Draft:The face-name matching effect

The face−name matching effect is a psychological phenomenon in which a perceiver (human or computerized) is able to accurately choose the correct given name of a person, significantly above chance level, according to facial appearance alone. It is determined by culturally shared face–name prototypes − representations of how a person with a certain name looks like on average.

A possible underlying mechanism of this effect is a self-fulfilling prophecy: our name induces particular social expectations of our appearance, behaviour and personality; over time our facial features adjust to match stereotypical expectations. As a result, the more name stereotypes manifest in our appearance, the more our faces match the names, and the more frequently we are perceived like our names.

The term was proposed in 2017 by Zwebner, Y., Sellier, A.-L., Rosenfeld, N., Goldenberg, J., and Mayo, R.

Background
Face–name association was first confirmed in the 2007 study ''“Who do you look like? Evidence of facial stereotypes for male names”'' by Melissa A. Lea et al. Prior to the study, the researchers conducted a field experiment, in which they showed pictures of two men to undergraduate students and individually asked them to guess which one is named “Tim” or “Bob”. The participants' answers turned out to be consistent. This experiment formed the basis for their 2007 study, in which the previously assumed association between facial characteristics and names was formally documented.

Melissa A. Lea et al. (2007)
The 2007 study ''“Who do you look like? Evidence of facial stereotypes for male names”'' by Melissa A. Lea et al. provides evidence that people use facial prototypes when they encounter different names. In order to document a face–name association and possible cognitive consequences, the researchers conducted three experiments with Caucasian American students:


 * 1) Experiment 1: Face Prototype Creation for Specific Names. Experiment participants were divided into two groups. The first group digitally created face exemplars for 15 common male names, with each participant generating one face for one name randomly assigned out of pre-chosen 15 names, using only eyes, nose, and mouth features, creating a pool of 10 generated faces for each name. The second group was then asked to determine which of the 10 exemplars, in their opinion, represented each name the best. Faces that were considered to be the most representative by the majority of the participants were kept and further morphed into 15 face−name prototypes.
 * 2) Experiment 2: Matching. Each participant matched the same 15 male names with each of the 15 prototype faces from Experiment 1, depending on which they found more suitable for each other. In most cases (10 out of 15) faces were correctly labeled with names initially assigned to the exemplars during the previous experiment. Participants’ matching choices showed convergence in answers as some names are assigned to certain faces more frequently than chance. Therefore, face–name pairings vary in terms of association strength.
 * 3) Experiment 3: Face–Name Learning. Utilizing the same stimuli from Experiment 2 participants were asked to learn 15 face–name pairings, by matching names with random faces, throughout many learning trials that lasted until the person got 45 correct responses in a row. It was found that participants learned face–name pairings that had a higher association faster than pairings with a low association.

In these experiments, several conclusions were drawn. Firstly, there is a connection between facial features and names. Secondly, this relationship determines or even eases the learning process: highly associated (well-matched) face–name pairings are learned faster. Thirdly, names carry their own facial prototypes : when guessing the true name of a depicted individual, people evaluate which name option best meets their expectations of what it should look like (i.e., matches its facial representation) and compare it to the image.

Prior to these findings, other researchers have also observed particular implications of the face–name association. For instance, in 1993, Gillian Cohen and Deborah M. Burke revealed that names that are subjectively considered matching are recalled better. This conclusion was based on the experiment where participants were required to learn faces and their corresponding names that were either assigned by the experimenter or freely chosen by themselves so that one fits another. The explanation is that choosing a suitable name in accordance with a certain meaning constructs semantic (i.e., meaningful) associations which in turn improve name memorization.

Recent studies
The term “face–name matching effect” was proposed by Yonat Zwebner, Anne-Laure Sellier, Nir Rosenfeld, Jacob Goldenberg, and Ruth Mayo in their 2017 study ''“Who do you look like? Evidence of facial stereotypes for male names”''. They established the existence of this phenomenon by conducting several experiments, in which participants (mostly young adults) were serially shown headshot images on a computer screen with four to five name options suggested below. The participants were then asked to choose the true name of the depicted person (target) from the offered list, which also included filler names, not belonging to any face in a trial. All names appeared only once for each participant and the order of faces as well as filler names was randomized each time.

In their most basic experiment, researchers set Israeli students to guess the names for 20 photos of Israelis of around the same age. Used pictures were blindly drawn by an independent assistant from social network profile pages. Participants accurately matched the target’s true name in 28.21% of the cases, which is significantly greater than random chance (20%). The results remained even after using target names as fillers and reshuffling the stimuli (29.91% with a chance level of 25%), which proves that the face–name matching effect does not stem simply from popularity differences between target and filler names.

Zwebner et al. also succeeded in replicating this effect on a sample of Caucasian French young adults. The procedure was the same, except that a new set of stimuli was created. Participants were presented with 10 photos of previously unfamiliar Caucasian French students, who were pre-selected based on the frequency of their names. Targets' faces were accurately matched with their names significantly above the chance level (40.52% vs. 25%). These findings indicate that the face–name matching effect is cross-cultural as it can emerge in different countries and cultures.

The effect has also been found when a perceiver was represented by a computer. Researchers trained a learning algorithm (Deep Convolutional Neural Networks in particular) by presenting a set of facial images and their corresponding labels (i.e., the correct names). Based on the dataset, consisting of 36,013 photos for 15 female names and 58,018 photos for 13 male ones, the algorithm was supposed to produce a classification rule for further matching names with unobserved faces. Used headshots were taken from a business-oriented French social platform, whose members were around the same age and social status. The trained algorithm was able to accurately match true names with targets' faces significantly above the chance level, same with testing all possible combinations of names with all facial images available. These results show that the face−name matching effect is not defined by possible human bias or measurement inaccuracy; it is driven by the information conveyed by faces, rather than personal, social, and historical context, which is irrelevant for a computer performing a face−name matching task.

The fact that the effect stems from associating a name with facial information rather than names’ socioeconomic background (e.g., ethnicity, age ) was also proven in another experiment. There were two experimental conditions introduced: the first group was suggested to match names with basic headshots (all targets were of the same age and ethnic group), while the second one was proposed to guess without seeing faces. For the latter, targets’ faces remained hidden behind a black square until a choice was made. A face–name matching effect occurred only in the basic condition, and there was no difference in the frequency of guessing the filler and target names in the control group, which suggests that, unlike faces, names alone do not lead to correct matching.

Shared face–name prototypes
The ability to accurately match a name to its face could be determined by face–name prototypes, shared within a culture : due to being familiar with local names, people develop a prototype representing each of them − part of a name schema as an idea of how a name looks on average. The availability of said schema is one of the necessary conditions for the face–name matching effect occurrence. If a social perceiver lacks it because they are not familiar with facial appearances of people with a specific name, the necessary prototype will not exist, and they, consequently, will not be able to correctly guess a target's name.

The evidence for that theory was suggested by Zwebner et al. in a cross-cultural study conducted in France and Israel, in which participants matched the names with the faces of both French and Israeli targets. The face–name matching effect was found to be culture-dependent as it only occurred in situations when people completed the tasks with targets from their own culture; otherwise, their guesses were below or equal to the chance level. The data of the participants who admitted to be familiar with some target names (and therefore had appropriate name schemas) were not included in calculations.

Culture-dependence indicates that the face–name relationship alone is not enough to create an effect, as opposed to the bouba-kiki effect, in which it is the only mechanism for the accurate matching (e.g., a round face will be given a name that has a “round” universal sound, such as “Bob”).

Self-fulfilling prophecy mechanism
It suggested that the self-fulfilling prophecy is a possible underlying mechanism of the effect. Throughout life a person receives social signals related to their name expressing stereotypes and other social expectations of their personality, behaviour, lifestyle, etc. Some names are found to be more or less associated with such individual characteristics as disruptive behavior, popularity and successfulness , competence and hireability. Since the way we are socially perceived can influence our personality and personality impacts our appearance , social expectations regarding our given name (social label) are similarly hypothesized to affect our facial appearance. Thus, over time stereotypical expectations that come with a name can manifest in appearance and a face will look more suitable, which will in turn enhance the guessing accuracy in face–name matching tasks. Accordingly, expectations will not derive from a name if it is not used socially ; otherwise, perceiver's name schemas will not apply, one's appearance will not be influenced, and a face–name matching effect will not take place."Take, for example, an American girl named Katherine. Research found that people share a stereotype and perceive a woman with this name to be more successful than others (e.g., compared with a woman named Bonnie). Thus, from the day she is born, a person named Katherine receives signals from society that relate to her name stereotype, leading her to act accordingly and eventually making her more and more similar to our shared representation of a stereotypical “Katherine.” Thus, social signals related to a name might affect one’s self perception and personality, and to some extent, one’s facial appearance."

- Zwebner, Y., Sellier, A.-L., Rosenfeld, N., Goldenberg, J., & Mayo, R. To gather evidence of the self-fulfilling prophecy theory, Zwebner et al. conducted a study, in which they manipulated the social usage of a name. For this purpose, they recruited targets who did not use their name in real life and who were typically addressed by their nickname that was unique and not related to the given name or appearance (e.g., "Bob" as a short version of "Robert" is not exclusive). There were two experimental conditions: the first group was shown images of nicknamed people with their real name, the second one – with their nicknames, and both were also exposed to targets that went only by their given names. Two given names and two nicknames appeared as choice options, each one being a filler. A significant difference was found between matching faces and names of targets that went by their given names and those who used nicknames on a regular basis: the face–name matching effect for the latter was less evident and strong.

Another argument in favour of the self-fulfilling prophecy idea is that the face–name matching effect prevails when being exposed to facial features that humans control most (e.g., their hairstyle ), as these regions are very informative for matching task execution. This was observed in a study with three facial information conditions: the first group was exposed to basic target headshots, the second group – to faces excluding any exterior features (e.g., hair, neck, ears), the third was shown merely targets' hairstyles. 10 standard images were photoshopped, removing all external facial features or leaving a hairstyle only. While the effect persisted in all conditions, the hairstyle alone was enough to generate a greater effect than inner facial features. Being able to control face features allows people to fulfill themselves according to the societal expectations with regard to their name.

Although inner facial features are less easily controlled, they are active from various facial expressions emanating from emotions, individual characteristics, or social interaction responses, resulting in physical changes of appearance. Facial features were explored in another research, in which a previously trained algorithm was used to generate heat maps, illustrating how informative different inner facial components are in matching them with French names. Each name has specific facial areas that are crucial for accurate matching, most of which are the parts related to our facial expressions (around the eyes and the mouth, etc).

Criticism
The decreased matching ability for targets who socially used their nicknames more often than given names can be explained alternatively. An infrequent choice of a nickname as a target's true name can arise not only from the absence of a name schema, required to associate a name with a face, but simply from a tendency to avoid choosing an uncommon or even bizarre option as it seems less likely to be true[1]. Further research must be conducted to eliminate this uncertainty and examine whether the self-fulfilling prophecy truly is the mechanism responsible for the face–name matching effect.

Moreover, there is a contradiction in the research. For instance, in the 2015 study Robin S. S. Kramer and Alex L. Jones denied the existence of an association between names and faces in their 2015 study "Do people’s first names match their faces?". They cross-culturally observed two samples of Caucasian participants from the USA and UK and found little or no effect: while American women were able to accurately match American faces with names significantly above chance level, American men and the British of both genders failed to perform the matching task. Researchers then considered the initial result to be false positive. The reason for the contradiction can be the differences in methodology : instead of presenting a photo along with a name list, they showed two images at the same time and asked which target matches a certain name more.