Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

## Introduction

The role of attention in modulating perceptual processes has been a long-standing topic in Psychology and Neuroscience. As attention operates on distinct stimulus/perceptual/visual properties, attention can be described as a collection of specialized mechanisms operating on the basis of stimulus properties such as space, time, features, etc. (see a review1). On the other hand, the load theory of attention describes attention as a more general processing resource that spans across specialized perceptual mechanisms and considers how the availability of such fluid resources could gate perceptual processing2,3. Whether attention is treated as something which is directed to a specific entity or as something utilized as general cognitive resources, studies have established the benefits of attention in various tasks over the past decades. For instance, when an identical feature (e.g. motion direction) is shared between two sets of moving dots, behavioral performance on comparing the speed of the two improves4. This is interpreted as feature-based attention enhancing performance on the feature-sharing stimuli. Similarly, orienting attention to a specific location improves visual performance5 and even boosts stimulus contrast6. These findings suggest that attentional modulation on visual information begins very early in the visual pathway and affects rudimentary visual features. Indeed, neurophysiological evidence has shown that attentional modulation begins early both anatomically (primary visual cortex (V1)7; lateral geniculate nucleus (LGN)8,9) and temporally (C1/P1 ERP component10,11: 100 ms after stimulus onset).

Attention not only benefits the processing of conscious stimuli, it has also been shown that attention operates near the border of consciousness. In an example of feature-based attention extending to the processing of near-threshold stimuli, Rossi and Paradiso12 found that observers who were focused on a foveal Gabor patch would detect a near-threshold peripheral grating at higher rates if the two had similar orientations and spatial frequencies. In an experiment treating attention as a fluid resource, Cartwright-Finch and Lavie13 showed that a peripheral distractor was more likely to enter conscious awareness and be detected when the observer was under a condition of low perceptual load, compared to a high-perceptual-load condition. Finally, in the phenomenon of inattentional blindness, focusing attention on a specific aspect of a visual scene prevents the conscious awareness of salient stimuli, even if they are presented foveally14,15. These studies have established that attention operates at the boundary of consciousness and may serve as a gating system determining whether visual stimuli enter into conscious awareness. Clearly, attention is tightly intertwined with consciousness both from our subjective experiences and experimental findings.

The next question is whether attention operates on unconscious visual content. Studies addressing this question provide a critical window for examining the relationship between the two. Are attention and consciousness one and the same, or can attention still modulate our visual system in the absence of conscious awareness? More and more findings suggest attentional modulation on subliminal stimuli. For example, both Bahrami et al.16 and Kanai et al.17 showed orientation adaptation from interocularly suppressed subliminal orientation. As in the previous examples, the unconscious effect was modulated by either feature-based attention elicited from attending a visible orientation17 or the perceptual load induced by a concurrent central task16. Further, Bahrami et al.16 showed that the availability of attention strengthened primary visual cortex signal for a subliminal distractor. Finally, Hsieh et al.18 showed that subliminal singletons elicited a location-specific cuing effect, prompting higher accuracy on a subsequent visual task. The effect disappeared when participants were instructed to perform a demanding concurrent task. These studies together suggest that deployment of attention to the unconscious stimuli enhances unconscious processes while depletion of attention drains the resources necessary for unconscious processes. That is, (1) under a high-load condition or (2) when attention is directed away from unconscious stimuli, the effect from such stimuli either weakens or disappears.

However, past studies showing that attention operates on subliminal visual content have been constrained to low-level visual features such as simple orientation and stimulus saliency (low-level visual processing refers to light-based, retinotopic, early stages simple feature processing, such as contrast, orientation and color (e.g.19), while high-level visual processing refers to categorical or semantic extraction of visual input, which is relatively invariant to the viewing conditions such as luminance and angles). These past studies thus suggest that, unlike conscious processes, the attentional effect on subliminal stimuli may be confined in the early visual cortex and to simple low-level features, leading to a dichotomy of attentional modulation on conscious and unconscious stimuli. On the other hand, it remains possible that attention may modulate subliminal visual content from early to late stages, leading to attentional modulation on a wide spectrum of subliminal visual information. That is to say, whether attentional modulation operates on the high-level processing of subliminal stimuli remains largely unknown. The interaction between attention and subliminal low-level/high-level features also tackles an important question: the robustness of unconscious effects in the scientific literature. The existence of certain high-level unconscious processes has been shown sporadically, often without stable replication. One possibility would be that the strength of subliminal high-level visual information is inherently weaker, and more attentional resources have to be directed to such information in order to maintain a stable representation and to give rise to an experimental effect.

## Results

### Conscious task demands gated unconscious processes

To first establish that the suppressed prime was invisible, we examined the accuracy of the location task in Experiment 1 (word-naming). The mean accuracy was 48.86% (2.17%) and not different from chance (paired t(19) = −0.53, p = 0.61), indicating that the suppression was successful. Moreover, the performance on the blank and visible trials was 94.06% (1.40%) and 97.19% (1.15%), respectively, showing that the participants responded with high accuracy and consistency. A similar pattern was found in Experiment 2. The mean accuracy on the location task was 50.07% (1.69%) and not different from chance rate (paired t(19) = 0.04, p = 0.97). Moreover, the performance on the blank and visible trials on the detection task was 97.81% (1.04%) and 98.44% (0.77%), respectively, showing high accuracy and consistency. Table 1 summarizes these objective measures in all experiments.

To examine whether prime-target word and color congruency affected target responses, we put in three different factors into our analysis: prime-target color congruency, prime-target word congruency, and target word-color congruency (that is, whether target is a Stroop word (word-color incongruent) or a non-Stroop word (word-color congruent)). In this article, we refer to prime-target color congruency as color congruency, prime-target word congruency as word congruency, and target word-color congruency as target Stroop/Non-Stroop throughout the manuscript. A three-way repeated measures analysis of variance was performed on the target word reaction time (please note that in all experiments the average accuracy on the word or color-naming task was near ceiling (all > 97%). Hence the analyses were focused on the reaction time). In Experiment 1, the main effect of target (reverse) Stroop was found, F(1, 19) = 37.00, p = 0.0000, ηp2 = 0.66, with a main effect of color (F(1, 19) = 7.63, p = 0.01, ηp2 = 0.29) and but not word congruency (F(1, 19) = 0.73, p = 0.40, ηp2 = 0.04). Furthermore, there was an three-way interaction between target Stroop, word and color congruency, F(1, 19) = 5.81, p = 0.03, ηp2 = 0.23. Post hoc comparisons showed that double word-color incongruency significantly slowed down target response only when the target was not a Stroop word (paired t(19) = −2.43, p = 0.02, marginally significant after correction), but not when it was a Stroop word (paired t(19) = −1.96, p = 0.06) (Fig. 2a, b).

The same analysis was performed on the data of Experiment 2 (color-naming). The main effect of target Stroop was also found, F(1, 19) = 56.01, p = 0.0000, ηp2 = 0.75. However, no further main effects or interactions from the prime-target congruency were found (word: F(1, 19) = 1.36, p = 0.26, ηp2 = 0.07, Fig. 2c; color: F(1, 19) = 2.84, p = 0.11, ηp2 = 0.13). Experiments 1 and 2 thus showed that the same set of unconscious stimuli could exert an interfering effect in one context but not another, even with identical time sequence of stimulus delivery and the subsequent influenced target.

### Replication of unconscious word-induced interference

Our first two experiments clearly showed that task-induced attentional load modulated the extent to which unconscious stimuli exerted an interfering effect. Experiment 1 showed word-induced and color-induced interference between a suppressed prime and a visible target slowed down the response time to the non-Stroop target. Critically, such interference disappeared when the task was of high load (i.e. color-naming) in Experiment 2. Surprisingly, a further analysis separating early and late trials revealed word-induced but not color-induced interference in the later trials, suggesting that word-induced semantic interference may be more resilient to the current conscious task demands. These findings showed an asymmetry of attentional modulation on prime interferences. To further replicate these findings with a cleaner design, we isolated the word and color components in the later experiments.

In Experiments 3 and 4 we focused only on the word aspect and aimed to re-examine and replicate the word-induced semantic incongruency effect. The experiments had identical trial sequence and design as Experiments 1 and 2, except for two changes. (1) The prime words were made colorless. (2) Half of the primes were made blank. These blank trials later served as our baseline to compare against and allowed us to calculate the percentage reaction time changes between prime absent and present trials.

In Experiment 3 (word naming), the reverse Stroop effect in the target responses was evident (Stroop and non-Stroop trials: t(19) = 4.21, p = 0.00, Cohen’s dav = 0.35). In Experiment 4, Stroop effect was found (Stroop and non-Stroop trials: t(19) = 6.25, p = 0.00, Cohen’s dav = 0.58).

To isolate the effects of congruency/incongruency, prior to comparing between the congruent and incongruent conditions, we first calculated target response time change with against without the prime. Therefore, the following results all appear in reaction time percentage changes from the baseline blank trials. More general ANOVA results in the style of Experiments 1 and 2 are included in supplementary information (Supplementary Note 2). In Experiment 3, a paired t test directly compared between semantically incongruent vs. congruent trials after normalizing against the blank trials. A significant word interference effect was found (t(19) = 2.43, p = 0.03, Cohen’s dav = 0.65, Fig. 3a), showing that when the invisible prime and visible target were incongruent, there was a slowing effect on target response of 4.74% (compared to blank trials: word congruent trials: 0.62% slower, word incongruent trials: 5.36% slower).

In Experiment 4 (color naming), a paired t test directly compared semantically incongruent and congruent trials after normalizing against the blank trials showed null results in overall trials (t(19) = 1.54, p = 0.14, Fig. 3b). Similar to our analysis in Experiment 2, we split the data into 1st quarter and 4th quarter trials. A significant word interference effect was found only in the 4th quarter trials (t(19) = 2.30, p = 0.03, Cohen’s dav = 0.57, Fig. 3d) but not in the 1st quarter trials (t(19) = 0.16, p = 0.87, Cohen’s dav = 0.03, Fig. 3c). This result again showed that when the task load decreased in a high-load task (color-naming), the incongruency between an invisible prime and a visible target slowed down target response of 7.29% (compared to blank trials: semantically congruent trials: 1.35% slower, semantically incongruent trials: 8.64% slower).

The results of Experiments 3 and 4 replicated what we found in Experiments 1 and 2, showing that semantic interference from an unconscious incongruent word exerted a slowing effect on the subsequent target response. Importantly, this effect was modulated by the conscious task demands. Similarly, the late emerging semantic interference effect was again found under a more demanding color-naming task, when extended practice had reduced the task load.

### No conclusive unconscious color interference

In the next two experiments, we re-examined the color interference between the invisible prime and the target, which was evident in Experiment 1 with a three-way interaction with word interference and target Stroop but disappeared in Experiment 2. Experiments 5 and 6 set out to examine whether color congruency alone exhibits an interference effect on target response. In these experiments, the primes were colored symbols (i.e. XXXX in blue or red) to cleanly isolate and test the effect of color in the current paradigm.

The reverse Stroop effect and Stroop effects were also evident in Experiments 5 (Stroop and non-Stroop trials: t(19) = 4.97, p = 0.00, Cohen’s dav = 0.32) and 6 (Stroop and non-Stroop trials: t(19) = 6.80, p = 0.00, Cohen’s dav = 0.78).

In Experiment 5, the RTs in each condition were first normalized against blank trials. No slowing effect was found in the color incongruent trials with t(19) = 0.7, p = 0.49, Cohen’s dav = 0.21, Fig. 4a. In Experiment 6, similar to Experiment 2, no color incongruency effect was found in all trials (t(19) = 0.28, p = 0.78, Cohen’s dav = 0.17 Fig. 4b) with both conditions showing a slowing effect on target response (compared to blank trials: color congruent trials: 4.23 % slower, color incongruent trials: 4.72 % slower). Such null effect was found in the 1st and 4th quarter trials: (1st: t(19) = 0.02, p = 0.98,, Cohen’s dav = 0.01, Fig. 4c, 4th: t(19) = 0.26, p = 0.80, Cohen’s dav = 0.07, Fig. 4d) with both conditions showing a slowing effect on target response (compared to blank trials: incongruent vs. congruent trials 1st Q: 4.70 vs. 4.64%; 4th Q: 5.84 vs. 5.10% slower).

These results indicate that non-semantic perceptual stimulus congruency (i.e. color) under interocular suppression did not conclusive yield an interference effect on target response in the current paradigm.

The critical difference was that the reverse Stroop effect was evident in Experiment 7 (Stroop and non-Stroop trials: t(19) = 3.97, p = 0.00, Cohen’s dav = 0.32) while the Stroop effect was not significant in Experiment 8 (Stroop and non-Stroop trials: t(19) = 2.04, p = 0.06, Cohen’s dav = 0.16).

In Experiment 7, the RTs in each condition were first normalized against blank trials. A direct paired comparison between semantically incongruent and congruent trials showed null effect (t(19) = −0.82, p = 0.42, Cohen’s dav = 0.19, Fig. 5a). On average, compared to the blank trials, semantic congruent trials were 2.1% slower while incongruent ones 0.9% were slower. We also examined whether prime-target location (same-different) had interacted with the semantic congruency effect. A two-way (semantic congruency; prime-target location consistency) repeated measures analysis of variance was performed on the normalized target word reaction time. Neither the main effect of semantic congruency, F(1, 19) = 0.09, p = 0.77, ηp2 = 0.005, nor the main effect of location congruency was found, F(1, 19) = 1.32, p = 0.26, ηp2 = 0.07. There was no interaction between the two, F(1, 19) = 0.21, p = 0.65, ηp2 = 0.01. Similar results were found in the 4th quarter trials (direct comparison: t(19) = 1.21, p = 0.24, Cohen’s dav = 0.33; semantic congruency × location consistency 2-way ANOVA: main effect of semantic congruency F(1, 19) = 0.10, p = 0.75, ηp2 = 0.005; main effect of location: F(1, 19) = 0.14, p = 0.72, ηp2 = 0.007; interaction: F(1, 19) = 1.60, p = 0.22, ηp2 = 0.08).

In Experiment 8, a direct paired comparison on the normalized RT between semantically congruent and incongruent trials showed null effect (t(19) = −1.12, p = 0.28, Cohen’s dav = 0.24). On average, compared to the blank trials, semantic congruent trials were 1.32% slower while incongruent ones 2.7% were slower. However, a two-way (semantic congruency; prime-target location consistency) repeated measures analysis of variance showed an interaction between the prime-target location and semantic congruency with F(1, 19) = 8.13, p = 0.01, ηp2 = 0.30. The main effects were not significant (semantic congruency: F(1, 19) = 1.1, p = 0.29, ηp2 = 0.06; location congruency: F(1, 19) = 0.05, p = 0.83, ηp2 = 0.003). A planned post hoc comparison between semantic congruent and incongruent trials when the prime-target were co-localized showed a significant effect with t(19) = −2.45, p = 0.02, Cohen’s dav = 0.63, Fig. 5b). When the prime and target were co-localized, semantic congruent trials were 0.98% slower while incongruent ones were 5.36% slower.

### General discussion

Two important questions were left unanswered after our first four experiments. Firstly, although the general predictions of the load theory are compatible with our findings, a feature-based explanation is also possible. Task-induced attention might selectively trigger different cognitive sets under different tasks, which in turn would gate how an unconscious prime interfered with subsequent target performance. For example, in the word-naming experiments (Experiments 1 and 3), participants’ attention was selectively oriented to the word semantics. Such attention tuning would have been applied further to the suppressed prime, allowing semantic interferences to occur. In contrast, in the color-naming experiments, attention was deployed to the word color, which would have interfered the semantic processing of the target (i.e. the Stroop effect) as well as hindered the semantic interferences from the suppressed prime. Since our interfering effects were semantic in nature, word-induced semantic incongruency effects could have been gated by this cognitive-set-induced-tuning to different levels: an incongruent word interference was evident in word-naming experiments; while under color-naming task, such effect only re-emerged after the participants became more fluent on the task (i.e. 4th quarter trials in Experiments 2 and 4). We also provide a discussion regarding the classical negative priming in the Stroop paradigm in Supplementary Note 3. Secondly, one could argue that since identical word-forms were used as primes and targets, what appeared to be a semantic incongruency effect could be driven merely by orthographic dissimilarity. The initial experiments did partially address this by jittering the prime-target locations and font sizes, and our further analyses focusing on the interaction between the semantic effect and location yielded null effect, which excluded the possibility of low-level adaptation (Supplementary Note 4). Nevertheless, this does not distinguish between orthography and semantics.

The relationship between attention and consciousness has been a heated debate involving researchers in Neuroscience, Psychology, and Philosophy. Koch and Tsuchiya24 posit that attention and consciousness are doubly dissociable, that is, one can show conditions where attention is summoned but consciousness is yet to emerge and vice versa. In contrast, Cohen et al. propose that attention gates consciousness, arguing that a stimulus/scene/object only enters our consciousness when some amount of attention is deployed to the item25. Thus, according to this attention gating theory, there is a causal relationship between attention and consciousness. Our data exhibit attention without consciousness, showing task-induced attentional modulation on unconscious processes. Together with previous evidence showing that bottom-up attention can be directed to an unconscious stimulus (e.g. salient singleton18; random motion26; attractive face27), our results further show that task-induced top-down attention constraints an interfering effect elicited by an unconscious stimulus. However, this interpretation is not necessarily at odds with the attention gating consciousness theory as these results can be seen as an expansion of the attention gating system from the realm of consciousness to unconsciousness, which is previously acknowledged25.

Since the invention of continuous flash suppression28, whether high-level subliminal information survives strong interocular suppression has been a matter of debate. Jiang et al.29 reported familiarity effect from both face and linguistic stimuli, showing that upright faces broke through interocular suppression faster than inverted faces. Similar faster breaking time was shown between words from one’s own native language and words from a foreign language. More directly, Costello et al.30 showed that a word broke through suppression faster if the preceding word was semantically or orthographically related, compared to the condition in which the suppressed word was preceded by an unrelated word. Recently, Hung and Hsieh31 showed that subsequent to a setential context, syntactically incongruent words break through interocular suppression faster than the congruent counterparts. However, these studies potentially suffer from the disadvantage of the breaking-suppression paradigm32, unable to distinguish a pure unconscious effect from an access-to-consciousness effect. That is, as breaking-suppression relies on the conscious detection of a suppressed stimulus, it is thus hard to tease apart the actual origin of the effect. Current study provided a cleaner paradigm in which the interference effect was assessed by how a precedent suppressed word (prime) affected the following target response even when the participants had not broken suppression and performed at chance on localizing the suppressed stimulus. Therefore, the semantic interference from an interocularly suppressed word provided clear evidence for high-level subliminal semantic processing. Moreover, as our findings show that the attentional requirement of a task serves as a gating mechanism to an unconscious effect, it is worth reconsidering the inconsistent results from recent interocular suppression studies under this framework. Specifically, subliminal stimuli very often suffer from poor stimulus signal-to-noise ratio, even more so in the interocular paradigm as experimenters actively suppress the visibility of a dim and static stimulus. It is thus difficult to distinguish a true null finding where a subliminal stimulus does not intrinsically elicit an effect from a more trivial explanation: the underpowered nature of subliminal stimuli. Here we provide evidence for another critical factor: the necessity of attention.

The current findings shed light on the attentional gating of subliminal information, expanding the capacity limitations of attention outside the realm of consciousness. More critically, we show that such attentional modulation occurs with high-level semantic information, opening up the possibility that the attentional load of concurrent task modulates a wide spectrum of information processing in the absence of consciousness. If attention is a limited resource that is shared and competed for, not only by what we are conscious of but also what we are not, the unconscious processes are potentially constrained by our conscious deliberations. Real-world implicit influences in the surroundings are thus gated by how we distribute our attention, indicating a mechanism in which unconscious contents can be consciously and voluntarily enhanced or weakened.

## Methods

### General experimental apparatus

In all experiments, the visual stimuli were generated with MATLAB (The MathWorks, Inc., Natick, MA) and PsychToolbox36,37. Participants viewed the dichoptic images through a mirror stereoscope and rested on a chin rest, from a distance of 42 cm. The stimuli were presented against a black background on a 30-in. Apple M9179LL/A LCD monitor with a resolution of 2560 × 1600 pixels and a refresh rate of 60 Hz. Throughout the experiment, a white frame (subtending 5.4° × 5.4°) remained on-screen to facilitate proper fusion.

### Participants

Throughout all experiments, all participants (age range: 18–36) reported normal or corrected-to-normal vision. They reported no history of language deficits and were proficient in English. They gave written informed consent prior to the experiment and were reimbursed \$15 for participating in a 60-min session. This study was approved by the institutional review board of the California Institute of Technology. Participants that (1) had performance 3 standard deviations away from the group mean on the catch tasks or/and (2) failed to achieve successful calibration on prime luminance or/and (3) brokethrough suppression yet failed to indicate prime location (see below experimental design and procedure) were removed before entering analysis. (see Stimuli and Procedure of each experiment; Experiment 1: n = 3; Experiment 2: n = 2; Experiment 3: n = 1; Experiment 4: n = 2; Experiment 5: n = 0; Experiment 6: n = 0; Experiment 7: n = 1; Experiment 8: 2). A total number of 20 participants was targeted in each experiment based on our 80% power calculation in our previous study31. All participants were naïve to the purpose of the experiments.

### Reaction time data pre-analysis processing

All trials that had reaction time longer than 5 s or shorter than 500 ms were pre-excluded. Furthermore, the reaction time data underwent per-participant per-condition outlier removal to remove data points 3 standard deviations away from the average.

### Experiments 1 and 2: experimental design and procedure

Prior to the experiment, participants’ eye dominancy was determined by the Miles test38. Two words and two colors were selected to create word-color (in)congruency in the prime and target: word BLUE or RED in color blue or red. Each trial began with a blank screen lasting for a varied SOA ranging from 0.1 to 1 s. After which the dominant eye received a series of colorful flashing Mondrian suppressors consisted of orange, yellow, green, indigo, and violet. No blue and red colors were in the suppressors to prevent confusion between the suppressor and the suppressed. On the non-dominant eye, the prime was presented with the contrast ramping up from 0% to the designated contrast determined by a trial-by-trial thresholding procedure. Each color had its own 3-up-1-down contrast calibration staircase: when a suppressed stimulus was detected, the contrast decreased in the next trial, while if a suppressed stimulus was not detected three trials in a row, the contrast increased in the next trial. The initial contrast was chosen according to participants’ performance on the 20 practice trials before they proceeded the actual experiment.

Both the suppressor and the suppressed were presented in an on-and-off manner with 400 ms on and 400 ms off to ensure stronger suppression. During 400-ms stimulus presence, the suppressed was sandwiched by the suppressor temporally, leaving the first and last two frames absence to prevent sudden breakthrough and afterimage, respectively. This led to a 333-ms prime presentation time in each on-and-off cycle. This on-and-off cycle lasted five times, resulting in 4-s suppression period (in a recent study (Hung and Hsieh, under review), we showed that intermittent presentation, coined discontinuous flash suppression, during the suppression period delayed breakthrough of the stimulus and hence potentially increased subliminal signal of the stimulus. Thus we applied such interocular suppression paradigm here to achieve longer suppression and exposure duration of the prime). The suppressed prime was presented either above or below the fixation point. The location was counterbalanced across the conditions. After which, the target was presented to the suppressor eye in a different font size and a slightly jittered location to prevent simple adaptation until response.

In total, 352 trials were completed with 320 experimental trials (2 prime colors × 2 prime words × 2 target colors × 2 target words × 2 prime location × 10 repetitions) and 32 catch trials, including 16 blank trials where no stimulus was delivered during the suppression period and 16 visible trials where the stimulus was delivered to the dominant eye and superimposed with the Mondrians. These catch trials allowed us to further gauge if participants developed any responses bias in the course of the experiment.

### Experiments 3 and 4: experimental design and procedure

The experimental design and procedure of Experiments 3 and 4 were identical to Experiments 1 and 2 except for two alterations: (1) The prime words were made colorless. We focused only on the semantic effect and aimed to re-examine and replicate the semantic incongruency effect of the suppressed prime and visible target. (2) Half of the trials were made blank. In total 336 trials were performed in each individual (2 prime presence/blank × 2 prime words × 2 target colors × 2 target words × 2 prime locations × 10 repetitions + 16 visible catch trials). These blank trials not only allowed us to gauge the false alarm rates of breaking suppression but also allowed us to normalize the reaction time of suppressed prime presence trials against.

$${\mathrm{RTnomalized}} = \frac{{{\mathrm{RTwith}}\,{\mathrm{prime}} - {\mathrm{RTwithout}}\,{\mathrm{prime}}}}{{\left( {{\mathrm{RTwith}}\,{\mathrm{prime}} + {\mathrm{RT}}\,{\mathrm{without}}\,{\mathrm{prime}}} \right)/2}}.$$
(1)

Participants in Experiment 3 were asked to name the word of the target while participants in Experiment 4 were asked to name the color of the target.

### Experiments 5 and 6: experimental design and procedure

The experimental design and procedure of Experiments 5 and 6 were identical to Experiments 1 and 2 except for two alternations: (1) The primes were meaningless symbols with color (XXXX in color blue or red). (2) Identical to Experiments 3 and 4, half of the trials were made blank. In total 336 trials were performed in each individual 2 prime presence/blank × 2 prime colors × 2 target colors × 2 target words × 2 prime locations × 10 repetitions + 16 visible catch trials.

### Experiments 7 and 8: experimental design and procedure

The experimental design and procedure of Experiments 7 and 8 were identical to Experiments 3 and 4 except for one alternation: the targets were replaced by scarlet and navy so that the prime-target relationship was semantic but not orthographic. Similarly, in total 336 trials were performed in each individual 2 prime presence/blank × 2 prime colors × 2 target colors × 2 target words × 2 prime locations × 10 repetitions + 16 visible catch trials.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

## Data availability

The raw data are available from the corresponding author upon reasonable request.

## References

1. 1.

Chun, M. M., Golomb, J. D. & Turk-Browne, N. B. A taxonomy of external and internal attention. Annu. Rev. Psychol. 62, 73–101 (2011).

2. 2.

Lavie, N., Hirst, A., de Fockert, J. W. & Viding, E. Load theory of selective attention and cognitive control. J. Exp. Psychol.: Gen. 133, 339–354 (2004).

3. 3.

Lavie, N., Beck, D. M. & Konstantinou, N. Blinded by the load: attention, awareness and the role of perceptual load. Philos. Trans. R. Soc. B: Biol. Sci. 369, 20130205–20130205 (2014).

4. 4.

Saenz, M., Buracas, G. T. & Boynton, G. M. Global effects of feature-based attention in human visual cortex. Nat. Neurosci. 5, 631–632 (2002).

5. 5.

Posner, M. I. Orienting of attention. Q. J. Exp. Psychol. 32, 3–25 (1980).

6. 6.

Carrasco, M., Ling, S. & Read, S. Attention alters appearance. Nat. Neurosci. 7, 308–313 (2004).

7. 7.

Schwartz, S. et al. Attentional load and sensory competition in human vision: modulation of fmri responses by load at fixation during task-irrelevant stimulation in the peripheral visual field. Cereb. Cortex 15, 770–786 (2004).

8. 8.

O’Connor, D. H., Fukui, M. M., Pinsk, M. A. & Kastner, S. Attention modulates responses in the human lateral geniculate nucleus. Nat. Neurosci. 5, 1203–1209 (2002).

9. 9.

Ling, S., Pratte, M. S. & Tong, F. Attention alters orientation processing in the human lateral geniculate nucleus. Nat. Neurosci. 18, 496–498 (2015).

10. 10.

Rauss, K. S., Pourtois, G., Vuilleumier, P. & Schwartz, S. Attentional load modifies early activity in human primary visual cortex. Hum. Brain Mapp. 30, 1723–1733 (2009).

11. 11.

Giattino, C. M., Alam, Z. M. & Woldorff, M. G. Neural processes underlying the orienting of attention without awareness. CORTEX 102, 14–25 (2018).

12. 12.

Rossi, A. F. & Paradiso, M. A. Feature-specific effects of selective visual attention. Vis. Res. 35, 621–634 (1995).

13. 13.

Cartwright-Finch, U. & Lavie, N. The role of perceptual load in inattentional blindness. Cognition 102, 321–340 (2007).

14. 14.

Simons, D. J. & Chabris, C. F. Gorillas in our midst: sustained inattentional blindness for dynamic events. Perception 28, 1059–1074 (1999).

15. 15.

Macdonald, J. S. P. & Lavie, N. Load induced blindness. J. Exp. Psychol.: Hum. Percept. Perform. 34, 1078–1091 (2008).

16. 16.

Bahrami, B., Lavie, N. & Rees, G. Attentional load modulates responses of human primary visual cortexto invisible stimuli. Curr. Biol. 17, 509–513 (2007).

17. 17.

Kanai, R., Tsuchiya, N. & Verstraten, F. A. J. The scope and limits of top-down attention in unconscious visual processing. Curr. Biol. 16, 2332–2336 (2006).

18. 18.

Hsieh, P.-J., Colas, J. T. & Kanwisher, N. Pop-out without awareness: unseen feature singletons capture attention only when top-down attention is available. Psychol. Sci. 22, 1220–1226 (2011).

19. 19.

Groen, I. I. A., Silson, E. H. & Baker, C. I. Contributions of low- and high-level properties to neural processing of visual scenes in the human brain. Philos. Trans. R. Soc. B: Biol. Sci. 372, 20160102 (2017).

20. 20.

Macleod, C. M. Half a century of research on the stroop effect: an integrative review. Psychological Bull. 109, 163–203 (1991).

21. 21.

Scheibe, K. E., Shaver, P. R. & Carrier, S. C. Color association values and response interference on variants of the Stroop test. Acta Psychologica 26, 286–295 (1967).

22. 22.

Kiefer, M. & Martens, U. Attentional sensitization of unconscious cognition: Task sets modulate subsequent masked semantic priming. J. Exp. Psychol.: Gen. 139, 464–489 (2010).

23. 23.

Martens, U. & Kiefer, M. Specifying attentional top-down influences on subsequent unconscious semantic processing. Adv. Cogn. Psychol. 5, 56–68 (2009).

24. 24.

Koch, C., Koch, C., Tsuchiya, N. & Tsuchiya, N. Attention and consciousness: two distinct brain processes. Trends Cogn. Sci. 11, 16–22 (2007).

25. 25.

Cohen, M. A., Cavanagh, P., Chun, M. M. & Nakayama, K. The attentional requirements of consciousness. Trends Cogn. Sci. 16, 411–417 (2012).

26. 26.

Rahnev, D. A., Huang, E. & Lau, H. Subliminal stimuli in the near absence of attention influence top-down cognitive control. Atten., Percept. Psychophys. 74, 521–532 (2011).

27. 27.

Hung, S. M., Nieh, C.-H. & Hsieh, P.-J. Unconscious processing of facialattractiveness: invisible attractivefaces orient visual attention. Sci. Rep. 6, 1–8 (2016).

28. 28.

Tsuchiya, N. & Koch, C. Continuous flash suppression reduces negative afterimages. Nat. Neurosci. 8, 1096–1101 (2005).

29. 29.

Jiang, Y., Costello, P. & He, S. Processing of invisible stimuli: advantage of upright faces and recognizable words in overcoming interocular suppression. Psychological Sci. 18, 349–355 (2007).

30. 30.

Costello, P. et al. Semantic and subword priming during binocular suppression. Conscious. Cognition 18, 375–382 (2009).

31. 31.

Hung, S. M. & Hsieh, P.-J. Syntactic processing in the absence of awareness and semantics. J. Exp. Psychol.: Hum. Percept. Perform. 41, 1376–1384 (2015).

32. 32.

Gayet, S., Van der Stigchel, S. & Paffen, C. L. E. Breaking continuous flash suppression: competing for consciousness on the pre-semantic battlefield. Front. Psychol. 5, 460 (2014).

33. 33.

Kanai, R., Walsh, V. & Tseng, C.-H. Consciousness and cognition. Conscious. Cognition 19, 1045–1057 (2010).

34. 34.

Eo, K., Cha, O., Chong, S. C. & Kang, M. S. Less is more: semantic information survives interocular suppression when attention is diverted. J. Neurosci. 36, 5489–5497 (2016).

35. 35.

Lin, Z. & Murray, S. O. Unconscious processing of an abstract concept. Psychological Sci. 25, 296–298 (2014).

36. 36.

Pelli, D. G. The VideoToolbox software for visual psychophysics: transforming numbers into movies. Spat. Vis. 10, 437–442 (1997).

37. 37.

Brainard, D. H. & Brainard, D. H. The psychophysics toolbox. Spat. Vis. 10, 433–436 (1997).

38. 38.

Miles, W. R. Ocular dominance in human adults. J. Gen. Psychol. 3, 412–430 (1930).

## Acknowledgements

The authors thank the support of James Boswell Postdoctoral Fellowship and Caltech Biology and Biological Engineering Divisional Postdoctoral Fellowship to S.-M.H and the research funding from the Japan Science and Technology Agency (JST) (JST.CREST2014) to S.S.

## Author information

Authors

### Contributions

Conceptualization, S.-M.H. D.-A.W. and S.S.; Methodology, S.-M.H.; Software, S.-M.H.; Investigation, S.-M.H.; Writing—Original Draft, S.-M.H.; Writing—Review and Editing, S.-M.H. D.-A.W. and S.S.; Supervision, S.S.

### Corresponding author

Correspondence to Shao-Min Hung.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

Peer review information Nature Communications thanks Min-Suk Kang, Robert Kentridge and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Hung, SM., Wu, DA. & Shimojo, S. Task-induced attention load guides and gates unconscious semantic interference. Nat Commun 11, 2088 (2020). https://doi.org/10.1038/s41467-020-15439-x

• Accepted:

• Published:

• ### Subliminal temporal integration of linguistic information under discontinuous flash suppression

• Shao-Min Hung
•  & Po-Jang Hsieh

Journal of Vision (2021)