We recruited 15 illiterate (14 female) and fifteen literate adults (14 female) from the same rural village in central China. Illiterate adults had an average age of 64.40 years old (range: 53−77 years; SD = 6.50 years), and the average age of literate adults was 61.00 years old (range: 56−68 years; SD = 3.35 years). The G-Power 3.1 (Faul et al., 2007) was used to assess the required sample sizes based on a prior study (Cao et al., 2019), in which the effect size (η_p² = 0.21) of interaction between Literacy and Orientation was applied to the present experiment. A power analysis showed that a sample size of 5 in one group would be required at the 0.05 alpha level with 90% power. Thus, we had enough sample size to detect this effect size. The illiterate adults were identified by the Grade one literacy questionnaire (Wang and Tao, 1993), the Mini-Mental State Examination (MMSE, Wu et al., 2002), and the Word-color Stroop task. These illiterates were those who could not recognize Chinese characters in the one, two, and three lists or less than five Chinese characters in succession in the Grade One literacy questionnaire (Wang and Tao, 1993; Cao et al., 2019), and could pass the test of MMSE (the score is above 14). We used the scores of MMSE to ensure that none of the participants experienced cognitive impairment, and used the scores of the Word-color Stroop task to identify that the illiterate participants indeed could not recognize common Chinese characters. These critical cognitive abilities measured in these tasks are considered core consequences of literacy acquisition, which has been used to distinguish between illiterate and literate individuals (Cao et al., 2019). All of them were orally reported as right-handed, with normal or corrected-to-normal vision. None of the participants has a history of mental illness or cognitive impairment. All participants and their relatives received verbal or written informed consent, and the ethical committee at Zhejiang Normal University approved the study.

The original materials were twenty Chinese adults’ upright face pictures with neutral expressions (half of them were male faces) (Cao et al., 2016, 2019; Zhang et al., 2021). Contour features, such as hair and ears, were cropped out using Photoshop (PS), and the isolated face pictures were put against a neutral gray background. Twenty upright houses were used as control materials, taken from the local village where participants lived (Cao et al., 2019). The production process of the upside house pictures was the same as that of the face pictures. The lightness and brightness of the pictures were generally identical. We used the software Adobe Photoshop CS5 (San Jose, CA) to adjust the lightness and brightness of visual stimuli and kept two types of visual stimuli similar in lightness and brightness (houses: 124 ± 19.58 cd/m²; full-spatial frequencies faces: 147 ± 1.84 cd/m²). We rotated the upright face/house pictures 180 degrees to make the inverted face/house pictures (see Figure 1). The visual angle of each picture is 4.0^° × 6.2^°. The experiment was carried out on Lenovo’s 14 inches G470 laptop with a screen resolution of 1024 × 768 (Zhang et al., 2018). The experiment was run based on the E-prime 1 software¹.

To begin with, a 500 ms black fixation cross appeared against a white background at the center of the screen, followed by a study stimulus for 600 ms. Then, a blank gray screen was displayed for 1,000 ms, followed by a test stimulus, which was displayed in the same orientation as the study stimulus. The test stimulus would disappear as soon as the participant initiated the response (see Figure 2). The task of the participants is to judge whether the study stimulus and the test stimulus were identical as quickly and accurately as possible. Half of the participants pressed “A” for “same” or “L” for “different”; for the other half, the key response was counterbalanced. The inter-trial interval was 1,000 ms. Participants completed 16 practice trials before the formal experiment. Only when the accuracy of the practice procedure was higher than 75% could they be allowed to start the formal experiment. Otherwise, they needed to complete the practice procedure again and again. The formal experiment consisted of four blocks with a total of one hundred and sixty trials, and each block had twenty trials for each condition (i.e., upright and inverted conditions).

We used SPSS 22.0 software² to analyze the accuracy and response time with a 2 (Literacy: illiterate vs. literate) × 2 (Stimulus Category: face vs. house) × 2 (Orientation: upright vs. inverted) mixed analysis of variance (ANOVA). In this model, we took Stimulus Category and Orientation as within-subject factors and Literacy as the between-subjects factor. We also performed independent sample t-tests to examine the difference in the size of the face inversion effect between literate and illiterate groups when the interaction of Literacy and Orientation was significant. We deleted 1.2% of trials accounting for all correct trials whose response times were less than 300 ms or longer than 3,000 ms (Cao et al., 2019).

We recruited eleven illiterate (ten female) and eleven literate adults (11 female) from a rural area of central China. Note that eight illiterate adults (one male) and all literate adults participated in Experiment 1 because illiterate adults were not readily available in the region where the study was conducted. Illiterate adults had an average age of 63.27 years old (range: 53−69 years; SD = 4.96 years), and the average age of literate adults was 61.00 years old (range: 56−68 years; SD = 3.90 years). The G-Power 3.1 was used to assess the required sample sizes in experiment 2. In this analysis, the mean effect size (η_p² = 0.17) of Cao et al. and experiment 1 in the present study was adopted to calculate the sample size, which demonstrated that a sample size of six in one group would be required at 0.05 alpha level with 90% power. Thus, the sample size used in the present experiment is necessary to detect this effect size. All of them were right-handed native Chinese with normal or corrected-to-normal vision. All participants and relatives received verbal or written informed consent, and the ethical committee at Zhejiang Normal University approved the study. The screening of subjects was the same as in experiment 1.

The face stimuli of experiment 2 are all from experiment 1. MATLAB software³ was used to filter all the face materials with low-spatial frequency (2−8 cycles per face) and high-spatial frequency (32−128 cycles per face) (Goffaux and Rossion, 2006; Cheung et al., 2008). Figure 4 shows an exemplar stimulus for each condition.

The experiment procedure is the same as experiment 1.

We used SPSS 22.0 software (see text footnote 2) to analyze the accuracy and response time with a 2 (Literacy: literate vs. illiterate) × 2 (Spatial Frequency: high vs. low) × 2 (Orientation: upright vs. inverted) ANOVA. In this model, we took Spatial Frequency and Orientation as within-subjects factors and Literacy as the between-subjects factor. In addition, we performed independent sample t-tests to examine differences in the size of the face inversion effect between literate and illiterate groups when the interaction of Literacy and Orientation was significant. We deleted 5.1% of trials accounting for all correct trials whose response times were less than 300 ms or longer than 3,000 ms.

Consistent with the neural recycling hypothesis (Dehaene and Cohen, 2007; Dehaene et al., 2010), which claims that literacy acquisition “invades” these cortexes typically for processing faces or other objects, the present study showed a larger inversion effect for faces and houses in literate adults compared with illiterate control. Moreover, the literate performed better in recognizing upright faces and houses presented sequentially than the illiterate. The present study, combined with a prior study in which paired faces were presented simultaneously (Cao et al., 2019), suggested that the way of stimuli presentation (simultaneous vs. sequential) seems not to change the outcome of literacy facilitation effect on face inversion effect. However, these two presentations may tax different cognitive loads (Menon et al., 2015) and require distinct processing strategies (Hole, 1994; Richler et al., 2009). The previous literature demonstrated that simultaneous presentation required less cognitive effort compared with sequential presentation; simultaneous presentation facilitated participants to process faces holistically, and even decreased the preference for the use of the local features (Yang and Schwaninger, 2010), while participants in the sequential presentation may employ memory-based-implicating strategies or the outcome-of-an-attention strategies. In sum, the findings suggested that the literacy effect on the inversion effect for faces was stable.

In addition, the literacy effect on face processing also extended to non-facial stimuli (i.e., house) processing, which suggested that the literacy effect might be domain-general, supporting the neural recycling hypothesis. More precisely, these original cortexes that learning to read “invades” may subserve domain-general cognitive ability.

In the present study, we also found partial evidence that the illiterate participants reacted faster to faces with high spatial frequency than the literate participants, while a comparable response speed on faces with low spatial frequency was observed between illiterate and literate adults. The findings seem to suggest that literacy acquisition attenuated the processing speed to the facial information conveyed by the high spatial frequency. In other words, the illiterate seem to be inclined to employ the featural information to process faces as compared with the literate, given that the high spatial frequency transmits more features or local information. These findings were not inconsistent with the view that literacy acquisition improved configural face processing (such as Cao et al., 2019). Additionally, Patching and Jordan (2005) found that good readers showed higher sensitivity to low spatial frequency as compared with poor readers. Although poor readers and illiterate adults had similarities in terms of reading, they were essentially different groups. To sum up, combined with these studies mentioned above, the present study provided insights into the complex relationship between literacy and the sensitivity to distinct spatial frequencies.

Finally, we found that the literacy effect on face processing was not modulated by spatial frequency. This is consistent with previous research, which shows that low and high spatial frequency evoked holistic face processing in a complete composite face paradigm (Cheung et al., 2008). These studies suggested that literacy acquisition seemed not to drive different sensitivities to spatial frequency to support configural face processing.

The two experiments in the present study both showed the facilitated face inversion effect of literacy, (partially) consistent with the findings of Cao et al. (2019) and Ventura et al. (2013). Cao et al. (2019) used the spatial configuration task and showed that illiterate adults were less sensitive to the relationships between facial features than literate controls in upright rather than inverted faces. However, Ventura et al. (2013) found that literacy acquisition decreased the holistic processing indexed by the composite face effects. Two possibilities could account for the inconsistent result patterns between Ventura et al. (2013) and the present study. One obvious possibility may be associated with the differences in these experimental paradigms (composite face paradigm and face inversion paradigm). The composite face paradigm reflects the failure of selective attention to facial parts, whereas the face inversion paradigm demonstrates the disruption of the typical arrangement of facial features. Different experimental paradigms could tap into the different dimensions of configural face processing (Maurer et al., 2002; Richler et al., 2012), given the complex relationships among different paradigms (Teunisse and De Gelder, 2003; Wang et al., 2012; DeGutis et al., 2013; Rezlescu et al., 2017). Therefore. future studies are encouraged to perform a face inversion task among western literate and illiterate adults or to conduct a composite face task in Chinese illiterates and literates.

An alternative possibility is an attention-shaped-by-language hypothesis (Yang et al., 2021), which suggests that language shapes how its user deploys the attentional resources to the visual processing of words, given the extensive and longstanding experience with the visual features of words. In the current case, literates acquired Portuguese scripts in Ventura et al.’s (2013) study, whereas literate adults acquired Chinese scripts in the present study and Cao et al.’s (2019)study. Chinese characters may be special cases as they possess a highly non-linear visual complexity of shape (Huang et al., 2004), while Portuguese words are composed of basic letters and are linear. Long-term and extensive exposure to the differences in visual features between Chinese and Portuguese scripts may induce distinct processing demands for Chinese compared to Portuguese scripts and further transfer to the processing of configural faces. Recently, Ma et al. (2022) showed that N170 amplitudes elicited by naturalistic and Mooney faces were larger for Chinese children than for German children, which suggested superior holistic processing in Chinese children. Using the Framed-Line Test in Portugal and Thailand, Ventura et al. (2008) found that Portuguese literates showed higher accuracy in the absolute task (reflecting the analytic processing) compared to the relative task (reflecting the holistic processing). However, Portuguese illiterates and ex-illiterates showed higher accuracy in the relative task than in the absolute task. However, all Thai groups showed consistently higher accuracy on the relative task than on the absolute task. The attention-shaped-by-language hypothesis seems to account for the domain-general mechanism underlying the literacy effect on object inversion processing (such as faces and houses). In a word, these findings demonstrated that schooling/literacy imposed different impacts on the participants with distinct script systems. Future studies may find it worthwhile to explore new experimental designs to examine the attention-shaped-by-language hypothesis.

The current data adds ingredients to the theoretical development of the relations between literacy and face processing by examining the first-order configural processing (i.e., face inversion effect) between Chinese literate and illiterate adults. Following the neural recycling hypothesis, previous studies mainly centered on the correlations between face processing and word processing, or between face processing and the amount of literacy. Recently, few studies (including the present study) with different paradigms of configural face processing shed light on how literacy acquisition affects configural face processing, particularly in different script systems (Ventura et al., 2013; Cao et al., 2019). Combined with the existing findings, the present study also proposed that the attention-shaped-by-language hypothesis could account for the cognitive mechanisms underlying literacy acquisition on configural face processing. Moreover, the present study also has some implications for further studies, which should be conducted by recruiting those literate and illiterate adults trained with different script systems.

Some limitations may be considered in the present study. First, socialization may reduce the credibility of making causal inferences. Participants in both groups were matched as closely as possible on age and gender, but they may differ in the amount of their exposure to faces. For instance, the literate may experience more distinct faces than the illiterate when attending regular school (Cao et al., 2019). They have more access to a broader range of people after acquiring literacy. The experience of faces may lead to differential effects on configural face processing (Balas and Saville, 2015).

Second, previous studies demonstrated that females performed better face perception and cognition than males (Sommer et al., 2013). However, in the present study, almost illiterate adults were female (only one male) because males had more opportunities to attend school in China in the 20th century. Therefore, we shall be cautious to generalize the current findings to male illiterate adults, or other populations.