As illustrated in Figure 1 and in Table 1, we integrate findings in the field of face adaptation research into a conceptual framework that includes three dimensions. The first dimension of this framework represents different types of potential facial information which are susceptible to adaptation; we call this dimension adapting information. Instances of adapting information are identity information (e.g., Leopold et al., 2001; Jiang et al., 2006), configural information (e.g., Carbon et al., 2007b; Little et al., 2008), gaze information (e.g., Jenkins et al., 2006), emotional information (e.g., Webster et al., 2004; Ng et al., 2008; Adams et al., 2010), age information (e.g., Schweinberger et al., 2010), gender information (e.g., Kovács et al., 2007; Bestelmeyer et al., 2008), ethnicity information (e.g., Jaquet et al., 2008; Ng et al., 2008), attractiveness information (e.g., Anzures et al., 2009), or viewpoint information (e.g., Chen et al., 2010). The present list of adapting information is completed by face information investigated in the context of face distortion aftereffects (FDAEs; e.g., Webster and MacLin, 1999). This method of manipulating faces may affect types of facial information that are listed above (e.g., configural information, age, identity, gender; please see also later sections). However, FDAEs are realized by unique manipulation algorithms (i.e., distortions by expanding or contracting the frontal-view original face image relative to a midpoint on the nose). Further, these algorithms relax the controlling of which specific information types are affected. For example, manipulating faces in the context of FDAEs affect facial features (e.g., eyes, mouth, etc.) while manipulations of configural information (i.e., spatial distances between these features) rather leave these features unaffected. Generally, these different types of instances of adapting information realize different levels of ecological validity. While differences in age, viewpoint, gaze, or emotion are plausible and realistic in an ecological context, manipulations of identity or configural information have less validity since such changes do not typically occur in the ecosystem.

The second dimension of the present framework, time, orders adaptation effects according to different types of temporal information. The first information type, delay, is related to the robustness and sustainability of adaptation effects; basically, the time interval between an adaptation and an adaptation test phase. Delays range from milliseconds (e.g., Leopold et al., 2001; Rhodes et al., 2003) to minutes (e.g., Carbon and Leder, 2006; Kloth and Schweinberger, 2008), but also include days and even weeks under typical laboratory (e.g., Carbon et al., 2007b; Carbon and Ditye, 2011) as well as ecologically more valid test contexts (Carbon and Ditye, 2012). The delay characteristics of adaptation effects are essential for providing useful information about the decay of adaptation effects and thus the “recalibration” and “readaptation” abilities of the visual system (Carbon and Ditye, 2011). Furthermore, they allow inferences about the robustness and consistency of perceptual information in general. In parallel to the “time” information delay, we focus on adaptation duration, the time span during which the adapting stimulus is presented (e.g., Strobach et al., 2011). Adaptation duration information provides insights into how this time span can modulate the adaptation effect size or the adaptability of faces. Moreover, this type of time information was compared with simple adaptation effects (e.g., with tilt information; Leopold et al., 2005; Rhodes et al., 2007) to explore the dynamics of adaptation effects at different levels of cortical visual hierarchy. Finally, we focus on “time” information of the test duration type, establishing the time span during which the test stimulus is presented (e.g., Rhodes et al., 2007). Similar to the adaptation duration, test duration can give insights into to the dynamics of adaptation in face stimuli in contrast to simple adaptation effects.

The third dimension in the present framework is associated with the transfer of adaptation effects. This transfer dimension reflects the range and limits of adaptation transfer effects providing important inferences about the nature of processing being linked with specific adapted stimuli or being of more general quality. In this way, the investigation of adaptation is a tool (rather than a topic) for localizing the plasticity and pointing out common coding principles of various levels of visual processing (from retinotopic to high and possibly face-specific levels of visual processing; Webster, 2011; Webster and MacLeod, 2011). There exists two systems of structuring transfer effects: transferring between different (manipulated) image versions of the identical identity during adaptation and test phases (e.g., variations in size or orientation) enables exclusive low-level perceptual effects of adaptation (e.g., on a retinal level; Zhao and Chubb, 2001) to be excluded. Additionally, as proposed by Carbon et al. (2007b), adaptation transfer can be systematically tested with face images used in the adaptation and test phases showing the same images of the same identity vs. different images of the same identity vs. different images of different identities².

The approach can be extended by investigating transfer effects, inter alia, across family members, gender, and/or ethnicity.

As will be seen later, not all studies in the field of face adaptation research allow a localization of its research in all dimensions of the applied framework. For example, many studies apply sets of face images of different identities during adaption and the identical set of images during a following test phase (e.g., Buckingham et al., 2006; Chen et al., 2010). Such a procedure prevents conclusions about the transferability of adaptation effects, since potential effects in an adaptation test phase may originate from the presentation of the identical image and/or the presentation of other identities’ images during the adaptation phase.

Basically, the result patterns in studies on adaptation effects showed an adaptation bias and were thus consistent in the following way: values of adaptation test ratings tend toward the (typically extreme) values of adapting information presented during the adaptation phase; in other words, average or neutral faces are perceptually biased away from the adapting face. After introducing findings in the context of FDAEs, we show findings of facial information loosely ordered with increasing abstractedness.

After reviewing adaptation effects of different face information and time characteristics (e.g., the sustainability of adaptation effects), it is essential to review the relationship between the adapting and test face images; i.e., to test transfers of adaptation effects to new face images (or new image versions) not presented during adaptation. Here, we review findings that investigated transfer effects between the same image of one identity (identical images or images differing in viewpoint, orientation, or size between adapting and test images) and different images of the same identity. Additionally, we also discuss adaptation transfer effects between images of different identities. As illustrated in Table 3, we test these transfer effects at the pictorial level (identical face image during adaptation and test), identity level (different face images of the same identity during adaptation and test), and novel level (different face images of different identities during adaptation and test, Carbon and Ditye, 2011). In the following section, we primarily discuss types of adapting face information that demonstrate transfer effects. We followed this strategy because, for the remaining types of face information (e.g., gaze, emotion), there was (A) no investigation of transfer effects and/or (B) no conclusive evidence for such effects.

When focusing on FIAEs, Hole (2011) demonstrated adaptation effects with identical face images during adaptation and testing which also transfer onto new image versions changed in viewpoints, orientation, and vertically stretched versions from the adapting face images when using familiar; this confirmed Jiang et al.’s (2006) finding of viewpoint invariance of FIAEs who also added evidence of their transfers across shape and surface reflectance information. Anderson and Wilson (2005) supported Hole’s finding of size independent transfer with unfamiliar, synthetic faces while their study provided no support of a viewpoint-invariant FIAE for this face type. Guo et al. (2009) revealed limits to the transferability of the FIAEs by showing that this effect exclusively works from upright to inverted orientation, but not vice versa with unfamiliar faces. With familiar faces, however, Hole showed that the FIAE produced by inverted adaptation faces and upright test faces was similar to that produced by upright adapting faces. Furthermore, this type of adaptation effect seems to be gender-specific since there is an effect from adaptation to test faces when these faces are related via a gender-specific prototype, whereas there was no such effect with an androgynous face (i.e., combined male and female prototype; Rhodes et al., 2011). Leopold et al. (2001, 2005) showed that relations between a facial prototype and the individual faces in face space could be manipulated by face adaptations of different identities. In other words, this manipulation includes face images during adaptation that are not located on a trajectory crossing an original face and a face-space average (similar to Figure 1); thus, there is evidence for FIAEs on the novel level. This conclusion is consistent with the fact that FIAEs are rather high-level perceptual effects: composite faces (different views of a composite face comprised of the top half of a famous face and the bottom half of a non-famous face) either did or did not produce a FIAE depending on whether or not the famous face is explicitly recognized before the post-adaptation test phase (Laurence and Hole, 2012).

Conversely, adaptation to facial expressions (i.e., emotional information) was partly independent from the represented identities. That is, adaptation effects with a focus on facial expressions were transferred to different faces and thus include at least portions of novel-level processing (Fox and Barton, 2007). Interestingly, the above discussed FIAE is not affected by expressional information. That is, FIAEs were not modulated by congruency of facial expression during adaptation and test phases (i.e., same expression vs. different expression; Fox et al., 2008). Thus, expressional adaptation and FIAEs tend toward asymmetry with impact of identity information on expression adaptation, but there is no reverse effect.

For facial gender, Yang et al. (2011) demonstrated that gender discrimination enhancement induced by face adaptation can transfer across a substantial change in three-dimensional facial orientation. Additionally, gender-adaptation effects are position invariant effects (Kovács et al., 2007). These effects also seem to be age-independent, since Barrett and O’Toole (2009) demonstrated an effect of gender adaptation within sets of children’s and adults’ faces and also between these sets of faces. These age-independent effects also demonstrate that gender-adaptation effects may work on a novel level since adaptation and test faces were derived from different identities. However, this novel level is limited to the orientation of faces; that is, adaptation effects work independently with upright and inverted face presentations (Watson and Clifford, 2006). Alternatively, the limitation of emotion adaptation effects is set at race boundaries: adapting to one type of emotion in, for instance, a Caucasian face, affects the later processing of an alternative Caucasian face but not that of an alternative ethnicity (i.e., black faces; Otten and Banaji, 2012).

The adaptation effect realized in the form of viewpoint adaptation (i.e., adaptation to left or right-turned faces) occurs at the novel level as demonstrated by transfer effects across different identities, different gender, and different vertical orientations (Fang et al., 2007). In the case of face normality ratings and their adaptation effects, there exists evidence for at least orientation-transferable adaptation effects, i.e., between upright and inverted orientations of face images (Rhodes et al., 2003).

Transfer effects of adapting configural information are not only in action at the pictorial and identity levels, but also at the novel level (i.e., different face images of different identities during adaptation and test; Carbon et al., 2007b; Carbon and Ditye, 2011); even though the effect was slightly reduced compared with pictorial and identity conditions. This was demonstrated by the transfer effects of adapting configural information in combinations of identical adaptation and test facial images, of different adaptation and test facial images from the same identity and transfers to new identities, i.e., transfers to test face images of identities not presented during prior adaptation (Walton and Hills, 2012, for comparable results with exclusive eyes shifts in the vertical axis). Little et al. (2008) assumed that such transfer effects are gender-specific. FDAEs are not transferable to images mirrored after the adaptation phase (Morikawa, 2005), but there is evidence of the transfer of such effects to different ethnicities (Jaquet et al., 2007), between different viewpoints (Jeffery et al., 2006, 2007), different orientations of upright and inverted faces (e.g., adapting face is oriented 45° from vertically upright and the test face 45° in the opposite direction; Watson and Clifford, 2003) between different identities and orientations (Webster and MacLin, 1999) as well as different facial image sizes (Zhao and Chubb, 2001; Yamashita et al., 2005). Consequently, FDAEs occur up to the novel level. Consistently, adaptation effects of facial age on a novel level are demonstrated by transfers between different genders (O’Neil and Webster, 2011) and identities (Schweinberger et al., 2010).

In sum, there is clear evidence for adaptation effects across different identities for a first set of face information (e.g., gender, age, configural information), i.e., transfer effects on a novel level. Adaptation of this set of information seems to affect the high-order visual system and/or memory representations. In contrast to these transferable adaptation effects, there is no clear evidence of transfer effects for other face information, such as attractiveness. Furthermore, there is evidence that some face information is only transferable between different identities when the specific subgroups are not changed simultaneously (e.g., FDAE transfers are gender-specific). To learn about processing characteristics and mental representations of faces, this section indicates that face coding is hierarchically structured with an orchestration of common underlying structures. This common structure was demonstrated at the novel level and maybe theoretically represented in a prototype in face space (Valentine, 2001). However, the processing of some facial aspects is characterized by and related to specific modules (e.g., gender-specific modules of FDAEs) potentially working in parallel to a general face-space prototype.

A summarizing overview of existing and lacking research in the field of face adaptation effects is illustrated in Table 4. Future studies may apply the present framework’s dimensions or operational parameters (adapting information, time, and transfer) for a systematic continuation of investigating face adaptation effects. For instance, for a number of adapting information types (e.g., emotion, age, attractiveness, gender) there exists no test for the robustness of adaptation effects; that is, adaptation effects of these types of face information are demonstrated after short delays between adaptation and adaptation test phases. However, there are no studies that test these adaptation effects after long delays and their decay over time. To present one specific example, in accordance with time intervals applied by Carbon and colleagues (Carbon et al., 2007b; Carbon and Ditye, 2011), the adaptation effect of facial age should be tested after time intervals of 5 min, 24 h, and 1 week in order to cover a broad range of time periods and to test the robustness of the age adaptation effect. Likewise, testing the impact of adaptation and test time should be extended to forms of facial information beyond phenomena investigated in the context of FDAEs (Leopold et al., 2005; Rhodes et al., 2007). As illustrated, a first extension of investigations on adaptation time occurs in the context of adaptation of configural information (Strobach et al., 2011), but other contexts are definitely needed to generate a broader picture of timing aspects in face adaptation. In this way, our framework is able to characterize gaps in the adaptation research literature while combining findings along the dimension adaptation information and delay.

Similarly, such an expansion of tests should also be performed on the transfer dimension since this type of test is essential for investigating the functional level of adaptation effects and increasing the ecological validity of these studies. While transfer tests on the same face image of the same identity (e.g., varying orientation or context during adaptation and test; Carbon and Ditye, 2012) enable investigations on picture or “iconic” processing (Carbon, 2008), and thus not necessarily on face processing per se, transfer tests on different face images of the same identity as well as different identities’ face images allow for conclusions about the characteristics of face representations. So far, these two types of transfer tests [i.e., (1) different face images of the same identity, (2) different face images of different identities] were separately conducted in most studies. Future studies may combine these two transfer types.

An additional way to continue face adaptation research, in terms of gaining knowledge on the basis of adaptation effects, is to combine aspects (i.e., our framework’s dimensions) of time and transfer. For instance, future studies should systemically investigate the effects of manipulating face information across different delays between adaptation and test phases, combined with tests for the different transfer levels between adaptation and test faces. Additionally, the effects of gender adaptation can be investigated after relatively short and long delays between faces of the same or different age groups, ethnicities, or gaze points. A related investigation focused on transfer effects between different emotional expressions and gender in the context of FDAEs (Tillman and Webster, 2012). This systematic investigation of the interplay of delay and transfer may provide conclusions about the range of adaptation effects and their origin from similar or different methods of neural coding.

Additionally, future studies should relate investigations on the functional level of face adaptation effects to concepts applied to other types of processes or skills. One option might be in relation to improved skills acquired during cognitive training (e.g., working memory training) and their range of transferability. Existing studies on this issue (Li et al., 2008b; Karbach and Kray, 2009; Strobach et al., 2012a,b) categorize the range of skill transfer into near transfer (transfer between situations with common basic characteristics) and far transfer (transfer between situations with structural differences). Conversely, near transfer tests could investigate the adaptation effects of facial images on other facial images, while far transfer tests investigate adaptation effects on facial images after prior adaptation to specific properties of cars (Carbon, 2010) or adaptations to mental sets due to the presentation of gender-typical objects (e.g., lipstick vs. motor bike, Javadi and Wee, 2012). For instance, one could test whether there is a transfer of adaptation effects of configural information (i.e., spatial relations between features) from face stimuli to car stimuli (i.e., far transfer). In particular, front views of cars with a similar setting of parts as can be found in faces are favorable for performing such transfer tests (see Windhager et al., 2010). In fact, this type of future transfer study would cross boundaries between different types of visual objects and may show similarities and differences in these general object characteristics.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.