Face adaptation induces duration distortion of subsequent face stimuli in a face category-specific manner

Abstract

Studies have shown that duration perception depends on several visual processes. However, the stages of visual processes that contribute to duration perception remain unclear. This study examined the effects of categorical differences in face adaptation on perceived duration. In all the experiments, we compared the perceived durations of human, monkey, and cat faces (comparison stimuli) after adapting to a human face. Results revealed that the human comparison stimuli were perceived shorter than the monkey and cat comparison stimuli (categorical face adaptation on duration perception [CFAD]). The difference between the face categories disappeared when the adapting stimulus was rendered unrecognizable by phase scrambling, indicating that adaptation to low-level visual properties cannot fully account for the CFAD effect. Furthermore, CFAD was preserved but attenuated when the adapting stimulus was inverted or a 1,000-ms interval was inserted before the comparison stimuli, which implied that CFAD occurred as long as the adapting stimulus was perceived as a face and not simply based on conceptual category processes. These findings indicate that face adaptation affects perceived duration in a category-specific manner (the CFAD effect) and highlights the involvement of visual categorical processes in duration perception.

Introduction

Time perception, which is the ability to estimate event duration and temporal interval between events, is a fundamental dimension of perceptual and cognitive activities (Mauk & Buonomano, 2004). While the traditional models of time perception assume the existence of an internal clock dedicated to temporal processing (Treisman, 1963; Treisman, Faulkner, Naish, & Brogan, 1990; Zakay & Block, 1997), a growing body of research suggests that time perception is intrinsic in sensory processing (Buonomano & Laje, 2010; Eagleman & Pariyadath, 2009; Ivry & Schlerf, 2008). Indeed, studies have revealed that stimulus motion (Kanai, Paffen, Hogendoorn, & Verstraten, 2006; Kaneko & Murakami, 2009; Yamamoto & Miura, 2012), size (Ono & Kawahara, 2007; Xuan, Zhang, He, & Chen, 2007), luminance (Matthews, Stewart, & Wearden, 2011), spatial frequency (Aaen-Stockdale, Hotchkiss, Heron, & Whitaker, 2011), and novelty (Pariyadath & Eagleman, 2007; Tse, Intriligator, Rivest, & Cavanagh, 2004) affect the perceived duration of visual events, suggesting a close link between visual (intramodal) processing and duration perception. 

Adaptation studies have provided extensive data on the stages of visual processing involved in duration perception. Johnston, Arnold, and Nishida (2006) demonstrated that adaptation to a moving grating compressed the perceived duration of a subsequent moving grating presented at the adapted position. This duration compression effect has been suggested to occur only when the adaptor and comparison are presented at the same retinal position (e.g., Bruno, Ayhan, & Johnston, 2010). Moreover, isoluminant-color motion is known to reduce the duration compression effect compared to luminance-defined motion (Ayhan, Bruno, Nishida, & Johnston, 2011; Yoshimatsu, Murai, & Yotsumoto, 2022). These findings highlight the importance of the low-level visual processing stage. By contrast, other studies have reported evidence supporting the role of higher visual processing stages. For example, the duration compression effect is still observed when the adaptor and comparison are presented at the spatiotopically same but retinotopically different position (Burr, Cicchini, Arrighi, & Morrone, 2011; Burr, Tozzi, & Morrone, 2007; Morrone, Cicchini, & Burr, 2010). Furthermore, adaptation to global motion reduces the perceived duration of subsequent motion stimuli (Curran & Benton, 2012; Gulhan & Ayhan, 2019). These findings suggest that higher visual stages contribute to duration compression. Although these studies have demonstrated robust duration compression induced by motion adaptation, conflicting results show that the locus of the adaptation effect remains controversial. 

A possible reason for these inconsistent results is the hierarchical structure of motion adaptation. Although the importance of the middle temporal area (MT) for motion representation is not deniable, motion selectivity can also be observed early in the primary visual cortex (V1) (Smith, Greenlee, Singh, Kraemer, & Hennig, 1998), indicating that the MT is not the sole locus of motion processing. Indeed, studies on functional magnetic resonance imaging (fMRI) have demonstrated that prolonged viewing of motion reduces neural activity at multiple visual processing stages, including the MT and medial superior temporal area (MST), as well as at lower stages, such as V1, V2, and V3 (Huk & Heeger, 2002; Lee & Lee, 2012; Tootell et al., 1995), suggesting that motion adaptation occurs across various areas. This hierarchical structure of motion adaptation implies that the contribution of motion adaptation at one stage cannot rule out the involvement of other stages (Burr et al., 2011). Thus, identifying a more dissociable visual adaptation would be more effective in revealing the contribution of each visual processing stage to duration perception. 

Another family of high-level visual adaptations occurs during face processing and is known to have a more independent locus than motion processing (see Webster & MacLeod, 2011, for review). Prolonged exposure to visual facial stimuli alters the perception of subsequently presented faces. For example, adapting to one face for several seconds alters perceived identity of a subsequently presented face (Leopold, O'Toole, Vetter, & Blanz, 2001; Leopold, Rhodes, Müller, & Jeffery, 2005). These face adaptation effects have been reported to correlate with reduced neural responses in face-selective areas (Cziraki, Greenlee, & Kovács, 2010; Fang, Murray, & He, 2007; Kaiser, Walther, Schweinberger, & Kovács, 2013; Kovács, Cziraki, Vidnyánszky, Schweinberger, & Greenlee, 2008; Nagy, Zimmer, Greenlee, & Kovács, 2012), such as the occipital face area (OFA) (Gauthier et al., 2000) and fusiform face area (FFA) (Kanwisher, McDermott, & Chun, 1997; Kanwisher & Yovel, 2006). Unlike motion adaptation, face adaptation effects are highly robust to differences in the position (Leopold et al., 2001), size (Rhodes et al., 2004; Zhao & Chubb, 2001), and angle (Rhodes, Jeffery, Watson, Clifford, & Nakayama, 2003) of adapting and test faces. Moreover, face adaptation effects are observed even when the adaptor face moves during the adaptation phase (Fang, Ijichi, et al., 2007; Fang, Murray, et al., 2007). These findings emphasize that adaptation at the lower visual stages barely contributes to face adaptation effects. Thus, investigating the effects of face adaptation on perceived duration could provide insight into the role of higher visual stages in duration perception. 

In this study, we examined the effects of categorical differences in face adaptation on perceived duration. fMRI and MEG studies have reported that distinct neural response patterns and latencies in face-selective areas represent different subcategories of faces (Blonder et al., 2004; Cichy, Pantazis, & Oliva, 2014; Kriegeskorte et al., 2008; Looser, Guntupalli, & Wheatley, 2013). Furthermore, face adaptation effects have been reported to transfer across human faces of different identities and genders (Fang, Ijichi, et al., 2007), whereas no transfer occurs across different species (Little, DeBruine, Jones, & Waitt, 2008). These results imply category specificity of face adaptation effects. Therefore, if adaptation at higher visual stages played a role in duration perception, we could observe a shorter perceived duration when the adaptor and subsequent faces were in the same face category than in different categories. 

In Experiment 1, we compared the perceived durations of human, monkey, and cat faces (comparison stimuli) after adapting to a human face to determine whether the human comparison stimuli would be perceived shorter than the monkey and cat comparison stimuli (categorical face adaptation on duration perception [CFAD]). In Experiment 2, we phase-scrambled the human face adaptor to test whether adaptation to low-level visual properties accounted for the CFAD effect. In Experiment 3, we used an inverted human face adaptor to investigate the influence of configural information on the CFAD effect. In Experiment 4, we tested the life span of the CFAD effect by increasing the interval between the adaptor and comparisons. 

Experiment 1

Method

Participants

A total of 28 paid volunteers (mean age ± SD = 20.68 ± 1.66 years) participated in Experiment 1. The sample size was predetermined using the Python Pingouin package (Vallat, 2018) by computing the statistical power of the main effect of the category using analysis of variance (ANOVA). The power analysis revealed 22, which was sufficient to detect the main effect of a medium sample size (η² = 0.06), and a statistical power of 0.8, using a 5% significance level. All participants had normal or corrected-to-normal vision, were naive to the purpose of the experiment, and provided written informed consent. This study was approved by the internal review board of Waseda University. 

Apparatus

Stimuli were presented on a gamma-corrected LCD monitor (1,920 × 1,080 pixels, 23.5-in., and 100 Hz) controlled by an Apple Macintosh computer. A chin rest restrained the participants’ head movements at a viewing distance of 57.5 cm from the display. The stimulus presentation code was written in Python using the PsychoPy toolbox. 

Stimuli

We used 210 human face images, 105 monkey face images, and 105 cat face images previously used in Sarodo, Yamamoto, and Watanabe (2022). Human facial images were obtained from the Glasgow face data set (Burton, White, & McNeill, 2010). The monkey and cat facial images were obtained from the Pixabay database (www.pixabay.com). All images were grayscale and elliptical with equal size (500 × 360 pixels). The luminance of all images was calculated using the MATLAB (MathWorks, Natick, MA, USA) SHINE Toolbox (Willenbockel et al., 2010). Half of the human faces were used as adaptors and standards, and the other half were used for comparison. 

Procedure

A schematic of the stimulus sequence is shown in Figure 1. Each trial began with a white fixation cross presented at the center of the screen. The participants were instructed to fixate on the cross throughout the trial. In each trial, a human face was presented at the center of the screen as an adaptor for 2,100 ms. Either a different human, monkey, or cat face was then presented as a comparison stimulus, followed by the same human face as the adaptor for a standard stimulus for 500 ms. The duration of the comparison stimulus was randomly chosen from one of the seven durations (350, 400, 450, 500, 550, 600, and 650 ms). The interstimulus interval was 300 ms. Participants were asked to compare the durations of the comparison and standard stimuli and indicate whether the duration of the comparison was shorter or longer than that of the standard. The next trial began automatically 1,000 ms after the participants’ response. Participants were instructed not to count or tap to follow the rhythm. 

Figure 1.

 

Schematic illustration showing the time course of the adaptation paradigm. After an adaptation period of 2,100 ms, a human, monkey, or cat face was presented, and the participants compared the duration of the comparison to that of the standard.

At the beginning of the session, the participants were explicitly informed that a human, monkey, or cat face would be presented at the second of the stimulus sequence. There were three test blocks, each comprising 105 trials. Thus, each participant completed 315 trials, providing 105 responses for each face category and 15 responses for each comparison duration. The participants were asked to take a break of at least 5 min between the blocks. 

Analysis

The point of subjective equality (PSE) was calculated, which represented the comparison duration that the participants judged as equal to the standard duration. Sigmoid psychometric functions were fitted to the proportions of “comparison was longer” responses for each participant and category condition using the psignifit toolbox for Python (Wichmann & Hill, 2001a; Wichmann & Hill, 2001b). The 50% point of the psychometric function was used as the PSE for the comparison. We then calculated constant error (CE) using the equation CE = 500 ms – PSE to qualitatively measure the PSE shift from the standard duration for each condition. We used a one-way repeated-measures ANOVA to compare CEs of the face category conditions. Greenhouse–Geisser corrections were applied as appropriate adjustments for nonsphericity by changing the degrees of freedom. A significance threshold of p < 0.05 was chosen for all tests. We additionally performed a Bayesian statistical analysis on the CEs using JAPS (Wagenmakers, Love, et al., 2018; Wagenmakers, Marsman, et al., 2018) to qualitatively evaluate the data providing evidence for the null hypothesis (i.e., there is no main effect of category on the CEs). We used Jefferys interpretation of the Bayes factor to interpret the result. 

Results and discussion

Data from six participants were excluded from the formal analysis because their performance was not above the chance level, even when the comparison durations were longer than the standard durations by 30%. Figure 2A shows the averaged psychometric functions across the participants. The mean and individual CEs for each condition are shown in Figure 2B. Repeated-measures ANOVA revealed a significant main effect of category on the CE (F(2, 42) = 8.50, p < 0.001, η² = 0.29). Post hoc analyses with Bonferroni correction revealed that the CE in the human face condition was significantly smaller than that in the monkey and cat face conditions (human face–monkey face, t(21) = 3.53, p = 0.006, d_z = 0.75, BF₁₀ = 19.58; human face–cat face, t(21) = 3.66, p = 0.004, d_z = 0.78, BF₁₀ = 25.61); however, there was no significant difference in the CE between the monkey and cat face conditions (t(21) = 0.56, p > 0.99, d_z = 0.12, BF₁₀ = 0.26). A Bayesian repeated-measures ANOVA also provided very strong evidence for the alternative hypothesis (BF₁₀ = 39.03), which indicates that the CEs differ among the category conditions. 

Figure 2.

 

Result of Experiment 1. (A) The participant-averaged psychometric functions for each condition. Error bars indicate the standard error of the mean in the present study. Note that the averaged plot is only for visualization; analysis was based on individual psychometric functions. (B) The mean CE of the comparison compared to that of the standard and individual CEs in each condition (**p < 0.01).

Experiment 1 showed that the duration of human face comparisons was shorter than that of the monkey and cat face comparisons. These results suggest that face adaptation influences the perceived duration of subsequent faces in a visual category-specific manner; categorical face adaptation influences duration perception (CFAD). 

Experiment 2

Adaptation to low-level visual properties can alter the perceived duration of subsequent stimuli (Bruno & Johnston, 2010), which may explain the differences in perceived duration among the category conditions. To test this possibility, we performed a control experiment in Experiment 2 in which the configural category information of the adaptor was eliminated using Fourier phase scrambling. If adaptation to low-level visual properties accounted for the face adaptation effect, the CFAD effect would still be observed even when the category of the adaptor face was unrecognizable. 

Method

Participants

A total of 24 paid volunteers (mean age ± SD = 20.83 ± 1.97 years) were newly recruited and participated in Experiment 2. The sample size was predetermined to be the same as that of Experiment 1. 

Stimuli, procedure, and analysis

The stimuli, procedure, and analysis were the same as those used in Experiment 1, except that the adaptor was a phase-scrambled human face. A two-dimensional fast Fourier transform (FFT) was applied to each adaptor image used in Experiment 1 to obtain the magnitude and phase components of each image. Phase components were then randomized by adding a random value to the original phase from a uniform distribution across the range (–π, π). We applied an inverse FFT to the combined components of the amplitude and randomized phases to produce phase-scrambled images of the human face adaptor, as shown in Figure 3. The participants were not informed that the adaptor was a scrambled image of a human face. 

Figure 3.

 

Intact and phase-scrambled images of human face adaptor.

Results and discussion

Data from two participants were excluded from the formal analysis because their performance was not above the chance level, even when the comparison durations were longer than the standard durations by 30%. Figure 4A shows the averaged psychometric functions across the participants. The mean and individual CEs for each condition are shown in Figure 4B. A repeated-measures ANOVA revealed no significant main effect of category on CE (F(2, 42) = 0.08, p = 0.92, η² = 0.004). Post hoc analysis with Bonferroni correction also revealed no significant difference in CE among the three conditions (human face–monkey face, t(21) = 0.04, p > 0.99, BF₁₀ = 0.22; human face–cat face, t(21) = 0.36, p > 0.99, BF₁₀ = 0.24; and monkey face–cat face, t(21) = 0.45, p > 0.99, BF₁₀ = 0.24). A Bayesian repeated-measures ANOVA also provided moderate evidence for the null hypothesis (BF₁₀ = 0.133). 

Figure 4.

 

Result of Experiment 2. (A) Participant-averaged psychometric functions for each condition. (B) Mean CE of the comparison compared to the standard and individual CE in each condition.

Figure 5.

 

Schematic illustration showing the time course of Experiment 4. The adaptor was inverted.

Experiment 2 showed that the perceived duration of the comparisons did not differ when the adaptor category was rendered unrecognizable while preserving low-level image statistics. This indicates that adaptation to low-level visual properties alone cannot account for the CFAD effects observed in Experiment 1. 

Experiment 3

Phase scrambling eliminates both the recognizability of the adaptor as a human face and configural information of the face, which is important for face recognition (Helmut & Vicki, 2000; McKone, 2008). In Experiment 3, we used an inverted human face as the adaptor. Inverting a face is believed to disrupt configural processing (Fang, Ijichi et al., 2007; Rhodes et al., 2004) while the stimulus is still perceived as a face, and face category is easily recognized. Therefore, we expected to regain the pattern of CFAD (i.e., the difference in CEs) but to a smaller degree because of the reduction in configural processing. 

Method

Participants

A total of 27 paid volunteers (mean age ± SD = 21.59 ± 3.37 years) were newly recruited and participated in Experiment 3. The sample size was predetermined to be the same as that used in Experiment 1. 

Stimuli, procedure, and analysis

The stimuli, procedure, and analysis were the same as those used in Experiment 1, except that the human face adaptor was inverted (Figure 5). 

Results and discussion

Data from five participants were excluded from the formal analysis because their performance was not above the chance level, even when the comparison durations were longer than the standard durations by 30%. Figure 6A shows the averaged psychometric functions across the participants. The mean and individual CEs for each condition are shown in Figure 6B. A repeated-measures ANOVA revealed a significant main effect of category on the CE (F(1.57, 33.0) = 5.36, p = 0.015, η² = 0.20). Post hoc analyses with Bonferroni correction revealed that the CE in the human face condition was significantly smaller than that in the cat face condition (t(21) = 2.64, p = 0.046, d_z = 0.56, BF₁₀ = 3.51), whereas there was no significant difference in CEs between the other conditions (human face–monkey face, t(21) = 2.16, p = 0.13, d_z = 0.46, BF₁₀ = 1.53; monkey face–cat face, t(21) = 1.60, p = 0.38, d_z = 0.34, BF₁₀ = 0.67). A Bayesian repeated-measures ANOVA also provided moderate evidence for the alternative hypothesis (BF₁₀ = 5.60). 

Figure 6.

 

Result of Experiment 3. (A) Participant-averaged psychometric functions for each condition. (B) Mean CE of the comparison compared to the standard and individual CE in each condition.

Figure 7.

 

Schematic illustration showing the time course of Experiment 4. Comparison appeared 1,100 ms after the adaptation period.

The results of Experiment 3 revealed that adaptation to an inverted intact face caused a difference in perceived duration between the category conditions; hence, CFAD was observed. However, in Experiment 1, the CFAD effect appeared weaker when the adaptor face was upright. In particular, the effect size of the difference in the CE between the human and cat face conditions was smaller than that in Experiment 1, and there was no longer a significant difference between the human and monkey face conditions. 

Experiment 3 indicated that CFAD can be observed as long as an adaptor face and its category are recognizable but is reduced when configural information processing is disrupted, further supporting the visual category-specific, face-processing-based nature of the CFAD effect. Unlike the propositional representation of concepts with discrete spaces, perceptual or visual representations have been reported to show a quantitative or analog representation of categories along a continuum based on their neural distance (Cichy et al., 2014; Kiani, Esteky, Mirpour, & Tanaka, 2007; Kriegeskorte et al., 2008). Thus, the weaker CFAD effect—namely, the disappearance of the significant difference only between the human and monkey face conditions—may imply that adaptation to perceptual face information, rather than conceptual category information, contributes to duration distortion. 

Experiment 4

In Experiment 4, we tested the temporal aspect of the CFAD effect by increasing the interstimulus interval between the adaptor and comparison. The strength of adaptation effect on face perception is known to gradually decay over time (Burton, Jeffery, Bonner, & Rhodes, 2016; Kloth & Schweinberger, 2008; Leopold et al., 2005; Rhodes, Jeffery, Clifford, & Leopold, 2007). For example, Kiani, Davies-Thompson, and Barton (2014) demonstrated that increasing the interval between the adaptor and test faces from 300 ms to 1,650 ms significantly reduced the face adaptation effect. Similarly, Harris and Nakayama (2007) demonstrated that the reduction of the M170 component occurred only when the first and second faces were presented with an interval shorter than 600 ms. If the CFAD effect was also reduced with a longer interstimulus interval, it would support the idea that the CFAD effect is based on a process similar to the face adaptation effects reported previously (Burton et al., 2016; Kloth & Schweinberger, 2008; Leopold et al., 2005; Rhodes et al., 2007). 

Method

Participants

A total of 25 participants (mean age ± SD = 20.8 ± 1.65) were newly recruited and participated in Experiment 4. The sample size was predetermined to be the same as that used in Experiment 1. 

Procedure and analysis

The procedure and analysis were the same as those used in Experiment 1, except that the Inter-stimulus interval (ISI) between the adaptor and comparison was 1,100 ms (Figure 7). 

Results and discussion

Data from three participants were excluded from the formal analysis because their performance was not above the chance level, even when the comparison durations were longer than the standard durations by 30%. Figure 8A shows the averaged psychometric functions across the participants. The mean and individual CEs for each condition are shown in Figure 8B. A repeated-measures ANOVA revealed a significant main effect of category on the CE (F(1.58, 33.2) = 5.51, p = 0.013, η² = 0.21). Post hoc analyses with Bonferroni correction revealed that the CE in the human face condition was significantly smaller than that in the cat face condition (t(21) = 4.60, p < 0.001, d_z = 0.98, BF₁₀ = 184.49), whereas there was no significant difference in CEs between the other conditions (human face–monkey face, t(21) = 1.58, p = 0.36, d_z = 0.34, BF₁₀ = 0.66; monkey face–cat face, t(21) = 1.42, p = 0.51, d_z = 0.30, BF₁₀ = 0.54). A Bayesian repeated-measures ANOVA also provided moderate evidence for the alternative hypothesis (BF₁₀ = 5.82). 

Figure 8.

 

Result of Experiment 4. (A) Participant-averaged psychometric functions for each condition. (B) Mean CE of the comparison compared to the standard and individual CE in each condition (***p < 0.001).

Experiment 4 showed that the face CFAD effect was preserved even when the comparison was presented 1,000 ms after the disappearance of the adaptor. However, unlike in Experiment 1, the difference between the monkey and human face conditions was not significant, suggesting that CFAD was attenuated by increasing the ISI between the adaptor and comparison. The attenuation of the CFAD effect significantly corresponds with the temporal attenuation of the adaptation effects on face perception (Burton et al., 2016; Kloth & Schweinberger, 2008; Leopold et al., 2005; Rhodes et al., 2007). Thus, the results imply that the adaptation effects on face perception and CFAD effect may share similar mechanisms, such as the reduced neural response at face-selective areas (Cziraki et al., 2010; Fang et al., 2007; Kaiser et al., 2013; Kovács et al., 2008; Nagy et al., 2012). 

General discussion

To understand the role of higher visual processing in duration perception, we investigated whether face adaptation affected the perceived duration of subsequent faces in the same or different categories. We found that the human face image was perceived to last shorter in duration than that of monkey or cat face images after adaptation to the human-face image (CFAD, Experiment 1). CFAD was abolished when the adaptor face was phase-scrambled and, therefore, not recognized as a face (Experiment 2). CFAD was observed but was attenuated when the adaptor face was intact but inverted (Experiment 3). Although an intact upright face was used as the adaptor, CFAD was attenuated when the temporal interval between the adaptor and comparison was increased (Experiment 4). Our results indicated that face adaptation influenced the perceived duration of a subsequent face in a category-specific manner. This supports the notion that higher stages of visual processing and/or the consequential representation of visual objects may play a role in duration perception. 

There may be a concern that the CFAD effect may not reflect face adaptation given the relatively short adaptation period. To address this, we conducted an additional experiment to test if the 2,100-ms exposure to a human face affects categorical judgment of the subsequent faces, following the experimental paradigm of Webster, Kaping, Mizokami, and Duhamel (2004). We created morphed faces between human and monkey faces and between human and cat faces, and we asked participants to judge the category of the morphed test face after adapting to an intact human face or a phase-scrambled human face (Supplementary Figures S1 and S2). Sigmoid psychometric functions were fitted to the proportions of “test appeared monkey” and “test appeared cat” responses for each participant. The mean and individual PSEs are shown in Supplementary Figure S3. The results¹ revealed a significant main effect of adaptation condition (F(1, 20) = 24.67, p < 0.001), indicating that the morphed faces appeared more like monkey or cat faces after adaptation to the intact human face than after adaption to the scrambled human face. This bias toward the opposite category of the adapted face is consistent with Webster et al. (2004) and suggests that the adaptation period of 2,100 ms is sufficient to induce face adaptation. Thus, the CFAD effect is likely to reflect face adaptation. This was supported by our second additional experiment, which showed that CFAD disappeared when the adaptation period was as short as 500 ms.² 

The present results are in line with previous findings that adaptation to visual events, such as fast-moving stimuli (Burr et al., 2007; Johnston et al., 2006), decreases the apparent duration of subsequent stimuli. However, throughout the study, we observed positive CEs in the monkey and cat face conditions, indicating that the comparisons in these conditions were perceived to last longer than those in the human face standard. These results can be interpreted as a dilation of duration for different face categories (i.e., monkey and cat) rather than a compression of duration for the same category (i.e., human) faces. There are two possible reasons for this positive CE result. First, adaptation to the human face image might influence not only the comparison duration but also the standard duration. Experiment 4 showed that, although the CFAD effect was attenuated with increasing interstimulus intervals, the effect was maintained to some extent beyond 1,000 ms after the adaptation period. As the face image used for the standard was identical to that used for the adaptor, the perceived duration of the standard might have been compressed more than the perceived duration of the monkey and cat face comparisons, resulting in positive CEs. Second, the range of comparison durations may not be appropriate for detecting robust duration compression. In this study, comparison durations were symmetrically placed around the standard duration, and it was possible that the perceived duration of the comparison could be shorter than this range. Some excluded participants consistently reported that the comparison was shorter than the standard, even when the comparison duration was 30% longer rather than shorter, implying the possibility that some participants experienced duration compression beyond the comparison duration range. Although the influence of these factors did not obscure our conclusions, future studies should clarify this point. 

A plausible explanation for the duration distortion induced by face adaptation is the neural coding efficiency hypothesis (Eagleman & Pariyadath, 2009; Pariyadath & Eagleman, 2007). This hypothesis posits that the perceived duration of a sensory event reflects the amplitude of the neural response evoked by the sensory input. Prolonged exposure to a human face (Cziraki et al., 2010; Fang et al., 2007; Kaiser et al., 2013; Kovács et al., 2008; Nagy et al., 2012) and a successive presentation of different human faces (Coggan, Baker, & Andrews, 2019; Harris & Nakayama, 2007) are known to reduce neural responses at face-selective areas. These findings suggest that reduced neural responses induced by face adaptation correlate with the altered perceived duration of human face comparisons. This may explain the smaller CEs in the human face condition compared to those in other face category conditions. Additionally, neural response patterns evoked by human faces are more similar to those evoked by monkey faces than to those evoked by nonprimate faces (Kiani et al., 2007; Kriegeskorte et al., 2008). This implies that cat faces may show greater recovery from face adaptation than monkey faces. Thus, the coding efficiency account also fits with the disappearance of the significant difference between the human and monkey face conditions in Experiments 3 and 4, namely, the weaker CFAD effect (see Ulrich & Bausenhart, 2019, for a review of recent behavioral findings opposed to the coding efficiency hypothesis). 

The results of the present study support the previous suggestion that higher visual processing is involved in duration compression induced by adaptation (Burr et al., 2011; Burr et al., 2007; Curran & Benton, 2012; Gulhan & Ayhan, 2019; Morrone et al., 2010). However, because previous adaptation studies used a moving stimulus as an adaptor, it was more difficult to conclude that the previously observed duration compression could be attributed to higher visual stages. Indeed, several studies have suggested that adaptation during the early visual processing stage is important for duration compression induced by motion adaptation (Ayhan et al., 2011; Bruno & Johnston, 2010). Given the hierarchical nature of visual motion processing, it appears that both lower and higher visual processing stages are involved in duration distortions (Latimer & Curran, 2016). While the present study highlights the involvement of higher visual processing of visual faces, further studies are required to determine the type of visual processing that is important for duration perception. 

Similar to adaptation-based duration compression, repetitive presentation of a stimulus has been shown to reduce the perceived duration of a subsequent stimulus (e.g., Cai, Eagleman, & Ma, 2015). This repetition-based duration compression has been suggested to be the cause of the temporal oddball effect (Pariyadath & Eagleman, 2007; Pariyadath & Eagleman, 2012; Matthews & Gheorghiu, 2016), in which a novel stimulus presented within a sequence of repeated stimuli is perceived to last longer in duration compared to the preceding stimuli. We recently demonstrated that face-category processing contributes to the temporal oddball effect (Sarodo et al., 2022). In the experiments, a human face was repeated as a standard, and a different human, monkey, or cat face was presented as an oddball. The oddball changing in the face category induces a stronger temporal oddball effect. Given that the temporal oddball effect is primarily driven by the duration compression of repeated stimuli, it is possible that a common mechanism underlies the results of this study and those of Sarodo et al. (2022). However, Schindel, Rowlands, and Arnold (2011) demonstrated that stimulus-driven adaptation could not fully account for the temporal oddball effect and argued that predictive coding processes driven by the repetitive presentation of the standard stimulus were important for the temporal oddball effect. Future studies should investigate how neural adaptation and predictive coding affect perceived duration. 

The present study has some limitations. First, it is not clear whether nonhuman faces would also work as the adaptor for the CFAD effect. In the presented study, we used human faces as the adaptor because the neural representation of own-category faces is more distinct and detailed than that of other-category faces (Kreigeskorte et al., 2008), and thus an adaption to human faces was expected to be more robust than adaption to nonhuman faces. However, if CFAD is an effect based on category differences, the CE should be larger for human faces after adaptation to nonhuman faces. Second, the difference in the CE between the category conditions does not necessarily reflect only the difference in category processing, as the facial features also differ between categories. Although the results of Experiments 2 and 3 suggest that configural face processing is important for the CFAD effect, they cannot exclude the possibility that the CFAD effect reflects a different level of face processing such as face identity. Future study should address these issues. 

Acknowledgments

Supported jointly by JSPS Research Fellowships for Young Scientists 23KJ2029 to AS, KAKENHI JP21K03133 to KY, and KAKENHI (22H00090) to KW. 

Commercial relationships: none. 

Corresponding author: Akira Sarodo. 

E-mail: [e-mail]. 

Address: Faculty of Science and Engineering, Waseda University, Shinjuku, Tokyo, Japan. 

Footnotes

1  A 2×2 repeated-measures ANOVA also revealed a significant main effect of morphed category (F(1, 20) =111.7, p < 0.001), but no significant interaction was found between adaptation condition and morphed category (F(1, 20) =1.21, p = 0.286).

Footnotes

2  This additional experiment was conducted using the same method as in Experiment 1, except that the adaptation period was replaced by 500 ms. The results from 22 participants revealed no main effect of category on CE (F(1.45, 30.49) = 3.62, p = 0.052). Furthermore, post hoc analyses with Bonferroni correction revealed no significant difference in CEs between any of the pairs (monkey face–human face, t(21) = 1.53, p = 0.42, BF₁₀ = 0.61; cat face–human face, t(21) = 2.12, p = 0.14, BF₁₀ = 1.42; cat face–monkey face, t(21) = 1.70, p = 0.31, BF₁₀ = 0.76). A Bayesian repeated-measures ANOVA provided only anecdotal evidence for the alternative hypothesis (BF₁₀ = 1.75).