and accuracy. Thus, our findings lean towards the notion that these forms of contextual factors do not influence early perceptional processes or suggest insufficiency of our experimental design, demanding optimization. Despite not obtaining robust evidence for unconscious perceptional processing in the current experiments, the masking-priming paradigm as well as established early facial processing has previously shown to be opportunistic for the unconsciousness-field of study and is worth reconsidering as sufficient research remains limited to date.

I

NTRODUCTION

I

magine walking through a park in a new city. There are two situations: in one, you walk

in on a sunny day, dogs running round and a young couple smiles at you. An elderly man is sitting on a bench reading a book, there is live music. In the other it is dark and cloudy. The park is half-empty with merely a hooded person is sitting alone, while you hear some men screaming angrily at each other across the park. In both, someone seemingly ungroomed comes up to you asking for help. Would you instantly react the same, or would one make you hesitate for a longer moment?

It is widely investigated how context (e.g., information surrounding an experience or situation; Chen et al., 2020) influences (emotional) perception of our environment. Perception studies are aimed at discovering what processes underlie subjective experiences and how for instance (social) behavior is influenced as a result. A prominent way of exploring unconscious perception is through faces, facial processing being a sustained field of research which has provided established and reliable localized face-selective neural mechanisms (Haxby, Hoffman, & Gobbini, 2000). With the ability to display a great deal of character and identity as well as emotion, faces propose a relevant factor in the social world and theories of unconscious face processing are popular (Axelrod, Bar, & Rees, 2015; Lehmann et al., 2004; Stewart et al., 2012). Regardless of its expression, a face is a salient emotional stimulus allowing us to distinguish friend from foe and conveying crucial information for social interactions (e.g., identity, race, sex, attractiveness, direction of eye gaze). Thus, all faces, even so called “unexpressive” or “neutral” faces will have emotional significance and may have special access to visual attention (Palermo & Rhodes, 2007). Rapid facial processing serves as a good medium for investigating unconscious perception, since processes that occur in sensory cortices early on after stimulus onset that aim to process features of visual presentation are the basis of what we call perception (de Gelder et al., 2006).

Here we will test whether context in the form of linguistic or pictorial cues facilitates early (pre-conscious) visual processing of faces. Besides visual information, recognition as part of perception also requires additional knowledge such as memory, which could hold information about early perceptual properties of visual presentations, to which another image could be compared subsequently (Adolphs, 2002). Facial perception therefore seemingly uses knowledge regarding contingencies between visual facial characteristics and other associated stimuli, such as where a face is seen, what is said about or by the person or related events, in other words: context (Barrett & Kensinger, 2010; Barrett, Mesquita, & Gendron, 2011; Wieser & Brosch, 2012). We aim to research unconscious processing by looking at two context-face effects, namely emotion-congruence and bias-congruence. Both indicate positive effects of context on performance regarding a facial task when the prior is coherently associated with the latter. Emotion-congruence has been researched widely, whereas bias-congruence is quite novel. In the next section we will explain more on the current research method, whereafter we review a selection of the extensive literature on emotion-congruence effects in face processing followed by insights on bias-research.

U

NCONSCIOUS PERCEPTION PROCESSES

A way to study contextual modulation of especially unconscious perception is through priming. Priming effects quantify whether an invisible stimulus (usually the prime) facilitates the processing of another stimulus (usually the target) with similar features or associations (Kiesel, Kunde, & Hoffmann, 2007). The information similarity between prime and target enables inferring at what level the invisible stimuli has been processed, ranging from low-level feature (e.g., colors) to high-level (e.g., emotional and semantic information) processing (Izatt, Dubois, Faivre, & Koch, 2014). To achieve stimulus invisibility several techniques are available, of which the backward masking (BM) paradigm is a prominent one. In BM a mask is shown following, and sometimes also preceding, a briefly presented stimulus to hinder conscious stimulus processing (Izatt et al., 2014). Applied in experimental priming tasks, unconscious processing can then be detected by comparing the direct

measure of conscious awareness through cognitive tasks with indirect measures of unconscious processing of the same stimulus (Stein, Hebart, & Sterzer, 2011). For instance, one such study looked at likability judgements of novel objects after a preceding masked expressional face and showed that both invisible angry and happy facial expressions resulted in unconscious affective prime effects revealed by correspondingly inclining judgements (Almeida, Pajtas, Mahon, Nakayama, & Caramazza, 2013). These findings suggested influenced perceptional processing of the non-valanced object as well as partial processing of masked faces.

Another direct measure of consciousness without requirement to identify or recognize targets consciously, thereby diminishing any form of categorization as main focus, is face localization, where solely the location of a presented face-target is asked (Stein et al., 2011). In our study we aimed strictly at researching contextual influence on subconscious processing of faces to gain insight on early perception and wanted to exclude conscious perceptional intent. Therefore, we incorporated invisible face detection: we adapted the priming paradigm by presenting visible contextual primes and rendered targets in a face localization task invisible through BM. We created two condition types, “congruent” (i.e., the prime is presumably associated to the proceeding target) and “incongruent” (i.e., the prime is not associated with the target). Furthermore, we chose to present faces briefly to create a research frame in which individuals are (even more) subjectively unaware of the face stimuli disabling them of conscious report (we call this “suboptimal” presentation) but still allowing some early processing. In studies where affective priming was done in optimal and suboptimal conditions followed by a judgement-task of novel neutral targets it was found that emotional face priming, compared to empty (neutral) priming, led to valence congruent shifts in affective evaluation of targets in only suboptimal (shorter) presentation trials (Murphy & Zajonc, 1993). Suggestions are that the most direct effects of affective stimuli are obtained when minimizing interference by conscious processing (Rotteveel, de Groot, Geutskens, & Phaf, 2001). Whether this is the case with suboptimal face targets following contextual priming, we wanted to find out in our current experiments by opting for two differential presentation durations to compare congruence-priming effects.

E

MOTION

-

PROCESSING AND CONTEXT

There is a general assumption that unconscious emotion can be processed outside of awareness, even though the exact neural mechanisms of unconscious processing remain unclear (Tamietto & De Gelder, 2010). Formation of emotional perception is influenced in various ways and it is found that implicit integration of relevant contextual cues takes place during the process (Chen et al., 2020). This was hypothesized to be due to facilitated retrieval of information by activation of associated information within a contextual frame. We can hypothesize that if the observed context activates certain semantic networks associated with emotions, this potentially influences pre-conscious processing, such as attention and anticipation, and therefore early emotional perception, which in place shapes the meaning and responses given to events (Palermo & Rhodes, 2007). It is for instance found that which emotion is seen in another person’s face is strongly influenced by contextual factors (Barrett et al., 2011). An early study looking at contextual congruency effects found that emotional cartoons paired with undetectable congruent or incongruent faces over several learning trials were afterwards identified faster when congruently paired, suggesting that even with subliminal face presentation transferring of valence happens unconsciously (Niedenthal, 1990).

Another study similarly showed that when subjects performed a Stroop-task with categorizing emotional words superimposed over expressional faces, responses were slower for incongruent word-face pairs, indicating deep representations of emotion at semantic level (Preston & Stansfield, 2008). The authors hypothesized that words (as context) activated one’s own associated relevant emotional memories. Barrett, Lindquist, & Gendron (2007) likewise showed that emotion perception is presumably influenced by such context instead of solely by bottom-up information collected from emotional facial expressions. Their review summarized earlier studies regarding language as main contextual factor, appointing the relevance of words in emotion perception, and it was discussed that linguistical shaping of sensory processing happens when seeing another’s face. That linguistic context can be encoded emotion was explicitly seen when labelling emotional

expressions improved the memory of context in which it was shown, whereas affective judgement based on merely structural features of the face did not (Barrett & Kensinger, 2010). Later it was found that words can be used to create perception of emotion through grounding various and ambiguous emotion categories (Lindquist & Gendron, 2013). In perceptional tasks of another study, it was found that showing emotionally familiar words (i.e., associated with known emotion concepts) that are temporarily made inaccessible by excessive repetition (i.e., the semantic meaning is disconnected from the term; Tian & Huber, 2010)before presenting emotionally related face-targets resulted in decrease of positive priming effects (Gendron, Lindquist, Barsalou, & Barrett, 2012). The authors specifically ruled out post-perceptional processes (e.g., decision making or implicit labelling that happen after perception is formed) by using non-identification or emotional judgement tasks and suggested that conceptual knowledge about emotion, encoded with emotion words, is a key element of generating emotion perception. Additionally, behavioral and electrophysiological data on recognition of facial expressions propose that perceptual integration of information appears automatically even before structural stimulus encoding and does not require later high-level semantic analysis (de Gelder et al., 2006).

Multiple findings mentioned above consistently suggest that structural facial configuration alone is not sufficient for emotion perceptions and that contextual information is automatically deployed for early perception. Therefore, we now focused on effects of context on early (pre-conscious) visual perception and in our first study we tested the hypothesis that language serves as context contributing to emotional perception forming of faces, even without emotion-based goal. Experiment 1 combined linguistic priming with sentences describing a valanced (positive or negative) event, with a face localization task using invisible emotional faces. We believed sentences could be constructive contextual primes by activating semantic knowledge about the event described which proposes possible facilitation of detecting emotional faces. It is expected that the face location task is performed faster and more accurate when using congruent face-sentences combinations (e.g., a positive priming sentence preceding a positive target face) in both long and (suboptimal) short presentation.

C

ONTEXTUAL BIAS

-

CUES AND FACE PROCESSING

Especially for mapping an individuals’ associative biases, contextual priming is an interesting focus with its’ ability to activate certain ideas, feelings or categories whereafter the effects on cognitive performance can be measured. It provides an opportunity to indicate mental content that was activated automatically or implicitly while bias of self-report is avoided (Cameron, Brown-Iannuzzi, & Payne, 2012). The last authors exemplified that similar contextual factors tend to activate differentiating associative links for different people: a passage from the bible may for example activate very distinctive ideas in the minds of Christians and Atheists and the face of a Black man might just evoke different feelings for individuals low or high in prejudice. As mentioned, facial perception is presumably always influenced by contextual variables and the situational form of context may be provided by the perceiver. Previously attained social information gained through affective learning and implicit processing biases that are formed, like race bias, can serve as such variables (Wieser & Brosch, 2012). The latter internal factor is a well-known social phenomenon and can be indirectly measured through affective priming effects.

While researching individual racial attitudes it was seen that even emotion perception is seemingly altered by racial biases, observed in a change detection task where subjects watched Black or White faces morphing from happy to angry or vice versa and indicated expression onset. Results showed later perceived offset and earlier onset of anger expressions in Black compared to White faces, demonstrating higher implicit bias associated with a higher likelihood to perceive anger in Black people (Hugenberg & Bodenhausen, 2004)

.

In another study, priming of Black and White faces followed by pleasant and unpleasant target-words influenced the evaluation of these words as good or bad. Prime race was seen to facilitate response time of target evaluation in both ways: Black primes facilitated negative words in White subjects, whereas White primes facilitated negative words in Black subjects (Fazio, Jackson, Dunton, & Williams, 1995). Conjointly, at neural level race biases on face

perception have been evident, as it was shown that higher amygdala (read: emotion center) activation patterns toward Black faces were found in participants with high implicit racial bias measured on the implicit association test (IAT) compared to those with lower bias (Cunningham et al., 2004). This establishes racial bias to play an important role in implicit perception, but processing mechanisms remain relatively unclear in situations without explicit emotional judgement.

A related significant social bias targeting implicit perception is the race-gun association linking Black people to weapons. This is supported by findings such as larger shooter bias against Black versus White targets, with participants observed quicker to shoot armed Black targets but slower to not-shoot unarmed Black targets (Mekawi & Bresin, 2015). Two experiments done by Payne (2001) showed that racial bias drove misidentification of tools as guns more often and made gun identification faster after Black face priming relative to White. Previous studies with similar experimental designs to our current used Black and Asian faces as primes for subsequent tool/gun localization to investigate coming in-to awareness(Stein, Ciorli, & Otten, 2020). No effect was found suggesting no alteration of early visual processing due to Black-gun association. However, the study was done with for Western people less-common Asian faces and priming with neutral ethnic faces does not serve as contextual factor in the way we aim to investigate. It is worth reevaluating potential race-gun association effects with an experiment using more familiar races and reversing the prime-target order to directly activate semantic networks in opposite ways. Our second experiment therefore replicated the experimental set-up of current experiment 1, consisting of contextual priming followed by face localization with backward masked invisible faces shown for short or long presentation duration, but used neutral Black and White faces as targets to investigate the impact of race bias and stereotypes on perception of faces. In light of the described race-gun bias contextual primes consisted of images of tools and weapons, assuming that semantic knowledge of these objects can shape early stages of perceptual processing. We primarily believed that strong associations between Black people and weapons would result in positive influence of gun-priming on Black face-processing and thus to observe an increase in performance of Black target-localization. Concretely, an improvement of accuracy and speed in face localization was expected for weapon-Black and potentially also for tool-White (as both congruent) conditions in all presentation durations.

E

XPERIMENT

1

In experiment 1, linguistic contextual priming was combined with implicit emotion perception of masked expressional faces. Target awareness was taken into account. Note that due to current Covid-19 related measurements individual experimental settings could not be controlled.

M

ATERIALS

&

METHODS

P

ARTICIPANTS

Thirty-five undergraduates were recruited from the University of Amsterdam via Lab.uva.nl to participate in the online experiment and received course credit. The description of the experiment informed about the purpose of researching influence of emotional context on processing of faces through a task. All had normal or corrected-to-normal vision and were without a history of psychiatric or neurological illness. Twelve were Dutch native speakers and eighteen were English native speakers, other native languages were excluded. The experiment was approved by the University of Amsterdam’s Institutional Review Board and all subjects gave informed consent in compliance with federal and institutional guidelines beforehand. Groups used in experiment 1 were nonoverlapping with experiment 2. Thirty subjects completed the entire task and were included in the initial analysis (21 females; mean age, 19.9 years; range, 18–29 years). For parts of the analysis, six participants were excluded due to matching the exclusion criteria for prime check-ups. Of the remaining 24 participants that did not fulfill any of the exclusion criteria, eighteen were female with mean age of 19.8years old, with ten Dutch and fourteen English native speakers.

S

TIMULI

&

DISPLAY

Scripting and executing of the experiment were done in NeuroTask. Qualtrics was used to introduce the subjects, obtain informed consent and forward the subjects to the task. Sentences were shown in black letters of 125% font size on a white screen. Due to obligatory social distancing, experiments are done in private at home. Therefore, display apparatus could not be controlled over participants. Primes consisted of 160 sentences describing an event, all written in third-person singular perspective and ranged from 3 to 14 words long. We used two separate sets of the same style sentences in Dutch and English depending on the subjects’ native language.

80 of the sentences described a positive event (e.g., She just received an A on

her paper) and 80 a negative event (e.g., He just missed his flight home). Of

both the positive and negative sentences, half were written with a female subject and half with male subject. Target images were upright faces of 4 male and 4 female individuals, obtained from the Radboud Face database (RaFD). Of each individual two pictures were used with both emotional expressions (happy and angry), resulting in 16 targets. The images were toned black-and-white and were blended out gradually at the contour of the faces, so that there were no hard edges visible from the outline of the face and contrast with the white background of the screen was minimized (see fig. 1).For backward masking 5 slightly differing black-and-white backward high-contrast random-noise patterns were used.

P

ROCEDURE

Prior to the task, subjects gave informed consent in compliance with

federal and institutional guidelines. The experiment was entered through Lab.uva.nl, where a brief description is shown explaining an experiment on unconscious processing in relation to context and overall duration of approximately 1 to 1.5 hours. Subjects were instructed to focus on the fixation cross and carefully read the sentence, followed by a masked target localization task. The “A” key is pressed if they suspect to have seen any target on the left of the fixation and the “L” if on the right, as accurately and fast as possible. If unable to categorize the targets, participants were instructed to guess. A trial started with 500ms of a fixation cross and primes displayed for 2s, followed by 200ms of fixation and then the target. Target faces were shown for either short presentation (SP) of 30ms or long presentation (LP) of 80ms based on earlier studies (Dimberg, Thunberg, & Elmehed, 2000). After every target 4 masks were shown for 100ms each. An extra pre-mask with differing black-and-white noise pattern was shown for 100ms before each target to secure masking-effects (see fig. 2). Emotional valence (positive vs. negative) of targets and primes was manipulated over trials resulting in four “prime-target” condition, two congruent and two incongruent: Positive-Happy, Negative-Happy, Positive-Angry, Negative-Angry. These four conditions existed separately for male/female and all conditions were doubled again for both presentation durations, resulting in eight conditions in SP and the same eight in LP. Sentences of each category (i.e., male/female and positive/negative) were shown in randomized order each trial and an individual sentence was shown 2 times over the whole experiment. Location of face targets and presentation (SP/LP) were also randomized in equal proportions, as were the target faces with each individual face being shown 5 times over the whole main experiment. There were 8 blocks of 40 trials with a break in between each block until the subject decided to continue through a keypress. A practice block consisting of 6 trials was done before starting the main experiment. Prime check-ups are implemented in the task, where after the response a screen appears asking participants to type in the gist (i.e., the main subject or essence) of the sentence they have lastly read. These were added because of the inability of physically monitoring participants to ensure their attention to the prime sentences, and avoiding inactive participation resulting in random responses. Each block of 40 trials contained 6 prime check-ups at random locations, but after similar trial numbers between participants.

Fig. 1: face target used in experiment 1, blended out white. Example of an angry male face.

Fig. 2: overview of experimental trial with short presentation and happy face-target in Experiment 1.

T

ARGET DISCRIMINATION TASK

After the localization task trials, there were post-trials of a target discrimination task to measure the extent of participants’ awareness of the masked targets. In this task the same set-up is used as the main localization task trials, with the exception that participants are instructed to determine solely whether the target is identified as angry face or happy face, so that location detection is irrelevant. 16 target faces were presented multiple times (equal in SP/LP). The A-key is used for “angry” and the L-key for “happy”. There are 4 blocks of 40 trials to be completed, again all targets and location were equally randomized over trials.

P

RE

-

PROCESSING DATA

A power analysis was done beforehand indicating a total sample size of 54 subjects for 80% power (effect size f=0.176). Since target location and target type (face category) were randomized and both independent of congruency, no association exists between the prime-target relationship and the left-right response (Stein et al., 2011). Obtained data could therefore be analyzed specifically for effects of prime congruence and presentation duration on the response characteristics, i.e., the reaction time (RT) and the accuracy rate. Prime check-ups were analyzed manually for answering the correct gist and answers were assessed correct if the main subject (noun) as well as the emotional connotation of the verb (e.g., “lost the game” vs. “won the game”) were in line with the previous prime sentence. A maximum of 40% was allowed to be inaccurate before subject exclusion and six subjects were excluded with 24 remaining for further analysis.Reaction time means and mean accuracy rates were calculated of all main and post trials per subject separately. Target awareness tasks focused on discrimination accuracy; therefore, RTs were not initially taken into analyzation and the sensitivity index d’ was calculated (Stein, Utz, & Van Opstal, 2020). A “hit” was defined by accurate categorization of angry targets answered as angry and “false alarm” was defined by happy targets answered as angry.

R

ESULTS EXPERIMENT

1

M

AIN ANALYSIS

R

EACTION TIME

Analysis was done in JASP. Mean RTs were submitted to a repeated-measures ANOVA (2 x 2 model) with prime congruency and presentation time as within-subject factors. The main effect of congruency was not significant (F(1, 23)=.013, p = .91) indicating that congruent trials were not responded to faster than incongruent trials. Neither could presentation time effects be observed (F(1, 23)=.56, p=.46), with no significant difference of RT between SP and LP targets (see fig. 3B). Target presentation time and congruency did not significantly interact with each other (F(1, 23)<0.001, p=.99). Following, interaction effects were analyzed. No significant effects of congruency in either SP (F=.071, p=.79) or LP (F=.003, p=.95) were found, and none of presentation time in congruent or incongruent trials (F=.26, p=.62; F=.23, p=.64).

A

CCURACY RATE

Mean error rates were submitted to the same model with prime congruency and presentation time as within-subject factors. There was no main effect of congruency found on accuracy (F(1, 23)=.30, p=.59). A marginally significant effect of the presentation time was observed (F(1, 23)=139.46, p<.001), with LP targets responded to 25.1% more accurate than SP targets (M(SP)=.64, SD=.14; M(LP)=.90, SD=.12). Target presentation time and congruency did not significantly interact with each other (F(1, 23)=.37, p=.85). Simple main effects were used to analyze interaction effects and found no congruency effect within SP or LP (F=.068, p=.80; F=.61, p=.44, respectively): there was no higher accuracy with congruent priming in either presentation durations found. A significant difference between performance in accuracy rates between presentation durations in congruent and incongruent trials was found (F(con)=102.50, F(inc)=159.40, both p<0.001)(see fig.3A).

A. B.

Fig. 3: Experiment 1 showed significant effects of presentation time on accuracy rates (A), accounted for by lower accuracy rates in SP trials (M=.64, SD=.14) as compared to LP trials (M=.90, SD=.12). No significant results were found for presentation or congruence effects on RTs (B), with relatively close means.

T

ARGET AWARENESS

Target discrimination accuracy was tested against chance using a binomial test with 160 trials and chance level of 50% (= 0.5) correct and indicated that target discrimination above 58% (= 0.58) correct was significantly above chance level (one-tailed). Calculated d’ sensitivity index was tested against chance and showed to differ significantly with t(23)=3.87, p<.001 (one-tailed), dz=.79.

C

ONCLUSION

1

In neither reaction time nor in accuracy analysis any priming effects. General performance on face localization does not seem to be for facilitated through semantic priming, as there was no improvement found for congruent priming. We did, however, find an effect of presentation duration showing that longer presentation duration positively influenced accuracy within congruent and incongruent trials. Comparing effects on RT and accuracy it became apparent that where accuracy increased with longer presentation, RT showed no effects of this sort. If RTs would slow down, this would be a performance trade-off: the effort put in accuracy would detract from fast responding and vice versa. Since RT remained the same, increased accuracy-performance is not explained by trade-off but simply by improvement of face localization accuracy. Contrary to what we expected to see, merely longer presentation seems to be beneficial for localization performance which leads us to assume that early low-level visual processing is used to employ face localization, without reason to believe semantic priming influences any unconscious processing of faces. Another reason for current failure to find evidence for the theory of contextual influence on early emotional processing could have been a lack of methodical sufficiency. Mainly, targets were not masked objectively invisible in longer presentation and sample size did not provide enough statistical power. We will elaborate on possible explanations and shortcomings in the discussion.

E

XPERIMENT

2

Experiment 2 followed the same layout and theoretical set-up as experiment 1, but prime images were shown serving as context instead of sentences and masked face localization now used neutral faces of Black and White individuals. As in experiment 1, participants’ primary task was to indicate as fast and accurately as possible the side on which the face or any facial part appeared. In addition, participants judged their subjective impression of face race in a discrimination task serving again to check target awareness.

M

ATERIALS

&

METHODS

P

ARTICIPANTS

Sixty-four new undergraduates were recruited from the University of Amsterdam through Lab.uva.nl to participate in the online experiment for course credit. The task was publicly described as face localization task with ethnic aspect to it. All participants had normal or corrected-to-normal vision and were without a history of psychiatric or neurological illness. The experiment was approved by the University of Amsterdam’s Institutional Review Board and all subjects gave informed consent in compliance with federal and institutional guidelines beforehand. All subjects completed the entire task and were included in the initial analysis (48 females; mean age, 20.9 years; range, 18–56 years). Six participants were excluded due to matching the exclusion criteria for the prime check-ups. In the subgroup of 58 participants that did not fulfill any of the exclusion criteria, 43 were female (mean age, 20.9; range, 18-56). Ethnic background was checked and 40 reported being

White, 4 Mixed race, 3 Asian, 1 Hispanic and 1 Black.

S

TIMULI

Scripting and executing of experiment 2 were similar to experiment 1, done using Qualtrics as introduction and NeuroTask for the task. Experiments were performed online in private, hence display apparatus could not be controlled. Primes consisted of images of guns and can openers, shown on white backgrounds. Both categories contained six distinctive pictures. The target set existed of upright faces of 8 male and 8 female individuals with neutral expressions, obtained from the Chicago Face Database. Half of both were White, and half were Black individuals. Contrast and saturation were lowered, and images were blended out white gradually at the contour of the faces to demolish hard edges on the

Fig. 4: face target used in experiment 2, blended out gradually at the edges. Example of one of four Black female

outline (see fig. 4). Backward masking was used to render the faces invisible, using the same black-and-white noise images as the prior experiment, combined with colorful Mondrian-masks.

P

ROCEDURE

The experiment is entered through Lab.uva.nl, where a brief description is shown explaining an experiment on unconscious processing in relation to context and ethnicity and overall duration was indicated at about 75 minutes. Subjects are asked to focus on the fixation cross and observe the image, where after a masked target localization task is performed. Again, one pre-mask before and 4 masks after a target were shown for 100ms each. The “A” key is responded if they suspect to have seen a target on the left side and “L” if seen on the right side of the fixation cross, as accurately and fast as possible (see fig. 5). All were encouraged to follow their spontaneous intuition or guess if necessary. Trials operated the same as in experiment 1, with faces shown for equally randomized SP or LP on randomized location. Main trials consisted of 8 blocks of 40 trials with a break in between each block until subject decides to continue through a keypress and again a practice block of 6 trials was performed before starting the experiment. All targets and primes were equally randomized over trials. Four conditions applied: female congruent, male congruent (both Black-Gun and White-Tool), female incongruent and male incongruent (both Black-Tool and White-Gun). Prime check-ups to ensure subjects’ attention were implemented in the task after several location-responses, asking whether a tool or gun was lastly shown. For “gun” the A-key was responded and for “tool” the L-key. Each block of 40 trials contains 6 prime check-ups at random locations, but after similar trial numbers between participants.

Fig.5: overview of experimental trial with short presentation and White neutral face-target in Experiment 2.

T

ARGET DISCRIMINATION TASK

To measure target awareness a target discrimination task was added similar to experiment 1, with instructions to determine solely whether the target is identified as Black or White, without relevance of location. 16 target faces were presented multiple times in equally random presentation durations. The A-key is used for “black” and the L-key is used for “white” and there were 4 blocks of 40 trials, again all targets and location were equally randomized over trials.

P

RE

-

PROCESSING DATA

The power analysis remained similar to experiment 1, indicating a sample size of 54 subjects for 80% power (effect size f=0.176). Prime check-ups are statistically analyzed for answering the right prime category (gun or tool) and a maximum of 40% of check-ups is allowed to be inaccurate before

subject exclusion. Six subjects were excluded with 58 remaining for further analysis. Mean RTs and mean accuracy rates of all responses per subject are calculated. In the target awareness task RTs are not initially analyzed, but mean accuracy as well as the sensitivity index d’ were calculated. Now, a “hit” was defined by accurate categorization of Black targets answered as Black and a “false alarm” was defined by White targets incorrectly categorized as Black.

R

ESULTS EXPERIMENT

2

M

AIN ANALYSIS

R

EACTION TIME

A repeated-measures ANOVA (2 x 2 model) was used on mean RTs in JASP, with prime congruency and presentation time as within-subject factors. Congruency main effects were not found to be significant (F(1, 57)<.001, p=1.00): congruent trials were not responded to faster than incongruent trials (see fig. 6). A main effect of presentation time could be observed (F(1, 57)=12.00, p=.0010), with significant increase of RT for SP (M=642.10ms, SD=452.95) opposed to LP targets (M=486.60ms, SD=528.00). Target presentation time and congruency did not significantly interact with each other (F(1, 57)=.11, p=.74). We looked at simple main effects to test significant interactions. No significant effect of congruency in either SP (F=0.80, p=.78) or LP (F(=.10, p=.75) was found. The presentation time effect was found within both congruent and incongruent trials (F=12.93, p>.001; F=6.60, p=.013) (see fig. 7).

Fig.6: mean reaction times of individual subjects per congruency condition (LP and SP) in experiment 2. No clear differences were seen between congruent or incongruent trials, with the means remaining somewhat similar within each subject.

Reaction time: Interaction effects - Presentation Level of congruence Sum of Squares df Mean Square F p Con 683266 1 683266 12.929 < .001 Incon 547840 1 547840 6.607 0.013

Note. Type III Sum of Squares

Fig. 7: Interaction effects of presentation time within congruent and incongruent trials, both showing significant increases of RT for short presentation (M=642.10ms, SD=452.95) compared to long presentation (M=486.60ms, SD=528.00) in experiment 2.

A

CCURACY RATE

Mean accuracy rates are submitted to a repeated-measures ANOVA with within-subject factors prime congruency and presentation time. The same repeated measures analysis performed on accuracy rates revealed significant effects of presentation time (F(1,57)=969.47, p<.001) but not of

(M(LP)=.98, SD=.037; M(SP)=.59, SD=.10) (see fig. 8) and congruent trials were not responded to more accurate than incongruent trials. Target presentation time and congruency did [not] significantly interact with each other (F(1, 58)=.93, p = .34). Again, simple main effects were analyzed to observe interactions and significant effects of presentation time within congruent (F=844.60, p<.001) as well as incongruent trials (F=790.20, p<.001) were found. Congruency effects were significant within neither SP nor LP trials (F=.73, p=.40; F=.37, p=.54).

In figure 9 the cumulative accuracy rates obtained of all subjects per condition shows that there is a clear difference between short and long presentation trials with (congruent and incongruent) SP trials showing lower but relatively distributed accuracy rates. LP trials lean mainly towards a high cumulative frequency for high accuracy rates, showing that participants were more likely to respond accurately in an invariable manner. No clear distinction between congruent and incongruent trials for both long and short presentation conditions in the cumulative accuracy rates can be observed visible in the similar lines of con-SP/incon-SP and con-LP/incon-LP, in line with the lack of congruency (interaction) effects found.

Fig.9: the cumulative frequency of the accuracy rates obtained in experiment 2 for all four conditions. No difference between congruent conditions, but accuracy distribution of all participants differs with generally more evenly accurate means on LP-trials.

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 0 . 3 5 0 . 4 5 0 . 5 5 0 . 6 5 0 . 7 5 0 . 8 5 0 . 9 5 CUMULATIVE FREQUENCY ACCURACY RATE

ACCURACY RATE: CUMULATIVE FREQUENCY PER CONDITION

Con-SP Con-LP Incon-SP Incon-LP

Fig. 8: Presentation effects in mean accuracy rates in Experiment 2. Each colored line represents an individual subject, mean SP to mean LP. A clear significant difference between accuracy rates on SP trials versus LP trials is seen, with mean rates being higher on LP trials (mean difference = 0.39).

T

ARGET AWARENESS

For target awareness, discrimination accuracy was tested against chance using a binomial test with 160 trials and chance level of 50% (= 0.5) correct, which indicated that target discrimination above 58% (= 0.58) correct was significantly above chance level (one-tailed). We performed a t-test and showed that sensitivity was significantly better than chance (M=0.79, SD=0.76; t(58)=8.00, p< .001 (one-tailed), dz=1.04), demonstrating that masking of targets in experiment 2, again, failed to render the (long presented) faces objectively invisible.

C

ONCLUSION

2

In experiment 2 again no priming effects were found. No improvement of face localization is seen with congruent gun-Black or tool-White prime-target sets, giving no clear reason to assume that early unconscious face processing can be semantically influenced through priming. Nonetheless, effects of presentation duration were found with longer presentation duration trials showing increased accuracy rates within congruent and incongruent conditions. Potential performance trade-offs were checked and comparing presentation effects on RT and accuracy showed short presentation increases reaction time, meanwhile accuracy is decreased. The opposite is visible for longer presentation, where responses are faster and more accurate. This showed no trade-off, but a clear overall performance increase for longer presented targets.

G

ENERAL DISCUSSION

In both our experiments people appeared to be more aware of, and thus better at identifying, longer presented faces. With short-presented faces identification remains at chance-level, independently of what prime is used. This shows us that masking is effective at rendering faces objectively invisible with short but not for longer presentation durations. Not surprisingly the presentation duration influenced localization tasks, since the target was more visible in long presentation conditions. With short-presented, effectively primed, targets we did not find any priming effects as expected based on subliminal priming paradigms(Kiesel et al., 2007). In our first experiment we looked at whether valanced context could influence early perceptional processing of emotional faces. A contextual factor was created using linguistic priming and face localization tasks were performed with masked happy and angry expressional faces. We hypothesized that valanced sentences could be constructive contextual primes by activating semantic knowledge about the emotional event and would serve as facilitation of detecting the masked emotional faces. Hence, we expected to find that congruent priming would result in more accurate and faster responses, but this was not observed: we could not conclude that congruency influenced task performance. Presentation time did influence performance with longer presentation resulting in increased accuracy without compromising speed. The second experiment focused mainly on bias-influence on unconscious perception using the Black-gun association. Images of guns or tools served as primes, and a masked face localization task with Black and White male and female faces were used in a similar experiment. We expected to find congruent priming (e.g., gun-prime with Black-target or tool-prime with White-target) would result in better performance on face localization. Congruency, again, did not influence speed or accuracy; however, longer presentation did show positive effects on both. Concludingly, we observe that people do not improve in detecting masked faces after seeing semantically related primes. These results suggest that early unconscious perception either does not take place or is not influenced by such contextual factors.

There were other considerable factors to these results. For instance, the lack of statistical power in experiment 1: our power analysis showed a desired sample size of 54 participants, but this was not reached in the limited publishing time. Additionally, in pre-analyzation we did not remove participants with mean accuracy below a percentage or reaction times above a limit as is done in earlier studies to prevent influence of extremities on statistical results (Van den Bussche, Notebaert, & Reynvoet, 2009a). Furthermore, contextual priming should be questioned as the valanced

sentences used in this experiment did not have direct semantic relations to target-face expressions and stronger evidence of unconscious priming with closely related prime-target sets is previously found (Van den Bussche, Smets, Sasanguie, & Reynvoet, 2012). A lack of specific emotional experience with the contextual information provided in the sentences might also result in insufficient priming (Fernández-Dols, Wallbott, & Sanchez, 1991). Alternatively, more evocative evidence of semantic word processing has been found, indicating that to increase semantic priming sensitivity, the use of highly related words should be considered (Gendron et al., 2012). Another factor could be the relatively small prime sets used in experiment 2 of only six distinctive objects per category, enabling formation of expectation of stimuli and it is possible that attempts at stimuli-response mapping influenced potential priming effects in longer presentation trials (Damian, 2001). Nevertheless, it was also shown that target set size is not a significant factor for effective subliminal semantic processing (Van den Bussche & Reynvoet, 2007). Next to this, lack of priming effects in our second experiment could also have been due to visual processing of pictures of objects. As mentioned in Stein et al. (2020) it could be that pictures are not adequate for researching unconscious processes due to the complexity of visual analysis.

Moreover, extensive evidence has found priming effects by studying subliminal primes on subsequent target processing (Van den Bussche, Van den Noortgate, & Reynvoet, 2009). Research on influence of especially (subliminally shown) faces and their characteristics on subsequent contextual perception has proposed promising results (Axelrod et al., 2015; Kiefer & Brendel, 2006; Van den Bussche, Notebaert, & Reynvoet, 2009b). Applying the same masking paradigm on investigating the influence on subliminal face-targets after visible context-priming did not deliver similar results - possibly due to differential processing of prime versus target. We should keep in mind that our subliminal target presentation might have been inadequate, given that invisible primes are passively attended, therefore processing might differ from active, attentive target-processing (Barbot & Kouider, 2012). Also, in masked priming studies it was seen that an increased stimulus onset asynchrony (SOA, i.e., the time interval between prime onset and target onset) influences semantic priming effects negatively (Kiefer & Brendel, 2006). Given our design with elongated prime presentation the SOA is notably increased and this potentially abolished priming effects in the current study. More research should be done on masked target paradigms of this structure. Lastly, earlier studies using face localization tasks did in fact find suggestive evidence of higher-level unconscious processing using CFS (i.e., masking-method where a target is projected to one eye while constantly changing pattern-mask are projected to the other) and, even though this evidence was not conclusive, that it could serve as sensitive method to study differences of awareness accessing between stimuli (Stein et al., 2020). With optimized presentation durations, replicating our experiments with the CFS-method might be insightful and could provide more suitable masking of face-targets.

Overall, we failed to collect the expected evidence for influence of context on unconscious processing of invisible emotional as well as racial faces with semantically related sentences and objects as prime. Backward masking provided adequate masking in short presentation trials, but whether the subliminal targets were shown effectively for enabling unconscious processing is unclear, since no priming effects were found. Current findings do not contribute to theories of early high-level visual processing, in line with earlier findings that failed to find evidence for such assumptions and insight in unconscious processing is still rather limited (Izatt et al., 2014; Stein et al., 2011). Notably, it has been previously found that unconscious processing of facial aspects such as race and gender is more complex since it requires higher level discrimination and cannot be processed outside awareness (Amihai, Deouell, & Bentin, 2011). It has been repeatedly shown that social cognitive functions and multisensory integration can clearly be processed unconsciously (Axelrod et al., 2015). The influence of context on unconscious perceptional processing of faces, however, remains a relatively untouched and valuable field of study to look into.

References

Adolphs, R. (2002). Recognizing emotion from facial expressions: Psychological and neurological mechanisms. Behavioral and Cognitive Neuroscience Reviews, 1(1), 21-62.

Almeida, J., Pajtas, P. E., Mahon, B. Z., Nakayama, K., & Caramazza, A. (2013). Affect of the unconscious: Visually suppressed angry faces modulate our decisions. Cognitive, Affective, &

Behavioral Neuroscience, 13(1), 94-101.

Amihai, I., Deouell, L., & Bentin, S. (2011). Conscious awareness is necessary for processing race and gender information from faces. Consciousness and Cognition, 20(2), 269-279.

Axelrod, V., Bar, M., & Rees, G. (2015). Exploring the unconscious using faces. Trends in Cognitive

Sciences, 19(1), 35-45.

Barbot, A., & Kouider, S. (2012). Longer is not better: Nonconscious overstimulation reverses priming influences under interocular suppression. Attention, Perception, & Psychophysics, 74(1), 174-184.

Barrett, L. F., & Kensinger, E. A. (2010). Context is routinely encoded during emotion perception.

Psychological Science, 21(4), 595-599.

Barrett, L. F., Lindquist, K. A., & Gendron, M. (2007). Language as context for the perception of emotion. Trends in Cognitive Sciences, 11(8), 327-332.

Barrett, L. F., Mesquita, B., & Gendron, M. (2011). Context in emotion perception. Current Directions

in Psychological Science, 20(5), 286-290.

Cameron, C. D., Brown-Iannuzzi, J. L., & Payne, B. K. (2012). Sequential priming measures of implicit social cognition: A meta-analysis of associations with behavior and explicit attitudes. Personality

Chen, J., Mizuno, A., Lyew, T., Karim, H. T., Karp, J. F., Dombrovski, A. Y., et al. (2020). Naltrexone modulates contextual processing in depression. Neuropsychopharmacology, 45(12), 2070-2078.

Cunningham, W. A., Johnson, M. K., Raye, C. L., Gatenby, J. C., Gore, J. C., & Banaji, M. R. (2004). Separable neural components in the processing of black and white faces. Psychological Science,

15(12), 806-813.

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(1), 154.

de Gelder, B., Meeren, H. K., Righart, R., Van den Stock, J., Van de Riet, Wim AC, & Tamietto, M. (2006). Beyond the face: Exploring rapid influences of context on face processing. Progress in

Brain Research, 155, 37-48.

Dimberg, U., Thunberg, M., & Elmehed, K. (2000). Unconscious facial reactions to emotional facial expressions. Psychological Science, 11(1), 86-89.

Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995). Variability in automatic activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of Personality and

Social Psychology, 69(6), 1013.

Fernández-Dols, J., Wallbott, H., & Sanchez, F. (1991). Emotion category accessibility and the decoding of emotion from facial expression and context. Journal of Nonverbal Behavior, 15(2), 107-123.

Gendron, M., Lindquist, K. A., Barsalou, L., & Barrett, L. F. (2012). Emotion words shape emotion percepts. Emotion, 12(2), 314.

Haxby, J. V., Hoffman, E. A., & Gobbini, M. I. (2000). The distributed human neural system for face perception. Trends in Cognitive Sciences, 4(6), 223-233.

Hugenberg, K., & Bodenhausen, G. V. (2004). Ambiguity in social categorization: The role of prejudice and facial affect in race categorization. Psychological Science, 15(5), 342-345.

Izatt, G., Dubois, J., Faivre, N., & Koch, C. (2014). A direct comparison of unconscious face processing under masking and interocular suppression. Frontiers in Psychology, 5, 659.

Kiefer, M., & Brendel, D. (2006). Attentional modulation of unconscious “automatic” processes: Evidence from event-related potentials in a masked priming paradigm. Journal of Cognitive

Neuroscience, 18(2), 184-198.

Kiesel, A., Kunde, W., & Hoffmann, J. (2007). Mechanisms of subliminal response priming. Advances

in Cognitive Psychology, 3(1-2), 307.

Lehmann, C., Mueller, T., Federspiel, A., Hubl, D., Schroth, G., Huber, O., et al. (2004). Dissociation between overt and unconscious face processing in fusiform face area. NeuroImage, 21(1), 75-83.

Lindquist, K. A., & Gendron, M. (2013). What’s in a word? language constructs emotion perception.

Emotion Review, 5(1), 66-71.

Mekawi, Y., & Bresin, K. (2015). Is the evidence from racial bias shooting task studies a smoking gun? results from a meta-analysis. Journal of Experimental Social Psychology, 61, 120-130.

Murphy, S. T., & Zajonc, R. B. (1993). Affect, cognition, and awareness: Affective priming with optimal and suboptimal stimulus exposures. Journal of Personality and Social Psychology, 64(5), 723.

Niedenthal, P. M. (1990). Implicit perception of affective information. Journal of Experimental Social

Palermo, R., & Rhodes, G. (2007). Are you always on my mind? A review of how face perception and attention interact. Neuropsychologia, 45(1), 75-92.

Payne, B. K. (2001). Prejudice and perception: The role of automatic and controlled processes in misperceiving a weapon. Journal of Personality and Social Psychology, 81(2), 181.

Preston, S. D., & Stansfield, R. B. (2008). I know how you feel: Task-irrelevant facial expressions are spontaneously processed at a semantic level. Cognitive, Affective, & Behavioral Neuroscience,

8(1), 54-64.

Rotteveel, M., de Groot, P., Geutskens, A., & Phaf, R. H. (2001). Stronger suboptimal than optimal affective priming? Emotion, 1(4), 348.

Stein, T., Ciorli, T., & Otten, M. (2020). Guns are not faster to enter awareness after seeing a black face: Absence of race-priming in a gun/tool task during continuous flash suppression.

Stein, T., Hebart, M. N., & Sterzer, P. (2011). Breaking continuous flash suppression: A new measure of unconscious processing during interocular suppression? Frontiers in Human Neuroscience, 5, 167.

Stein, T., Utz, V., & Van Opstal, F. (2020). Unconscious semantic priming from pictures under backward masking and continuous flash suppression. Consciousness and Cognition, 78, 102864.

Stewart, L. H., Ajina, S., Getov, S., Bahrami, B., Todorov, A., & Rees, G. (2012). Unconscious evaluation of faces on social dimensions. Journal of Experimental Psychology: General, 141(4), 715.

Tamietto, M., & De Gelder, B. (2010). Neural bases of the non-conscious perception of emotional signals. Nature Reviews Neuroscience, 11(10), 697-709.

Tian, X., & Huber, D. E. (2010). Testing an associative account of semantic satiation. Cognitive

Psychology, 60(4), 267-290.

Van den Bussche, E., Notebaert, K., & Reynvoet, B. (2009a). Masked primes can be genuinely semantically processed: A picture prime study. Experimental Psychology, 56(5), 295-300.

Van den Bussche, E., Notebaert, K., & Reynvoet, B. (2009b). Masked primes can be genuinely semantically processed: A picture prime study. Experimental Psychology, 56(5), 295-300.

Van den Bussche, E., & Reynvoet, B. (2007). Masked priming effects in semantic categorization are independent of category size. Experimental Psychology, 54(3), 225-235.

Van den Bussche, E., Smets, K., Sasanguie, D., & Reynvoet, B. (2012). The power of unconscious semantic prcessing: The effect of semantic relatedness between prime and target on subliminal priming. Psychologica Belgica, 52(1), 59-70.

Van den Bussche, E., Van den Noortgate, W., & Reynvoet, B. (2009). Mechanisms of masked priming: A meta-analysis. Psychological Bulletin, 135(3), 452.

Wieser, M. J., & Brosch, T. (2012). Faces in context: A review and systematization of contextual influences on affective face processing. Frontiers in Psychology, 3, 471.