Face-based categorization, context-based categorization, and distortions in the recollection of gender ambiguous faces

Abstract

In three studies, we examined the impact of face-based and context-based categorization in the recollection of gender ambiguous faces. Gender ambiguous faces were created by morphing male and female source faces. In Study 1, the recollection of moderately ambiguous faces (i.e., 70% male–30% female faces and 70% female–30% male faces) was accentuated towards face distracters that were more typical of the spontaneous (i.e., face-based) categorization of these faces. In Study 2, the recollection of extremely ambiguous faces (50% male–50% female faces) was accentuated towards face distracters that were more typical of the gender category suggested by context cues attached to these faces prior to the face presentation. Study 3 relied on the same design as Study 2, but this time context cues were provided after face encoding. In line with predictions, no accentuation effect emerged under the latter conditions. The theoretical and practical implications of these findings are discussed.

Introduction

Face memory is of utmost importance in everyday life. Accurate face memory allows individuals to recognize their relatives, friends, and colleagues, the people they like and those they would rather avoid. In sum, face memory is a prerequisite for individuals to behave appropriately in their social environment. Face memory, however, is not immune to biases, whose consequences may be disastrous, in particular in the context of eyewitness testimonies. Mistaken identification seems to be responsible for more wrongful convictions of innocent people than all other causes combined (Wells et al., 1998). As a matter of fact, the recent use of DNA analyses by forensic scientists exonerated numerous innocents who had been jailed on the basis of inaccurate eyewitness testimonies. The consequences of face memory for people’s social adjustment point to the necessity of better understanding the biases that impair face recollection and identification. In the present research, we focused on the impact of categorization on the recollection of gender ambiguous faces. We hypothesized that the recollection of gender ambiguous faces would be distorted towards faces more typical of the female (male) category when considered as belonging to the female (male) category. We also investigated the impact of both face-based and context-based categorization of the faces in driving memory distortions. Finally, we examined the impact of context cues provided before and after face encoding.

Accentuation effects due to categorization have received extensive attention in cognitive and social psychology. Overall, this literature shows that the categorization of stimuli that vary from each other by constant amounts on a given continuum enhances the perception of differences for cross-category stimuli and the perception of similarities for within-category stimuli. Classic work on categorical perception (e.g., Harnad, 1987) has shown that people discriminate better between cross- than within-category stimuli. For instance, keeping constant the physical differences that exist between two hues, people discriminate better between two hues that belong to two different colors than to a same one (e.g., Bornstein & Korda, 1984).

Earlier work in social psychology provided converging findings at the level of physical estimates (Tajfel & Wilkes, 1963; see also Corneille, Klein, Lambert, & Judd, 2002). This general accentuation principle was applied to a host of physical and social judgments, including body weight (Krueger, Rothbart, & Sriram, 1989), personality traits (Krueger & Rothbart, 1990), and attitudes (Eiser and Stroebe, 1972, Eiser and Van-der-Pligt, 1982, McGarty and Penny, 1988). As a whole, this literature suggests that categorization not only derives from, but also alters the perception of similarities and differences between objects (e.g., Corneille and Judd, 1999, Goldstone, 1994, Goldstone, 1995, Livingston et al., 1998).

Surprisingly, only a few studies investigated the impact of categorical accentuation on memory in general, and face memory in particular. As an exception, Krueger and Clement (1994) found that people overestimate temperature variations between different months of the year and underestimate those variations within a same month. Studies conducted under the ‘Who-saidwhat’ paradigm (Taylor, Fiske, Etcoff, & Ruderman, 1978) also revealed that people are more likely to misattribute a statement to within- than to cross-category individuals. For example, one would more readily misattribute a statement issued from a male speaker to another male than from a male speaker to a female speaker (and the opposite for a female speaker). This literature, however, did not examine face memory effects per se, and did not provide either evidence for accentuation effects in face memory.

Recently, however, a few studies have started to examine accentuation effects in the context of face memory. For instance, Halberstadt and Niedenthal (2001) investigated accentuation effects in the memory of facial expressions and showed that faces conceptualised in terms of anger or happiness are later remembered as displaying these emotions more widely than they actually did. Closer to our present research interest, using a morphing procedure, Corneille, Huart, Béquart, and Brédart (2004) showed that race ambiguous faces are recollected as more typical of their race category than they actually are. Specifically, these authors generated target faces lying at 30, 50, and 70% on continua involving face endpoints from two different ethnic categories (e.g., ethnic categories A and B). In a recollection task, participants were then presented with the previously seen target faces along with face distracters that were located closer to the A and B endpoints of the continua (i.e., faces that were more typical of the A and B categories than the target faces were). The task of the participants was to select the previously seen target faces among the distracters. The findings revealed that, when initially presented with a 30% target face (i.e., a face mostly composed of the A source face), participants were more likely to select distracters that were closer to the A source face (i.e., a 20% morphing distracter) than distracters that were closer to the B source face (i.e., a 40% morphing distracter). Conversely, when initially presented with a 70% target face (i.e., a face mostly composed of the B source face), participants were more likely to select distracters that were closer to the B source face (i.e., an 80% morphing distracter) than distracters that were closer to the A source face (i.e., a 60% morphing distracter).

The memory distortion effect reported by Corneille and colleagues (2004) was obtained on ambiguous Caucasian, Asian, and North African faces. This effect emerged with high levels of confidence and lasted over a period of one week. Additional findings confirmed that this accentuation effect related significantly to the participants’ categorization of these face. Hence, faces categorized as either Caucasian, Asian or North-African were recollected as being, respectively, more Caucasian, Asian or North-African than they actually were. Finally, this effect was shown to be independent from a mere distortion of memory towards distinctive or unmorphed faces. Indeed, recollections could have been distorted towards faces more extreme on the continua not because of accentuation effects, but because those faces were somewhat more distinctive (Rhodes, Byatt, Tremewan, & Kennedy, 1997), or simply because of morphing artefacts. Two studies, comparing recollections of faces morphed on continua that involved source faces from either different categories or from a same one, allowed ruling out these alternative accounts of data. Indeed, stronger memory distortions were obtained when the source faces were from two different ethnic categories than when they were from a same one.

The prediction that face recollection may be distorted towards more typical ethnic faces was based on the general idea that people rely on category and schematic information in the recollection process. In particular, Huttenlocher and her colleagues (Huttenlocher et al., 1988, Huttenlocher et al., 2000; see also Bartlett, 1932, Brewer and Nakamura, 1984) proposed a model of episodic memory according to which two kinds of information are encoded when presented with a stimulus: fine-grained information inherent to the stimulus and categorical information self-generated by the perceiver. The poorer the encoding or storage of the fine-grained information, the more likely people are to combine this information with category information at recollection (categorical information will suffer of a slower rate of forgetting than item information; see Dorfman & Mandler, 1994). This process was coined “weighting with a prototype” by Huttenlocher and colleagues.

As just explained, the general notion conveyed by this model is that categorical information is spontaneously attached to a stimulus at the encoding stage, which will be combined to fine-grained information at recollection, resulting in recollection distortions towards exemplars more typical of the category. This model was applied to basic stimuli so far (e.g., the localization of dots in a circle), but it is defined as a general model of episodic memory, and there is therefore no reason why it should not apply to more complex stimuli (Corneille et al., 2004), and to other kinds of percepts (Stern, Corneille, Huart, & Mullennix, 2004).

In studies by Corneille and colleagues (2004), accentuation effects were expected to emerge as a function of face-based categorization for moderately ambiguous faces (i.e., 70% Caucasian, Asian, and North African faces), and as a function of context-based categorization for extremely ambiguous faces (i.e., 50% Caucasian–Asian faces and 50% Caucasian–North African faces). Specifically, the authors predicted that participants would spontaneously categorize moderately ambiguous race faces according to the features inherent to these faces. In contrast, they predicted that participants would actively seek out context cues, such as social information externally associated with a face, to figure out the race of an extremely ambiguous race face. Whatever the cues used for categorizing, in both cases categorization was expected to result in recollection distortions consistent with the category assigned to the stimuli (i.e., an accentuation effect).

Whereas strong empirical support was obtained for the first prediction (i.e., accentuation effects due to face-based categorization of moderately ambiguous faces), the authors repeatedly failed to obtain any support for the second prediction (i.e., accentuation effects due to context-based categorization of extremely ambiguous faces). For instance, participants who were exposed to a 50% Caucasian–North African face were insensitive to social information suggesting that this face was that of a Caucasian (e.g., François) or North African (e.g., Mohammed) person. The context information associated with these extremely ambiguous faces had no impact on the categorization and later recollection of this face, irrespective of whether this information was delivered before, during, or after the face presentation, and irrespective of the nature of this information (e.g., stereotypic information, ethnic label attached to the face).

To explain the absence of effect of context information, Corneille et al. (2004) speculated that the extremely ambiguous race faces they used may have been spontaneously assigned to a third race category that appeared inconsistent with the context information associated with these faces. For instance, participants may have been reluctant to categorize a 50% Caucausian–50% North African face in any of these two categories, simply because these had spontaneously assigned these faces to a third ethnic category (e.g., Hispanic, South Italian) which had little to do with the social information provided about it.

If this reasoning is correct, then social information externally attached to extremely ambiguous faces may influence the categorization and recollection of these faces in the context of continua that exclude the presence of intermediate categories. We thought that gender was a good candidate here as gender obviously involves only two exclusive categories.

The assumption that continua of gender ambiguous faces would elicit more category exclusivity than the continua of race ambiguous faces examined in Corneille et al. (2004) was tested in the pilot study of the present paper. This was a prerequisite for investigating accentuation effects due to context-based categorization in face memory (Studies 2 and 3), since we reasoned that context-based accentuation towards face continua endpoints should be observed for faces varying on dimensions that exclude intermediate categories. Then, we report three studies concerned with the impact of face-based (Study 1) and context-based (Studies 2 and 3) categorization in the recollection of gender ambiguous faces.

In Study 1, we sought to test for the generality of Corneille and colleagues (2004)’s findings in the context of new, socially relevant, and highly familiar, face categories, i.e., gender faces. Participants were exposed to moderately or extremely ambiguous faces created by a morphing program. We predicted that moderately male faces (i.e., 30% female–70% male faces), and moderately female faces (i.e., 30% male–70% female faces), would be spontaneously categorized on the basis of their prevalent gender features. Hence, we anticipated that the recollection of moderately female (male) faces would be accentuated towards faces more typical of the female (male) category. No systematic classification, and therefore no systematic distortion, was expected for extremely ambiguous faces (50% male–50% female faces) in these conditions.

In Study 2, we sought to test for the possibility of context-based distortion effects in the recollection of gender ambiguous faces. We predicted that the categorization of an extremely ambiguous face would reveal assimilation to the gender information communicated prior to the face presentation. As classically shown in priming studies, people are known to rely on relevant cues that are accessible to them when facing uncertain categorization tasks (for a review, see Higgins, 1996). In turn, we anticipated that an extremely ambiguous gender face would be recollected as more male (female) than it actually was, when previously associated with a male (female) gender information.

In Study 3, we aimed to test for the idea that accentuation effects due to context-based categorization operate only when the context information is presented at the pre-encoding stage. As explained before, Huttenlocher and colleagues’ model posits that people spontaneously generate categorical information at the encoding. If so, context information that comes after the face encoding should have little impact on the categorization/recollection of a face. Indeed, people may have a very difficult time challenging retrospectively the category information they spontaneously attached to the face as they first saw it (a self-generated categorization of the face which may, in addition, have little to do with gender in the case of extremely ambiguous gender faces).

Study 3 relied on the exact same design as Study 2, except that gender information was this time delivered after exposure to the face. If we correctly assume that self-generated categorization takes place spontaneously at face encoding, then no impact of context information delivered after face encoding should be observed on face categorization and recollection measures. Because Study 3 basically tested for a null hypothesis, we increased the N size by 60% between Studies 2 and 3. Direct comparisons were also made between the two studies, in which we examined the impact of context information timing in the categorization and recollection of extremely ambiguous faces.

Thirty-six undergraduate students at the Catholic University of Louvain, Louvain-la-Neuve, were contacted on an individual basis. They were presented with one extremely ambiguous race face and one extremely ambiguous gender face (in a counterbalanced order). They were asked to report what they thought was the gender and nationality (or origin) of each of these two faces. The materials consisted in the extremely ambiguous race faces created by Corneille et al. (2004; i.e., four 50% Caucausian–50% North African faces, and four 50% Caucausian–50% Asian faces), as well as the four extremely ambiguous gender faces (i.e., 50% Female–50% Male faces) used in the studies of the present paper, which are described in details in Study 1.

Depending on participants’ response profile, we assigned them into one of three categories. In the first one, participants could assign both gender and race ambiguous faces to one of the source faces categories (e.g., the gender ambiguous face categorized as female, and the Caucasian-North African face categorized as North African). In the second, the gender ambiguous face could be attributed one of the source faces categories, whereas the race ambiguous face could be attributed a new face category (e.g., the gender ambiguous face categorized as female, and the ambiguous Caucasian-North African face categorized as Hispanic). Finally, in the third, the race ambiguous face could be assigned to one of the source faces categories, whereas the gender ambiguous face could be assigned to a new face category. The observed frequencies for these three categories of responses were respectively 22, 14, and 0, Chi_(2) = 20.67, p < .01. Thus, in line with predictions, the extremely ambiguous race faces used by Corneille and colleagues (2004) were frequently assigned to an intermediate category (i.e., Caucasian-North African faces were sometimes categorized as South Americans, Poles, Spanish, or Italians; Caucasian–Asian faces were sometimes categorized as South Americans, North Africans, Spanish, or Italians), whereas extremely ambiguous gender faces were always (not surprisingly) categorized either as male or female.