In this study, 37 undergraduate students from Hangzhou Normal University of China (9 men; range = 21–25 years old) were paid for participating in Experiment 1, and 30 students for Experiment 2 (7 men; range = 19–23 years old). All participants were right-handed with normal or corrected-to-normal vision. All were native Chinese speakers. Participants were provided informed consent, as approved by the institutional ethics committee of the Center for Cognition and Brain Disorders in Hangzhou Normal University.

Experiments 1 and 2 adopted the same materials. There were three forms of masks in the study (as shown in Figure 1A): patterns, metacontrast/paracontrast, and four dots. Pattern masking was random pattern of black oblique bars (each 12 × 2 mm in size) on the transparent background, used in the previous study (Oyama et al., 1983), containing 112 bars placed in positions randomly sampled with a probability of 0.5 from a 15 × 15 matrix (14 × 14 cm, 10 × 10 degrees). Half of the bars in the mask were oriented at 135 degrees, the other half were oriented at 45 degrees. Six different pattern masks for each bar orientation were used, as shown in Figure 1A. The metacontrast was four vertical and horizontal lines formed in two parallel pairs that are closely adjacent to the contours without superimposed with the Chinese characters. The width of lines used to construct the para/meta mask was 0.5 cm. The four-dot masking was modified with four squares surrounded by targets, and the square “dots” for a 2 cm × 2 cm.

Three hundred characters were chosen as target stimuli, whose orthography was manipulated to form character-like stimuli: real Chinese character (e.g., 响), orthographically correct pseudo-character (e.g., ), and stroke randomly combined non-character (e.g., ) (as shown in Figure 1B). A hundred Chinese real characters were selected from the Modern Chinese Corpus Centre for Chinese Linguistics, Peking University¹. All these real characters were left-right compound characters, with word frequency ranging from 877.7/million to 2,536/million (mean = 1,462.3, SD = 462.1). The stroke number of characters ranged from 7 to 11 (mean = 8.6, SD = 1.3). Pseudo-characters were made by combining radicals from two different characters according to orthographical rules but neither phonetic nor semantic components were present. Non-characters were constructed by combing the strokes of real characters randomly. There was a total of 100 pseudo-characters and non-characters. All the 300-character stimuli were presented randomly. All Chinese characters were Song font, black, and were presented on a gray background (RGB 128, 128, 128). The visual angle of stimuli was 4.4 × 4.4 degrees.

All stimuli were displayed on a 17-inch CRT (cathode-ray-tube) monitor (1,600 × 1,200 resolution) running at a refresh rate of 85 Hz by using JavaScript coded in the Visual Studio Code application. Participants were seated in a dimly lit room, attached to a chin rest at a distance of approximately 58 cm.

Each testing session began with several practice trials and was then followed by a formal experiment. Both Experiment 1 and 2 adopted 3 × 3 × 3 within-subject designs: mask form (patterns, metacontrast/paracontrast, and four-dot), character-likeness (real character, pseudo-character, and non-character), temporal sequence [forward (FM), backward (BM), and sandwiched masking (SM)]. The serial order of these manipulations was fully balanced across the observers in a Latin square. During the formal experiments, participants were asked to focus on the central fixation of a computer screen all the time. They were asked to respond as quickly and accurately as they could. Participants in Experiment 1 were asked to do a two-alternative forced choice task (2AFC task) and those in Experiment 2 had to finish a lexical decision task (LD task, which is to judge whether the target stimulus is a real character or not). Figure 2 showed experimental procedures in the two tasks. The position of the correct choice was randomly determined on each trial. The number of “yes” and “no” responses was balanced in each experiment. RTs and accuracies were recorded.

The current analyses included three dependent variables. First, mean RTs for all conditions from correct trials were computed and analyzed for each subject. Second, instead of accuracy, a bias-free sensitivities index (A′) was calculated with the below equation (1) for all conditions. A′ is a non-parametric measure of sensitivity according to the signal detection theory and is relatively unaffected by response bias when the assumption of normality and equal variances are violated (Stanislaw and Todorov, 1999). It has been indicated that A′ is a better estimation of performance in a 2AFC task (Wong et al., 2012). The formula of A′ is shown below. Third, the magnitudes of masking effects indexed by ΔA′ and ΔRTs were calculated by subtracting the RTs and the A′s for choosing targets (real character, pseudo-characters, and non-characters) under each temporal sequence from those under the control condition(ΔA′ = control − A′; ΔRTs = control − RTs). A higher value of ΔRTs and ΔA′ designates a larger size of the masking effect, demonstrating bad performance, and relative disruption by masking.

H = hit rate, F = false alarm rate

A 3 × 3 ANOVA of mask form and character-likeness was conducted separately for each of the FM, BM, and SM conditions. Post-hoc comparisons were corrected with Bonferroni correction(Armstrong, 2014).

To test if the task modulates the masking effect on character processing, the task demand was taken as a between-group factor. A 4-way ANOVA of ΔA′ revealed a significant main effect of mask form [F(2, 130) = 8.351, p < 0.001, η_p² = 0.114]. Significant main effect of temporal sequence was observed [F(2, 130) = 23.754, p < 0.001, η_p² = 0.268]. There was a significant 4-way interaction [F(8, 520) = 4.154, p < 0.001, η_p² = 0.06]. Post-hoc analyses revealed that greater masking effects were observed on real characters in the paracontrast and four-dot FM under the 2AFC than those under the LD task, t(66) = 4.501, p < 0.05; t(66) = 18.905, p < 0.001, and in the metacontrast SM under the 2AFC task than those under the LD task, t(66) = 18.877, p < 0.001. Masking effects by pattern and metacontrast BM were stronger for pseudo-character under the 2AFC than the LD task, t(66) = 5.166, p < 0.05; t(66) = 13.196, p < 0.01. Results of post-hoc comparisons are displayed in Table 3.

As for ΔRTs, there were significant main effects of character-likeness [F(2, 130) = 85.973, p < 0.001, η_p² = 0.569], mask form [F(2, 130) = 80.819, p < 0.001, η_p² = 0.554], temporal sequence [F(2, 130) = 16.573, p < 0.001, η_p² = 0.203]. A significant 4-way interaction was observed [F(8, 520) = 16.787, p < 0.001, η_p² = 0.205]. A post-hoc analysis for the interaction was conducted (as shown in Table 3 and Figure 5).

The task-oriented attention had their distinct influences on the target identification in all forms of temporal masking. The overall masking effects were stronger in backward masking under the 2AFC task, but greatly reduced or the target sensitivity was enhanced under the LD task. These task-related changes might reflect the attentional selection, which can bias the processing in favor of task-related features or objects. This is also evidenced by a previous study by Pilling and Gellatly (2010). They presented additional display items to influence the linking apparent motion seen between a target and a spatially separated mask and used placeholders to maintain the presence of target object during mask presentation, both manipulations resulted in a significant reduction of masking when the conditions promoted the target-mask individuation. It is suggested that attention involves not only the facilitation of needed information but also the suppression of unneeded information (Boyer and Ro, 2007; McMains and Kastner, 2009; Grootswagers et al., 2021). The unneeded stimuli (e.g., masks) or task-irrelevant information are decoded in the visual hierarchy, resulting from a degree of automaticity in visual processing (Grootswagers et al., 2019, 2021). This can explain why the masking across the tasks is active no matter how attention facilitates the target processing. Attention enhanced sensitivity of the targets in the masking by having the processing of the task-relevant information take priority over that of task-irrelevant one (Grootswagers et al., 2021). Moreover, we observed that the sensitivity enhancement of the targets in the masking was obvious in real characters and pseudo-characters under the LD task. The differences in task effect indicated that the magnitude of enhancement in sensitivity depends on the nature of the attended or task-relevant information. When attention was directed to lexical components under the LD task, processing was biased up to abstract properties of target characters and greatly suppresses the visual interference from masks. But when attention was directed to the perceptual contents under the 2AFC task, masking became more effective due to spatio-temporal alignments with the targets in the visual pathway.

Besides, Bachmann and Francis suggested that the target processing was facilitated mainly via conscious awareness (Bachmann and Francis, 2013). The conscious awareness happens because of the slower non-specific subcortical activity elicited by various mental preparations, such as pre-and post-target state or temporal expectancy modulates faster specific cortical activity generated by the target stimulus and participating in the representation of the target sensory-perceptual contents (Bachmann and Francis, 2013). In the case of the current experiment settings, as they suggested, awareness of the task requirements prior to the target-mask sequence or temporal expectancy induced by periodic stimulation might help upcoming targets upgraded to the conscious representation ahead. However, the task effects on target sensitivity in masking cannot be well-explained by this account. He et al. (1996) suggest that the availability of visual information to conscious awareness is restricted by the attention that acts in one or higher visual cortical areas. The differences in masking effectiveness between tasks are likely attributed to attentional focus on specific representation levels of the task-relevant information.

Our results revealed significant interactions between mask form and character-likeness across temporal sequences regardless of the task requirement. This suggested that the tempo-spatial influence on the target processing is unavoidable in visual masking. This influence was shown in all levels of processing. For instance, all forms of backward masks are observed to reduce the sensitivity of real characters under the 2AFC task and that of non-characters under the LD task, while pattern and metacontrast backward masking reduced sensitivity of pseudo-characters under the 2AFC task. The identification of real characters involves all levels of processing from visual analysis to lexical access, whereas non-characters and pseudo-characters well represent the featural and character form analysis, respectively. It seems that visual masking regardless of mask forms might involve either low or higher levels of processing, specific to characters corresponding to the feature analysis or lexical access. The masking effects on pseudo-characters by pattern and metacontrast might relate to their suppressions in the midlevel of character processing or character form analysis. It seems that these findings are compatible with the re-entrant account proposing that processing of visual characters in masking is a process of testing perceptual hypotheses through re-entrant iterative comparisons from the high-level areas to the ongoing low-level activity (Enns and Di Lollo, 2000). Alternatively, Chakravarthi and Cavanagh (2009) argued that all forms of backward masking lead to an abrupt attenuation of target signals at the earlier stage and thus result in the lack of later or higher-level feedback processing. Evidence has been shown that backward masking by noise and metacontrast have their maximally inhibitory effects on letter identification at a target-mask interval of 0–50 ms, whereas masking by four dots showed a similar effect at an interval later than 150 ms (Enns, 2004). The target-mask interval of 47 ms in our study is just at this interval when metacontrast and pattern masks maximize their spatial interference. The four-dot backward masking at our short interval indicated that its interference can occur as early as other forms of masking and influences the low level of processing. The masking at this level is associated with different mechanisms depending on the form of masks. The pattern masking works through spatio-temporal integration (Oyama et al., 1983; Enns, 2004), while the underlying mechanism of the metacontrast masking is transient-on-sustained inter-channel inhibition (Breitmeyer et al., 2006).

In contrast to the backward masking, forward masking by four dots and paracontrast had either facilitatory or inhibitory effects on target sensitivity across tasks in the current findings. These two forms of masks might have similar masking mechanisms. Kafaligönül et al. (2009) revealed that variations of paracontrast masking occur at different SOA ranges, modulated by the task requirement. The prolonged inhibition and the facilitation were explained as activities of slower cortical and subcortical systems (Breitmeyer et al., 2006; Kafaligönül et al., 2009). Breitmeyer (2007) suggested that paracontrast suppresses primarily the early feedforward processing and then indirectly weakens late re-entrant activity.

In closing, the present findings extend previous studies of this field by considering multiple interactions among factors, demonstrating contributions of spatio-temporal interaction and attentional demand in the identification of the masked target characters. Polat et al. (2007) have suggested that the masking effect depends on at least three factors: processing time of a target, temporal sequence, and spatial arrangement. Moreover, our findings might help to explain the variability of studies driven by differential attention mechanisms.

According to the recurrent processing model, visual recognition is a consequence of both feedforward and re-entrant processing, in which the visual input is transformed through cortical areas via feed-forward pathways and via re-entrant pathways from higher to lower areas (Enns and Di Lollo, 2000; Breitmeyer, 2007; Dux et al., 2010; Mohsenzadeh et al., 2018). Both forward and backward masking can cause suppression of early feedforward processing. This early suppression may be compensated by increased re-entrant processing at later stages (Mohsenzadeh et al., 2018). Future research should try to combine neuroimaging methods to clarify what are the contributions of cognitive processes (conscious awareness) supported through subcortical activities.