Although previous research in ERPs has focused on the conditions under which faces are recognized, less research has focused on the process by which face representations are acquired and maintained. In Experiment 1, participants were required to monitor for a target “Joe” face that was shown among a series of nontarget “Other” faces. At the halfway point, participants were instructed to switch targets from the Joe face to a previous nontarget face that is now labeled “Bob.” The ERP analysis focused on the posterior N250 component known to index face familiarity and the P300 component associated with context updating and response decision. Results showed that, in the first half of the experiment, there was increase in N250 negativity to the target Joe face compared with the nontarget Bob and designated Other face. In the second half of the experiment, an enhanced N250 negativity was produced to the now-target Bob face compared with the Other face. Critically, the enhanced N250 negativity to the Joe face was maintained, although Joe was no longer the target. The P300 component followed a similar pattern of brain response, where the Joe face elicited a significantly larger P300 amplitude than the Other face and the Bob face. In the Bob half of the experiment, the Bob face elicited a reliably larger P300 than the Other faces, and the heightened P300 to the Joe face was sustained. In Experiment 2, we examined whether the increased N250 and P300 to Joe was because of simple naming effects. Participants were introduced to both Joe and Bob faces and names at the beginning of the experiment. In the first half of the experiment, participants monitored for the target Joe face and at the halfway point, they were instructed to switch targets to the Bob face. Findings show that N250 negativity significantly increased to the Joe face relative to the Bob and Other faces in the first half of the experiment and an enhanced N250 negativity was found for the target Bob face and the nontarget Joe face in the second half. An increased P300 amplitude was demonstrated to the target Joe and Bob faces in the first and second halves of the experiment, respectively. Importantly, the P300 amplitude elicited by the Joe face equaled the P300 amplitude to the Bob face, although it was no longer the target face. The findings from Experiments 1 and 2 suggest that the N250 component is not solely determined by name labeling, exposure, or task relevancy, but it is the combination of these factors that contribute to the acquisition of enduring face representations.
People are “face experts” to the extent that we can distinguish hundreds, if not thousands, of unique faces that have been learned over the course of a lifetime (Bahrick, Bahrick, & Wittlinger, 1975). Although much of the research in ERPs has focused on neural components that index faces from nonface objects (Bentin, Allison, Puce, Perez, & McCarthy, 1996) and the recognition of previously familiar faces (Caharel, Fiori, Bernard, Lalonde, & Rebaï, 2006; Jacques & Rossion, 2006; Caharel et al., 2002; Schweinberger, Pickering, Jentzsch, Burton, & Kaufmann, 2002; Bentin & Deouell, 2000; Eimer, 2000), less attention has been paid to understanding how the representations of these faces are formed and reinforced and become permanent in memory. One goal of the current study is to investigate the neurophysiological processes of face learning; that is, the changes in brain activity that occur as a novel face becomes familiar as a consequence of practice and exposure.
Previous ERP studies have established a connection between the N170 posterior-occipital component and facial perception. Specifically, approximately 170 msec after stimulus onset, a larger negative deflection is generated in response to human faces relative to objects, animal faces, or body parts (Bentin et al., 1996). Although inversion has been shown to disrupt face perception (Lewis & Edmonds, 2003; Rousselet, Mace, & Fabre-Thorpe, 2003; Purcell & Stewart, 1988; Yin, 1969), it does not abolish the face N170 but increases its amplitude or delays its latency (Rossion et al., 2000). Other manipulations such as priming low-level facial features (Bentin, Sagiv, Mecklinger, Friederici, & von Cramon, 2002), isolating facial features (Bentin et al., 1996), or even changing features' positions in a face (George, Evans, Fiori, Davidoff, & Renault, 1996) are sufficient to evoke a robust face N170 response.
Although the N170 component is sensitive to the general category of faces, it is not sensitive to distinct faces within the category. For example, no difference in N170 amplitude was observed between famous faces (such as politicians and actors) and unfamiliar faces (Bentin & Deouell, 2000; Eimer, 2000). As a result of these findings, it has been suggested that the N170 component indexes the presence of a face stimulus at the category level of “face” as opposed to subordinate (individuated) categories. However, more recent work has suggested that the N170 can be responsive to individual facial identities, such as highly familiar faces (Caharel et al., 2006; Caharel et al., 2002) or after conditions of continuous habituation (Jacques & Rossion, 2006).
A more robust component of facial identity is the negative deflecting, posterior-occipital component elicited around 250 poststimulus onset. Initially, variation in this component was observed in repetition priming paradigms that presented target face stimuli successively, denoting the component as the N250r (repeated). Schweinberger et al. (2002) tested familiar faces (i.e., politicians and actors) and unfamiliar faces in a repetition paradigm that presented different pictures of the same face that varied randomly by perspective, age, expression, eye gaze, or hairstyle. Participants were shown a prime face and a target face and were told to categorize the latter as familiar or unfamiliar. The crucial trials utilized the same identity, but a different picture of the target face for the prime. Results revealed a significant priming effect for the same identity trials during the 200–300 msec postonset segments, resulting in larger, negative deflecting activation during this time epoch. The neural source of the N250r has been associated with activation in the anterior ventral temporal lobe corresponding to the fusiform gyrus (Schweinberger et al., 2002), an area that is responsive to individual faces (Gauthier et al., 2000).
One mechanism by which a face can be “learned” or familiarized is through the use of naming. The association of names and faces has a long-standing history in the literature, with many studies investigating the relation between faces and names during recognition (MacKenzie & Donaldson, 2009; Ellis, Young, Flude, & Burton, 1996; Burton & Bruce, 1992) and matching tasks (Huddy, Schweinberger, Jentzsch, & Burton, 2003; Carson, Burton, & Bruce, 2000) and in priming paradigms (Martin-Loeches, Sommer, & Hinojosa, 2005; Pickering & Schweinberger, 2003; Schweinberger et al., 2002; Ellis et al., 1996; Bruce & Valentine, 1985). The role of names has also been featured in prominent cognitive models of face processing (Ellis & Lewis, 2001; Carson et al., 2000; Valentine, Brennen, & Bredart, 1996; Burton & Bruce, 1992; Bruce & Young, 1986).
According to models of face recognition, two forms of semantic information are associated with a face. The first of these is perceptually derived and available from the face image, such as gender or age (Bruce & Young, 1986). The other kind of semantic information is abstract and associated with the person representation, such as occupations and names (Burton & Bruce, 1992; Cohen, 1990; Bruce & Young, 1986). Behavioral research focusing on face recognition has revealed surprisingly poor performance in face naming. Findings have shown that name retrieval is slower than judgments of familiarity (Young, Ellis, & Flude, 1988), mortality and nationality (Johnston & Bruce, 1990), and occupation (Young, McWeeny, Ellis, & Hay, 1986). Names are impaired in recall when compared with other semantic information such as occupation (Cohen & Faulkner, 1986) and more difficult to learn in association with a face than other semantic information such as occupations (McWeeny, Young, Hay, & Ellis, 1987; Cohen & Faulkner, 1986).
Whereas previous research has focused on abstract semantics and the nature of their association with face representations, few have attempted to investigate the process by which names and face representations become associated. In a series of ERP studies by Paller and colleagues, participants were required to learn a series of unnamed faces (exposure condition) and faces that were associated with names and abstract semantic information such as hobbies (semantic condition; Paller et al., 2003; Paller, Gonzalves, Grabowecky, Bozic, & Yamada, 2000). During the recognition phase, changes in ERPs were observed to learned faces as opposed to novel faces, with learned faces resulting in larger posterior midline activation than novel faces (Paller et al., 2000, 2003), and the faces accompanied by semantic information eliciting larger anterior midline activation than unnamed faces (Paller et al., 2000, 2003), providing support for memory retrieval of biographical information to the faces associated with semantic information. Other research has shown that episodic memory for names and faces activate different parietal regions, with faces eliciting activation in the right hemisphere and names activating the left hemisphere (MacKenzie & Donaldson, 2009).
A recent investigation of the neural correlates of face learning by Kaufmann, Schweinberger, and Burton (2009) attempted to associate face-specific ERPs to learning processes. Participants were required to study a series of video recordings displaying faces. Half of the faces were presented with no other semantic information associated (exposure condition) and the other half were presented with voice samples providing semantic information such as the person's name and other identity-specific information (e.g., occupation, place of residence). During a subsequent recognition test, differential ERP activation was found, revealing a significant difference between learned and novel faces at posterior occipital sites with more negative N170 and N250 amplitudes as a result of face repetition. Critically, no differences were found between the semantic and nonsemantic learned group, suggesting that the enhanced N250 negativity did not occur as a result of learning abstract semantic associations but as a result of perceptual learning.
It is important to note, however, that these experiments presented both name labels and other abstract semantic information together. In contrast, Tanaka, Curran, Porterfield, and Collins (2006) investigated the role of the name label uniquely in the acquisition of representations for previously unfamiliar faces. In their “Joe–No Joe” experiment, the participants' task was to monitor for an unfamiliar target face, “Joe,” which was presented among unfamiliar faces and unbeknown to the participant, their own face. Results showed that during the first half of the experiment, an N250 amplitude elicited to the participants' own face was significantly larger than the response to either the Joe face or a randomly assigned “Other” unfamiliar face. In the second half of the experiment, the acquired N250 amplitude produced to the Joe face increased to the same magnitude as the preexisting N250 to the participant's own face. It has been shown that highly overlearned faces develop visual representations, which facilitate perceptual fluency and recognition, or “robust representations” (Tong & Nakayama, 1999); therefore, the N250 amplitude elicited to the extremely familiar Own face and subsequently familiar “Joe” face is indicative of robust memory representations for both these faces (Tanaka et al., 2006).1
In addition to face familiarity effects, the N250 component has been shown to be sensitive to the acquisition of subordinate level representations. For example, a more negative N250 is produced when participants are trained to distinguish species of birds (Scott, Tanaka, Sheinberg, & Curran, 2006), models of cars (Scott, Tanaka, Sheinberg, & Curran, 2008), and other race faces (Tanaka & Pierce, 2009). These findings support the claim that the N250 indexes a finer, subordinate level representation that is characteristic of expert object recognition and everyday face recognition (Tanaka, 2001).
In contrast to the gradually accrued N250 component to Joe face, a prominent P300 component was in evidence at the very outset of the experiment, suggesting dissociation between the posterior N250 and P300 potentials (Tanaka et al., 2006). The P300 is a centroparietal midline component occurring at 300–600 msec postonset and has also become associated with processes of categorization. Whereas its elicitation in odd ball paradigms had led to theories suggesting that activation is related to the probability of a stimulus' presentation (Donchin, 1981; Duncan-Johnson & Donchin, 1979), other researchers have suggested that the P300 indexes changes in task requirements (Donchin & Coles, 1988; Donchin, 1981) and relevancy (Donchin & Coles, 1988).
The current “Joe–Bob” study investigated the role of naming in the formation and maintenance of representations for novel faces as measured by ERPs. Following the paradigm of Tanaka et al. (2006), electrophysiological activity was recorded while participants completed a Joe–No Joe task. In this task, participants were introduced to a previously unfamiliar target face (Joe) and then asked to monitor for the target “Joe” while shown a series of faces that included both target and other unfamiliar faces. To examine the effects of associating a name label and face representation, perceptually derived semantic information was kept constant across faces (i.e., all faces were the same age, race, and gender), and no other abstract semantic information, except the name, was associated with the face. Midway through the experiment, the target switched from the “Joe” face to a new target “Bob” face. We hypothesized that changes in the N250 component would be observed to the “Joe” and “Bob” faces as a result of practice, reflecting the acquisition of a face representation as a consequence of face–name associations. Critically, we expected that the N250 to presentations of the Joe face would be maintained even when the Bob face became the task-relevant face. We also expected increases in P300 activation to target stimuli such that a larger P300 amplitude would occur to “Joe” when the Joe face was the target face and to “Bob” when the Bob face was target face. Furthermore, no significant P300 was predicted for learned faces when those faces were not target, reflecting the effects of task relevancy.
Twenty-four first-year students (four men) from the University of Victoria, aged 17–25 years (M = 20.35 years) took part in this study. All but two individuals were right-handed, and all had normal or corrected-to-normal visual acuity. None had any history of brain injury or trauma. Informed consent was obtained from all participants before the experiment, and students were given the choice of either 2.5 credits to be put toward class requirements or $20 as compensation for their time.
The stimuli comprised images of 12 white male frontal faces with neutral expressions from the Karolinska Directed Emotional Faces database (Lundqvist, Flykt, & Öhman, 1998). Using Adobe Photoshop, the images were gray-scaled and cropped to a dimension of 122 × 184 pixels around the face area, allowing for a visual angle of 2.5° horizontally and 3.7° vertically. Assignment of faces to Joe, Bob, and Other face conditions was counterbalanced across participants. The remaining nine face stimuli were filler trials and not used in analysis but were equivalent to the Other condition from the participant's perspective.
After the EEG electrodes were applied, participants were introduced to the Joe target face and asked to inspect the Joe face until they felt confident that they could identify him. For the practice phase, participants viewed a series of faces on a computer screen, presented one at a time, and asked to identify the face as either “Joe” or “Other” via a key response. After the practice phase, if participants had no further questions, they proceeded to the experimental phase of the study. In the experimental phase, each trial consisted of a blank screen with a fixation cross at the center jittered for 500–1000 msec, followed by a blank screen lasting 250 msec, followed by a face stimulus for 500 msec, and ended with a prompt screen that read “Joe?” and two possible responses of “Joe” or “Other.” Participants were instructed to select the “Joe” key if the target appeared or the “Other” key if any other face than the target appeared (see Figure 1-1A). A delayed response procedure was employed to reduce motor-related artifacts in the ERP signal.
At the halfway point in the experiment, participants were shown one of the Other faces and were told that this face was “Bob.” For the second half of the experiment, participants were instructed to select the “Bob” key if they saw the target and the “Other” key if they saw any other face than Bob (including the Joe face). The prompt screen was altered to reflect the target change, with the prompt reading “Bob?” and the key selection marked “Bob” or “Other” (see Figure 1-1B). All other aspects of the trial were the same as in the first half of the experiment.
Participants completed 720 trials that were divided into six 120-trial blocks, with three blocks in the first half of the experiment and three blocks in the second half. The switch in target faces occurred at 360 trials (the halfway point), such that Joe was the target in the first half of the experiment and Bob was target in the latter half. Participants were given self-controlled breaks after each block, and an impedance check was performed at the halfway point before the introduction of Bob.
Scalp voltages were collected from 41 channels using the Brain Vision Recorder software (Version 1.3, Brainproducts, GmbH, Munich, Germany). Eye movements were recorded from electrodes placed on the left and right temples and beneath the right eye. All impedances were kept below 10 kΩ and sampled digitally at 250 Hz with a bandpass filter of 0.017–67.5 Hz on-line (Quick Amp, BrainProducts, GmbH, Munich, Germany). Data obtained were then subjected to several filtering and an ocular correction process off-line as follows; first, EEG was filtered using a bandpass filter of 0.1–20 Hz; thereafter, each ERP segment was divided into 800-msec epochs, beginning 200 msec before stimulus onset and subsiding 600 msec after stimulus onset. Thus, the start of each epoch coincided with the blank screen stimulus that preceded the presentation of the face stimulus. Potentials were referenced to an on-line average (Quick Amp, BrainProducts, GmbH, Munich, Germany), and then an off-line average reference was applied to minimize the effects of reference site activity and accurately estimate the scalp topography of the measured electrical fields (Dien, 1998). Ocular corrections followed, using the Gratton and Coles method (Gratton, Coles, & Donchin, 1983). Trials were rejected if voltage changes exceeded 35 μV or activation was less than 0.5 μV. All trials were baseline-corrected 200 msec before stimulus onset. There were 72 trials in the Joe, Bob, and Other conditions, respectively, with a minimum of 57 acceptable trials for each condition.
Accuracy was near ceiling with all conditions averaging 99% correct, and all incorrect trials were discarded from subsequent analysis. Because of the delayed response paradigm, RT was not analyzed. No decrease in accuracy was observed as a result of switching the target from Joe in the first half of the experiment to Bob in the second half. Three participants were excluded from analysis for failing to meet the minimum number of acceptable trials per condition, and a fourth participant was excluded because of a technical malfunction.
Analyses focused on channels P7 and PO7 in the left hemisphere and P8 and PO8 in the right hemisphere where the N170 and N250 amplitudes were maximal. A 2 × 2 × 3 × 2 ANOVA was carried out with Hemisphere (left, right), Location (lateral, medial), Condition (Joe, Bob, Other), and Half (first, second) as within-participants factors. Greenhouse–Geisser and Bonferonni adjustments were performed when appropriate. All reported post hoc comparisons were reliable at the p = .05 level.
The latency of the N170 peak amplitude was determined within a time window between 140 and 240 msec poststimulus onset. Analysis of the peak latency of the N170 component showed a significant effect of Hemisphere, F(1, 20) = 10.50, p < .01 (left hemisphere: M = 169 msec, SE = 2.13; right hemisphere: M = 163, SE = 1.74) and Half, F(1, 20) = 16.53, p < .01 (first half: M = 164 msec, SE = 1.64; second half: M = 168 msec, SE = 1.94). Although the Half × Condition interaction was reliable, F(2, 40) = 3.46, p < .05, the differences between halves for each condition did not exceed 5 msec and, thus, were not meaningful. No other main effects or higher-order interactions were reliable, p > .05.
Analysis of the N170 peak amplitudes showed a reliable effect of Location, F(1, 20) = 32.26, p < .01, where medial locations (PO7–PO8) were larger (M = 4.14, SE = 0.64) than lateral locations (P7–P8; M = 2.29, SE = 0.49). No other main effects or higher-order interactions were reliable, p > .05.
Following previous studies (Tanaka et al., 2006), a window of 230–330 msec was used to compute the mean amplitude of the N250. Whereas the N250 observed in this study is positive, we will use the term “negativity” to reflect a relative negativity as a result of decrease in N250 amplitude, as opposed to an absolute negativity. The main effect of Hemisphere, F(1, 20) = 4.63, p < .05, was reliable, such that the right hemisphere activation (M = 3.86, SE = 0.45) was larger than the left hemisphere activation (M = 2.79, SE = 0.56). Location was also reliable, F(1, 20) = 38.74, p < .01, with activity at the medial locations greater (M = 3.96, SE = 0.48) than the activity at the lateral locations (M = 2.69, SE = 0.44). The main effect of Location reliably interacted with Hemisphere, F(1, 20) = 5.19, and Half, F(1, 20) = 7.06, and a three-way interaction between Location, Hemisphere, and Half, F(1, 20) = 5.09, p < .05, was also found. The main effects of Condition, F(2, 40) = 31.60, and Half, F(1, 20) = 37.44, were significant, p < .01, and critically, the interaction between Condition and Half, F(2, 40) = 22.09, p < .01, was also reliable. No other interactions reached reliable levels, p > .05. Wave plots can be seen in Figure 2.
Further exploration of the Condition × Half interaction focused on electrode PO8, where activation was maximal. Findings showed that, in the Joe and Other conditions, there was an overall increase in the N250 negativity from the first half of the experiment (Joe: M = 2.40, SE = 0.30; Other: M = 4.74, SE = 0.30) to the second half (Joe: M = 1.98, SE = 0.30; Other: M = 3.92, SE = 0.33). In the Bob condition, an increase in the N250 negativity was also observed between the first (M = 4.82, SE = 0.31) and second halves (M = 4.82, SE = 0.31) of the experiment (see bar graph, Figure 3). However, Bob showed the largest shift in the N250 response. Although Bob was not reliably different from Other in the first half of the experiment, Bob was reliably different from Other in the second half, p < .01. In contrast, the N250 to Joe was reliably greater than the Other face in both the first and second halves of the experiment. The comparisons of the N250 to the Joe and Bob faces relative to the Other face are shown in the scalp topographies in Figure 3.
The mean amplitudes for channels across the midline (Fpz, Fz, FCz, Cz, CPz, Pz, POz, Oz) revealed the largest mean amplitude at the Pz electrode, and thus, analysis focused on this channel. A window of 400–600 msec was used to compute mean amplitude of the P300. A 2 × 2 ANOVA was performed with the within-participants factors of Condition (Joe, Bob, Other) and Half (first, second). Mean amplitude analysis showed a significant main effect of Condition, F(2, 40) = 26.68, p < .01. No significant effect of Half was found, F(1, 20) = 3.54, p > .05. A significant interaction between Half × Condition was observed, F(2, 40) = 17.26, p < .01. Direct comparisons showed that, for the Joe face, the P300 amplitude was larger in the first half of the experiment (M = 4.05, SE = 0.34) than the second half (M = 3.06, SE = 0.36). For the Bob condition, the P300 amplitude was smaller in the first half (M = 1.01, SE = 0.26) than the second half (M = 2.86, SE = 0.45; as shown in Figure 4).
In Experiment 1, it was found that a robust N250 response was elicited by the target Joe face in the first half of the experiment and by the target Bob face in the second half of the experiment. Interestingly, the increase in N250 negativity to the Joe face was maintained in the second half of the experiment when the Joe face was no longer a target. These findings indicate that the N250 is not task-dependent but may reflect a more enduring face representation of the Joe face. A more negative N250 response was also observed in the Other condition in the second half of the experiment relative to the first half, but the N250 negativity to Other was reliably smaller than both the Bob condition and the Joe condition. P300 amplitude was significantly larger in response to the target Joe face than the Bob or Other faces during the first half of the experiment. In the second half of the experiment, P300 amplitude increased in response to the target Bob face, although a somewhat reduced yet sustained P300 was found in response to the previous target Joe face.
One drawback of Experiment 1 was that the presentation of the labels was confounded with task relevancy, such that the name (i.e., Joe or Bob) was only assigned to a face when it became a target. Hence, it is not known whether the emergence of N250 and P300 components in the second half of the experiment was because of labeling effects, task relevancy, or a combination of these factors. To disentangle labeling and task relevancy, a second experiment was performed where the names for the Joe and Bob faces were presented at the beginning of the experiment. If the ERP effects observed in Experiment 1 were because of labeling, we would expect a more negative N250 and a more positive P300 to the Bob face during the initial “Joe–No Joe” phase of the experiment. In contrast, if the name label and task relevancy are critical in the formation of the N250 and P300 components, we would expect an N250 and P300 to the Bob face to emerge only in the second half “Bob–No Bob” phase of the experiment, when this face becomes task relevant.
Twenty-two first-year students (5 men), aged 18–24 years (M = 19.25), took part in this study. None of the participants took part in Experiment 1. Informed consent was obtained before the experiment from all participants who participated in return for course credit. All participants had normal or corrected-to-normal vision and were right-handed. None had any history of brain injury or trauma.
All procedures were the same as in Experiment 1, except for the presentation of the target stimuli and instructions. In the current study, participants were shown pictures of Joe and Bob at the beginning of the experiment and told to study both faces as they would have to identify both targets. Participants moved on to the practice phase after indicating that they were confident with the targets' identities. Participants were told to monitor only for Joe in the first half of the experiment, and they were to monitor for Bob in the second half of the experiment. All other instructions, materials, and EEG methods were identical to the first experiment.
Accuracy was near ceiling with all conditions averaging 99% correct. Incorrect trials were excluded from any subsequent analysis. No decrease in accuracy was observed as a result of switching the target from Joe in the first half of the experiment to Bob in the second half. RT was not analyzed because of the delayed response paradigm employed in this study.
Similar to Experiment 1, analyses focused on channels P7 and PO7 in the left hemisphere and channels P8 and PO8 in the right hemisphere where the N170 and N250 amplitudes were maximal. A 2 × 2 × 3 × 2 ANOVA was carried out with Hemisphere (left, right), Location (lateral, medial), Condition (Joe, Bob, Other), and Half (first, second) as within-participants factors. Greenhouse–Geiser and Bonferonni adjustments were performed when appropriate. All reported post hoc comparisons were reliable at the p = .05 level.
The latency of the N170 peak amplitude was determined within a time window of 140–240 msec poststimulus onset. For peak latency, Hemisphere, F(1, 21) = 6.54, p < .05 (left hemisphere: M = 168 msec, SE = 2.33; right hemisphere: M = 164, SE = 2.28) and Half, F(1, 21) = 12.65, p < .01 (first half: M = 164 msec, SE = 2.23; second half: M = 167 msec, SE = 2.22) were reliable.
Analysis of the N170 peak amplitudes showed an effect of Location, F(1, 21) = 39.24, p < .01, with medial locations (PO7–PO8) larger (M = 3.69, SE = 0.69) than lateral locations (P7–P8; M = 1.68, SE = 0.54). The Location × Hemisphere × Half interaction, F(1, 21) = 6.51, p < .05, was also reliable. No other main effects or higher-order interactions were found, p > .05.
Similar to Experiment 1, a window of 230–330 msec was specified to calculate the mean amplitude of the N250. As aforementioned, we use the term “negativity” to reflect a relative negativity as a result of a decrease in N250 amplitude, not an absolute negativity. The main effect of Location was reliable, F(1, 21) = 21.20, p < .01, with amplitude at the medial locations greater (M = 3.95, SE = 0.48) than lateral locations (M = 2.79, SE = 0.36). The Location × Hemisphere, F(1, 21) = 5.23, and Location × Half, F(1, 21) = 4.86, interactions were reliable, p < .05. The main effects of Condition, F(2, 42) = 22.00, and Half, F(1, 21) = 55.12, as well as the Condition × Half interaction, F(2, 42) = 7.70, were also reliable p < .01. No other interactions reached reliable levels, p > .05. See Figure 5 for wave plots.
To investigate the source of the Condition × Half interaction, a series of post hoc comparisons were performed on electrode PO8, where activation was maximal. Similar to Experiment 1, for Joe and Other faces, there was an overall increase in the N250 negativity from the first half of the experiment (Joe: M = 2.69, SE = 0.30; Other: M = 4.73, SE = 0.29) to the second half (Joe: M = 1.95, SE = 0.28; Other: M = 3.90, SE = 0.26). In the Bob condition, a large increase in the N250 negativity was also observed between the first (M = 4.53, SE = 0.27) and second halves (M = 2.40, SE = 0.28) of the experiment (see bar graph, Figure 6). As a nontarget, Bob was not reliably different from Other in the first half of the experiment, p > .05, but the change in N250 negativity to Bob was reliably greater than that to the Other condition when the Bob face became a target in the second half of the experiment, p < .01. In both the first and second halves of the experiment, the N250 negativity to Joe was reliably larger than the response to the Other face. The differences between the Joe and Bob conditions relative to the Other condition are reflected in the scalp topographies shown in Figure 6.
The P300 analysis focused on the Pz electrode where maximal activity of the midline channels was observed, during a time window of 400–600 msec. The mean amplitude analysis revealed a reliable main effect of Condition, F(2, 42) = 72.63, p < .01, such that the Joe condition (M = 4.08, SE = 0.28) was significantly larger than the Bob condition (M = 2.65, SE = 0.26) and the Bob condition was reliably larger than the Other condition (M = 1.38, SE = 0.16). A reliable effect of Half was found, F(1, 21) = 4.67, p < .05, with P300 amplitude smaller in the first half of the experiment (M = 2.52, SE = 0.19) than the second half (M = 2.89, SE = 0.24). A significant Half × Condition interaction was also found, F(2, 42) = 30.29, p < .01.
In the Joe condition, a larger P300 was found in the first half of the experiment (M = 4.36, SE = 0.27) than the second half (M = 3.80, SE = 0.32; see Figure 7). The opposite pattern was found in the Bob condition, in which a greater P300 was produced in the second half of the experiment (M = 3.65, SE = 0.37) than the first half (M = 1.64, SE = 0.26). There was no change in P300 amplitude in the Other condition between the first and second halves of the experiment.
In summary, both Joe and Bob faces were given name labels at the start of Experiment 2. In the first half of the experiment, only the target Joe face showed a significantly more negative N250 response. Only in the second half of the experiment when the Bob face became a target did the Bob face elicit a N250 that was more negative than the Other face. These findings indicate that name labels are not sufficient for eliciting a robust N250 response but that the N250 is acquired through practice and experience. P300 amplitude was significantly larger for the Joe condition during the first half of the experiment, decreasing reliably during the second half. An increase in both P300 latency and amplitude was found for the Bob condition in the second half of the experiment, resulting in amplitudes of equal magnitude for both Joe and Bob faces.
The current study was intended to test the effects of name labels, task relevancy, and exposure on the formation of face representations. In both Experiments 1 and 2, participants were required to monitor for a target face, “Joe,” in the first half of the experiment and a target face “Bob” in the second half of the experiment. In Experiment 1, the target Bob face was introduced halfway through the experiment, whereas in Experiment 2, the Bob face was previewed at the beginning of the experiment. An increased N250 negativity was found in response to the Joe face during the first half of the experiment when Joe was the target and an enhanced N250 negativity to the Bob face when Bob was the target. Critically, in both experiments, the enhanced N250 negativity to the Joe face was maintained during the Bob phase of the experiment. A more negative N250 was also observed to the Other face in the second half of the experiment. Although not as large as the response to the past target Joe and current target Bob faces, it is likely that repeated exposure to the Other face enhanced a N250 response.
These studies suggest conditions under which a more negative N250 is maximized. First, our results show that simple stimulus exposure is not sufficient to produce a large increase in the N250. In our experiments, repeated presentations of “Other” faces produced a modest increase in N250 negativity, although this enhancement did not reach the same proportions as the target Joe and Bob faces. Nor is the N250 a simple labeling effect in which a face is made distinctive by assignment of a proper name. Specifically, in Experiment 2, Bob was introduced at the beginning of the experiment and was distinguished from the “Other” faces by its name. However, name knowledge was not sufficient to generate a more negative N250 response as indicated by the absence of the N250 to Bob in the Joe–No Joe phase of the study. Only in the second half of the experiment, when the Bob face was task relevant, was a reliable increase in N250 negativity produced. Thus, individuating a face with a name label is not sufficient to elicit the N250.
Although task relevance appears to be critical in the formation of the N250 response, it is not necessary for maintaining the N250 component. In the second half of Experiments 1 and 2, when Bob was the target face, the Joe face continued to produce an enhanced N250 negativity, despite no longer being task relevant. These results suggest that the N250 to highly familiar faces is obligatory and independent of task demands. The automatic activation to the nontarget Joe face in the Bob–No Bob phase of the experiments is similar to the more negative N250 shown to other highly familiar faces, such as one's own face (Tanaka et al., 2006), and the N250r to task-irrelevant famous faces (Neumann & Schweinberger, 2008).
Tong and Nakayama (1999) suggested that highly overlearned faces, such as one's own face, result in robust representations that are characterized by rapid recognition and reduced attentional resources. These representations are a result of extensive visual experience with a face in which the dynamic properties, such as expression, viewpoint, and even lighting contribute to optimal learning conditions (Tong & Nakayama, 1999). These representations are robust in that they contain abstract and view-invariant information that facilitate visual and decisional processes that generalize across tasks and contexts (Tong & Nakayama, 1999). The current study has highlighted qualitative differences between mere exposure, name labels, and deliberate practice in the formation of subordinate-level face representations. As indexed by the N250, robust memory representations were formed by repeated practice associating a name label to a single face. Thus, it is possible that the task demands of this experiment mirrored the extensive visual experience necessary for generating robust representations for highly, personally familiar faces (e.g., one's own face).
It is important to note, however, that the current study made use of static images that were repeated many times during the experiment. Thus, it is possible that the observed N250 is the result of picture learning rather the formation of a robust representation of a familiar person. However, a study by Bindemann, Burton, Leuthold, and Schweinberger (2008) demonstrated that, in a priming paradigm, repeated faces of famous people that were vertically or horizontally distorted elicited the same N250r priming effects as the priming effects found for identical images. In a recent study, Kaufmann, Schweinberger, and Burton (2009) showed dynamic video clips of faces from varying viewpoints. Although the accuracy in identifying a newly familiarized face was far below ceiling, the N250 amplitude to the multiple views of the learned face was larger than the response to novel faces and the magnitude of the N250 grew with increasing repetition, suggesting that a more stable face representation was formed over time. These results support the idea that the N250 is an indicator of identity or memory representations for faces. In future research, it will be important to test the robust representations of the N250 response to target faces in the Joe–No Joe paradigm by varying their viewpoints in a procedure similar to the Kaufmann et al. study (2009).
The N250 findings are consistent with previous training experiments in which a more negative N250 was elicited to faces from a specific racial group that were individuated with a name label (Tanaka & Pierce, 2009). In this study, participants who were required to individuate black or Hispanic faces at the subordinate level showed an enhanced N250 negativity to novel exemplars from the subordinate racial group (Tanaka & Pierce, 2009). Importantly, participants demonstrated better recognition performance to faces from the subordinate level race, suggesting that the N250 might index a more durable memory representation that facilitates recognition. Further evidence indicating a connection between this ERP component and recognition comes from object training studies where the increase in N250 negativity and recognition advantage are still present 1 week following training (Scott et al., 2006).
One difference between the current experiments and the original Tanaka et al. (2006) study is the immediate N250 response to the Joe face as opposed to the accrued N250 findings of the original study. In the Tanaka et al. (2006) experiments, participants may have been anticipating the presentation of their own face, which, in turn, hindered their learning of the Joe face. In Experiment 1, there were no such distractions in learning the task-relevant Joe face, perhaps resulting in an earlier onset of N250 component. It is worth noting that, in Experiment 2, when both Joe and Bob were introduced at the beginning of the experiment, the N250 to Joe—although reliably different from Bob and Other in the first half of the experiment—continued to increase in negativity in the second half of the experiment. Presentation of the Bob face might have acted as a distraction in the learning process similar to the distraction of one's own face in the original Tanaka et al. (2006) study.
These experiments also shed light on the role of the P300 in face–name learning. The P300 amplitude was consistently larger to task-relevant target faces than nontarget faces, a finding that is congruent with those of the previous Joe–No Joe experiment (Tanaka et al., 2006). In the second half of the experiment, when the target face switched to Bob, the increased P300 to the target Bob face was offset by small but reliable decrease in P300 amplitude to the Joe face. The net result was that the Joe and Bob faces produced a P300 of equivalent magnitude in the second half of the experiment, although the Joe face was no longer task relevant. Despite requiring two distinct responses in the second half of the experiment (i.e., “No Bob” to Joe face and “Bob” to the Bob face), participants may have grouped the Joe face and Bob face together as a special category that is distinct from the “Other” nontarget faces. This interpretation is consistent with other work showing exemplars that belong to the same psychological category will elicit a similar P300 response (Johnson & Donchin, 1980; Kutas & Donchin, 1979). Because the magnitude of the P300 is inversely proportional to the probability of the stimulus (Mecklinger & Ullsperger, 1993;Johnson & Donchin, 1980;Kutas & Donchin, 1979), the reduced amplitudes to the Bob and Joe faces can be explained by the increase in probability from a 10% (e.g., Joe face) to a 20% (i.e., Joe or Bob face) chance of occurrence in the second half of the experiment.
So what is in a name? The current study had found reliable changes in the N250 as a result of name–face associations. One factor that may have influenced the importance of the name in the learning process was the way in which names were used to familiarize a face; whereas the previous experiments required participants to identify named faces as familiar or unfamiliar (Kaufmann et al., 2009), the current experiment required the participants to use the name label to categorize a face as a target. As a result, the name-label in the current study may have anchored the acquisition of a face representation by facilitating the individuation process, comparable with the N250 findings in training expert object recognition (Scott et al., 2006, 2008) and other-race face recognition (Tanaka & Pierce, 2009).
In summary, the reported findings reveal that the process of acquiring robust face representations is not a passive but an active process in which the percept and label are associated through repeated practice. Essentially, the N250 component showed no changes to the simple act of labeling and minor changes in negativity as a result of mere exposure, but when combined with task relevancy produce alterations that persisted even beyond task relevancy. These findings reinforce the notion that the N250 component is a result of a persistent change in brain function because of an active process of learning that combines all three factors: name labels, exposure, and task relevancy. The acquisition of face representations are also corroborated by reliable changes in the P300 component that were evident even when faces were no longer task-relevant. Future research may investigate the permanence of these changes using paradigms that test for face familiarity across longer retention intervals (days, weeks) to fully understand the robustness of face representations and their manifestation in ERPs. Other research may also look to compare the acquisition of face representations using stimuli from both static and varying viewpoints and its effects on long-term retention.
Reprint requests should be sent to Iris Gordon, Department of Psychology, University of Victoria, PO Box 3050 STN CSC, Victoria, British Columbia, Canada V8W 3P5, or via e-mail: firstname.lastname@example.org.
In this experiment, the participant's own face failed to influence the earlier N170 component in contrast to other studies showing an own face N170 effect (Caharel et al., 2002).