Self priming from distinctive and caricatured faces.
The Burton et al. IAC model comprises four pools of units: face recognition units (FRUs), name input units (NIUs), person identity nodes (PINs) and semantic information units (SIUs). Each unit is connected to all other units in the same pool by bidirectional inhibitory connections. Additionally, equivalent units in different pools are linked by bidirectional excitatory connections.
There is one FRU to represent the pattern formed by each of the faces with which we are familiar. Similarly, the pattern formed by a familiar person's name is represented by an NIU. Hence, the presentation of Stan Laurel's face or name will activate different input units; however, these input units are connected to a single PIN that represents Stan Laurel's identity. The PINs act as a gateway to the access of semantic information which is stored separately as SIUs, which are accessible only via the PINs. Burton et al. propose that the recognition of familiarity takes place at one common level, the PINs, regardless of the input domain (face, name, voice, etc.). This function of the PINs is central to Burton et al.'s account of priming effects.
Studies of priming effects in person recognition have identified two main types of effect (Young & Bruce, 1991). First, there is a long-lasting benefit of repetition, which is domain specific (Bruce & Valentine, 1985; Ellis, Young, Flude & Hay, 1987); seeing a face primes subsequent recognition of that face some considerable time later, but seeing a person's name has no benefit on the recognition of their face after an appreciable interval. Similarly, seeing a person's name will prime the recognition of the same name presented some time later but not the same person's face. Burton et al.'s (1990) IAC model accounts for these long-lived but domain-specific effects of repetition in terms of the strengthening of input-PIN connections (FRU-PIN for face input, NIU-PIN for name input).
The second type of widely studied priming effect is a short-lived benefit on the recognition of an associated person, usually referred to as semantic priming (Bruce & Valentine, 1986; Young, Flude, Hellawell & Ellis, 1994). For those old enough to remember, seeing the face of John Lennon primes recognition of Paul McCartney's face or recognition of Paul McCartney's name. However, although this priming (unlike repetition priming) can cross input domains, it is short lived.
Burton et al. suggest that a face primes recognition of a semantically associated face or name because they are associated (via the PINs and SIUs) with the same semantic information. In the case of Lennon and McCartney, the presentation of John Lennon's face activates his FRU and PIN. This in turn leads to the activation of SIUs connected to Lennon's PIN (e.g. songwriters, Sergeant Pepper, etc.). Because McCartney shares many of the same SIUs with his partner, activation feeds back to McCartney's PIN. Hence, when McCartney's name (or face) is presented after Lennon's face (or name), the recognition of McCartney's name (or face) is facilitated. Furthermore, semantic priming should be short lived because an input to a third semantically unrelated PIN (e.g. Margaret Thatcher) causes Lennon and McCartney's PINs to return to their resting activation.
In addition to these accounts of established priming effects the IAC model predicts a new form of priming. Burton et al. refer to this new form of priming as 'self-priming' and it depends on the same underlying mechanisms as semantic priming. In the above example of Burton et al.'s IAC account of semantic priming, Lennon was shown to prime McCartney because they are associated with the same semantic information. However, Burton et al. also note that Lennon's PIN remains active after McCartney's PIN has become active. Therefore, the model predicts that the presentation of Lennon's face should prime the recognition of his own name, provided that no subsequent items have abolished the activation at Lennon's PIN. Hence, there should be a short-lived form of cross-domain priming from a person's face to recognition of his or her name - self-priming.
The existence of this cross-domain (face prime-name target) self-priming effect has been confirmed in neuropsychological work (de Haan, Bauer & Greve, 1992a; de Haan, Young & Newcombe, 1992b) and studies of normal subjects (Young et al., 1994). However, little as yet is known about self priming, and one aspect that has not been explored is its relation to the distinctiveness of the face prime.
Valentine & Bruce (1986a, b) found that familiar faces that are distinctive in appearance are recognized faster than familiar faces that are typical in appearance. Burton et al. have included a 'front end' component in their model that offers an account of Valentine & Bruce's finding. Figure 2 shows Burton et al.'s model of distinctiveness effects with faces.
In front of the FRUs Burton et al. have added a number of units to code visual characteristics of the seen face [ILLUSTRATION FOR FIGURE 2 OMITTED]; we will call these VCUs. Each FRU is connected to six VCUs (only three sets of VCUs are shown in [ILLUSTRATION FOR FIGURE 2 OMITTED]), which, for the sake of simplicity, are given a photofit-like status. Hence, there are VCUs for eyes, nose, hair, etc. VCUs can be shared by a number of faces or they can be connected to only one FRU. A measure of distinctiveness for each FRU is calculated by counting the number of other FRUs that share the same VCUs. Consequently, the FRU for a distinctive face shares very few, or none of its VCUs with other FRUs while the FRU for a typical face shares its VCUs with a large number of other FRUs. When the VCUs corresponding to a typical face are activated, a number of other FRUs besides the target FRU become activated. This leads to large levels of inhibition within the FRU pool and the target's FRU and PIN have a low activation level. In contrast, simulations with the model show that the FRUs corresponding to distinctive faces produce higher and more rapid activation of their PINs.
Recall that in Burton et al.'s account of cross-domain self priming, the face prime and name target activate the same PIN. Recall also that the decision that a name (or face) is familiar is made at the PIN level. Hence, because distinctive face primes produce higher PIN activation than typical face primes it follows that target names preceded by their corresponding face primes should be categorized as familiar faster if the face prime is distinctive than if the face prime is more typical in appearance. In other words the Burton model predicts that distinctive faces should produce more cross-domain self priming than typical faces. Figure 3 shows a simulation of this effect.
The solid black squares illustrate the activation in a PIN connected to a distinctive face. The first section of the graph (0 to 80 cycles) shows the PIN activation following an input to each VCU connected to a distinctive face (corresponding to the presentation of a familiar distinctive face). After 80 cycles these inputs are turned off and the system is left to run without any external inputs for 30 cycles (corresponding to a blank interval). Finally, at the 110 cycle point there is an input to the NIU for that same distinctive face (corresponding to the presentation of the name matching the distinctive face). The hollow squares illustrate the same course of events for a typical face.
Note that for the distinctive face condition the PIN activation reaches recognition threshold in fewer cycles following the initial input to its VCUs and again later following the input to its corresponding NIU. In this way, the Burton et al. model accounts for the finding that distinctive faces are recognized faster than typical faces and predicts that a distinctive face will prime recognition of its name more than a typical face.
In Expt 1, therefore, we tested this prediction of enhanced self priming from distinctive faces with typical and distinctive face sets.
Distinctiveness and familiarity ratings
Subjects. Twenty subjects from the undergraduate and postgraduate populations of the University of Durham acted as raters. They were between the ages of 18 and 31 years and had normal or corrected to normal vision. All were paid for participating.
Materials. Black and white photographs of 82 faces of famous celebrities were prepared. The prints were approximately 120 x 180 mm.
Procedure. The 82 faces were given to two groups of 10 students to rate. One group rated the faces for distinctiveness on a scale from 1 to 7, where 1 indicated 'highly typical' and 7 'highly distinctive'. As a guideline, distinctiveness was defined as something that distinguishes a face from the general population - a large nose, particularly close-set eyes, a thin face, etc. Subjects were told that they should base their ratings purely on facial features and not on the persons' personalities. The second group rated the faces for familiarity on a scale from 1 to 7, where 1 was defined as 'never seen before' and 7 as 'highly familiar'.
The distinctiveness ratings assigned to the 82 faces were analysed using Kendall's Coefficient of Concordance. The result showed that there was a significant amount of agreement between subjects' distinctiveness ratings (W = 0.48, [[Chi].sup.2](81) = 386.04, p [less than] .001).
Two sets of 12 male faces were selected from the rated faces (see Appendix 1). One group contained faces that had been given a high distinctiveness rating (mean = 5.5, SD = 0.57) and the other group faces that had been given a low distinctiveness rating (mean = 2.3, SD = 0.34). A t test confirmed that the distinctive and typical face sets were significantly different in terms of their distinctiveness ratings (t(22) = 16.374, p [less than] .001). Care was taken to ensure that the mean familiarity ratings of the typical and distinctive face sets were as similar as possible (mean = 6.0, SD = 0.76 for high distinctiveness set, and mean = 6.1, SD = 0.72 for low distinctiveness set).
Self-priming with distinctive and typical faces
Subjects. Twelve subjects from the undergraduate and postgraduate populations of the University of Durham participated as subjects. Subjects were between the ages of 18 and 32 and had normal or corrected to normal vision. Subjects were paid for participating.
Apparatus. A three-field projection tachistoscope and three Kodak SAV 2050 projectors were used to present the stimuli.
Design and procedure. Two experimental factors were examined, distinctiveness of face prime (which had two levels: distinctive and typical) and prime type (which had three levels: same, neutral and unrelated).
The 12 distinctive and 12 typical faces selected as stimuli were used as prime sets. Each of the faces (distinctive and typical) was transferred to a black and white slide transparency so that, as far as possible, the face filled the 36 x 24 mm frame. Black and white transparencies were also prepared of the names of these people, printed in uppercase Helvetica font (e.g. TERRY WOGAN). These slides were back-projected onto a white screen and subtended a horizontal visual angle of approximately 8 [degrees].
On each trial of the experiment, following a verbal warning signal from the experimenter, a face (the prime) was presented for 250 ms followed by an inter-stimulus interval of 250 ms and then a name (the target) for 2.5 s.
Each of the familiar target names in the typical and distinctive sets was seen three times, once in each of the same, neutral and unrelated prime type conditions. Examples of the three levels of the prime type condition were as follows:
Same. The prime face and target name were of the same person; e.g. Terry Wogan's face was followed by the target name TERRY WOGAN.
Neutral. The target name was preceded by an unfamiliar face prime; e.g. an unfamiliar face was followed by the target name TERRY WOGAN. A single unfamiliar face was used as the neutral prime throughout the experiment.
Unrelated. The prime face and target name were of different famous individuals who were not semantically related; e.g. Roger Moore's face was followed by the target name, TERRY WOGAN. Unrelated prime-target pairs were produced by mixing the faces and names within the distinctive and typical sets.
Subjects were instructed to look at the prime but respond only to the target name by making a manual button-press response to indicate whether the name was familiar (i.e. a famous person's name) or unfamiliar. Reaction times were recorded on an electronic timer, activated from the onset of the target name and terminated by a manual response made by the subject pressing one of two horizontally located buttons. The buttons were marked Yes and No; subjects were instructed to press the Yes button if they thought the target name was familiar and the No button if they thought that it was unfamiliar. Half of the subjects made positive familiarity responses with their right hands and half with their left. Each of the three conditions (same, neutral and unrelated) contained 24 prime-target pairs; half were made up of the faces and names from the distinctive face set and the other half from the typical face set. Hence, there were 72 prime-target pairs to which the subjects were required to make a positive familiarity response to the target name. A further 72 prime-target pairs were added as 'No' response trials. These were created by replacing the 24 familiar target names with 24 unfamiliar invented names matched to the familiar target names in terms of number of letters (e.g. Terry Wogan [right arrow] Jimmy Green). Hence, there were 144 prime-target pairs in all. Each trial was separated from the next by approximately 3 s. Subjects saw all 144 prime-target pairs. Presentation of the stimulus pairs was pseudo-random with respect to distinctiveness of prime, prime-target condition and familiarity of the target names.
Before starting the experiment, the subjects were presented with slides of all 24 familiar and 24 unfamiliar target names written in Helvetica script (i.e. exactly as they were to appear in the main task itself). The names were pseudo-randomly presented with respect to familiarity and the subjects were required to make a familiarity decision to each name. The names were presented twice to ensure that the subjects were familiar with the target stimuli and practised in pressing the response keys. Following the target name presentation, a short practice trial was run containing 10 of the stimulus pairs described above. The practice trials included examples from each of the experimental conditions. Immediately after the practice trials the main experimental trials were run.
Mean reaction times for the familiarity decisions and mean error rates are shown in Table 1. Error rates were low ([less than] 4 per cent overall for familiar targets, [less than] 2 per cent for unfamiliar targets) and will not be considered further. They are presented in Table 1 only to show that the results do not represent a speed-accuracy trade-off.
Analysis by subjects
A two-factor ANOVA was carried out on the correct reaction times for familiar target names. The within-subject factors were distinctiveness (typical or distinctive face primes; repeated measure) and prime type (same, neutral and unrelated; repeated measure). There was a significant effect of distinctiveness (F(1, 11) = 8.07, p [less than] .02) indicating that overall, familiar target names preceded by distinctive primes were responded to faster than those preceded by typical primes. There was also a significant effect of prime type (F(2, 22) = 43.56, p [less than] .001). Newman-Keuls tests ([Alpha] = 0.05) showed that responses to target names preceded by same primes were significantly faster than responses to target names preceded by neutral and unrelated primes, which did not differ. Therefore, the overall priming effect was facilitation from the same primes without inhibition from the unrelated primes. This result is consistent with Posner & Snyder's (1975) criterion for automatic priming. Further, and of most interest, there was a significant interaction between distinctiveness and prime type (F(2, 22)= 4.75, p [less than] .05). Simple effects analyses showed that the significant interaction effect was attributable to a significant effect of distinctiveness for the same prime type condition only (F(1, 11) = 25.796, p [less than] .001); that is, distinctive face primes produced more self priming than typical face primes. Hence, the main effect of distinctiveness was qualified by the significant interaction effect; that is, the effect of distinctiveness was restricted to the same prime type condition.
Table 1. Mean correct reaction times (in ms) and mean error rates for correct familiarity decisions to familiar and unfamiliar target names preceded by distinctive and typical face primes in the three prime type conditions: same, neutral and unrelated. Maximum possible error rates (number of trials) for each of the distinctive and typical prime type conditions are shown at the bottom of the table
Familiar targets Unfamiliar targets Same Neutral Unrelated RTs Distinctive primes 492 598 612 676 Typical primes 540 608 616 652 Errors Distinctive primes 0.3 0.4 0.3 0.5 Typical primes 0.8 0.6 0.4 0.5 Max. possible errors 12 12 12 36
Analysis by items
The reaction time data were also submitted to a two-factor ANOVA by items. As with the subjects analysis the within-subject factors were distinctiveness (typical or distinctive face primes; repeated measure) and prime type (same, neutral and unrelated; repeated measure). The results showed no main effect of distinctiveness. There was a significant effect of prime type (F(2,22) = 60.394, p [less than] .001). Newman-Keuls tests ([Alpha] = 0.05) showed the same pattern of effects of prime type found for the analysis by subjects. There was also a significant interaction between distinctiveness and prime type (F(2, 22) = 3.555, p [less than] .05). Simple effects analysis showed that this was attributable to the effect of distinctiveness for the same prime type only (F(1, 11) = 7.576, p [less than] .05) that is, distinctive primes produced more self priming than typical primes. Thus, the analysis by items replicates the principal finding of the analysis by subjects.
Simulations with the IAC model demonstrated that for this model familiar distinctive faces are recognized faster than typical faces because the former produce lower levels of inhibition within the FRU pool. Consequently, distinctive faces produce more rapid and higher activation of their corresponding PINs, leading to the prediction of an enhanced cross-domain self-priming effect from a distinctive face prime to recognition of the same person's name. The significant interaction between the distinctiveness of the face prime (distinctive and typical) and prime type (same, neutral and unrelated) found in both the subjects and items analyses is therefore consistent with these simulations. Furthermore, the simple effects analyses (subjects and items) indicated an effect of distinctiveness for the same prime type only. This confirms that the source of the interaction effect was the greater degree of facilitation from distinctive primes in the same condition.
Consistent with Burton et al.'s IAC model, Expt 1 demonstrated that cross-domain (face prime-name target) self-priming interacts with the distinctiveness of the face prime. That is, distinctive face primes produce more self-priming than typical face primes. This forms a further example of how the IAC model offers useful accounts of priming and distinctiveness. Nevertheless, there are some face recognition effects for which the IAC model has not as yet been extended to accommodate.
One such effect is caricature. Perkins (1975) has defined caricatures as representations in which a face's characteristic features (i.e. those features of the face that distinguish it from the general population) are exaggerated. For example, in comparison to other faces, a face might have large ears, a large chin and close-set eyes. Consequently, in a caricature of this face the size of the ears and chin will be increased and the distance between the eyes will be decreased in relation to the rest of the face.
Brennan (1985) has developed an objective computer-based procedure for producing line-drawing caricatures of faces. Her software operates in two basic stages. Firstly, the features (e.g. the eyes, mouth and jaw-line, etc.) of a target face are outlined with a fixed number of manually positioned points to produce a database indicating their locations. Secondly, the target face database is compared to a norm database created by averaging the databases of a number of faces of the same sex. By exaggerating the differences between the locations of points on the target face and corresponding points on the norm database a caricature can be created. Similarly, by reducing the differences, the target face can be made to look more like an average face: this is referred to as an anti-caricature. More recently, Benson & Perrett (1991b) have developed a similar technique capable of producing continuous-tone caricatures of photographic quality from monochrome or colour pictures. Both Brennan (1985) and Benson & Perrett's (1991b) programs allow the user to generate caricatures (and anti-caricatures) of faces at any degree of exaggeration.
Using line-drawing stimuli, Rhodes, Brennan & Carey (1987) and Benson & Perrett (1994) have demonstrated that computer-generated caricatures of faces are named faster and judged to be better likenesses of the people than veridical (undistorted) or anti-caricature representations of the same faces. In its current form the IAC model does not present an obvious account of these caricature effects. Therefore, it is worth considering other models of face recognition that can.
Valentine and his colleagues have suggested that faces are encoded in terms of a number of visual features, stored as coordinates in a multidimensional face space (MDFS) (Valentine, 1991 a, b; Valentine & Endo, 1992; Valentine & Ferrara, 1991). Valentine has distinguished two types of model within the MDFS framework, one involving norm-based coding, the other exemplar-based coding. In a norm-based model faces are stored as vectors from a norm face (or average face). The norm face is abstracted from all faces a person has encountered and is updated each time a new face is encoded. In an exemplar model there is no norm and faces are encoded as discrete points in multidimensional space.
Valentine does not define which features of a face are encoded in MDFS but suggests that they may include hair colour, skin texture, configural information, etc. Furthermore, he makes the assumption that the values along each of the feature dimensions vary normally around a central tendency. Consequently, faces represented by a number of outliers in feature dimensions are stored in areas of low exemplar density (distinctive faces). In contrast, faces represented by feature dimension coordinates that approximate the central tendency are stored in areas of high exemplar density (typical faces). Valentine argues that face recognition is affected by the error associated with encoding the target stimulus and the number of neighbours surrounding the target representation in MDFS. Typical faces have a large number of close neighbours. Therefore, error at the encoding stage will have a more detrimental effect on their recognition. This is because a number of other representations besides the target representation will be activated. In contrast, distinctive face representations have few close neighbours. Therefore, they will be recognized more readily because relatively few representations in addition to the target (distinctive face) representation will be activated. In this way Valentine's MDFS conception accounts for the fact that distinctive faces are recognized faster than typical faces.
Valentine's exemplar-based and norm-based MDFS models present largely similar accounts of distinctiveness effects. However, the norm-based version provides the more appropriate metaphor within which to discuss caricature effects. Hence, if a veridical face A is represented in MDFS as vector A, then a caricature of face A might be represented as vector [A.sup.+], where vector [A.sup.+] has an identical direction to vector A but a longer length. Above we discussed Valentine's suggestion that distinctiveness effects arise because distinctive face representations have fewer neighbours than typical face representations. Consequently, there is less interference associated with recognizing a distinctive face than a typical face, because presentation of the former will activate fewer neighbours. In MDFS caricatures have a similar status to highly distinctive faces, because few faces (if any) will share the same values on the dimensions of the space as caricatures. In other words, a caricature of a face will have relatively fewer neighbours than its veridical representation. Therefore, although a caricature does not map directly onto its veridical representation, relatively less interference will be associated with recognizing a caricature. Exactly the same explanation applies in Valentine's exemplar-based model of MDFS.
In summary then, Valentine's model suggests that a caricature is more efficient at accessing a stored face representation because the process of caricaturing gives the face a distinctive-like status in MDFS. A number of authors have offered similar interpretations of caricature effects (Benson & Perrett, 1991a, 1994; Rhodes et al., 1987; Rhodes & McClean, 1990), but although this idea that caricaturing works by enhancing a face's distinctiveness seems theoretically plausible there is as yet little empirical evidence to support it. Distinctiveness effects and caricature effects have generally been studied separately using different paradigms, and this has made a direct comparison of the two effects difficult. However, the IAC model presents a very specific prediction; caricatured face primes should produce more cross-domain (face prime-name target) self-priming than typical face primes. This is because if caricatures are distinctiveness-enhanced faces, then they should produce more self-priming than anti-caricature or veridical representations of the same faces for exactly the same reasons that distinctive faces will produce more self-priming than typical faces. Experiments 2 and 3 investigated this prediction.
In Expt 1 face primes were monochrome photographs. In Expts 2 and 3 our stimuli were continuous-tone caricatures created using Benson & Perrett's (1991b) software.
Twelve subjects from the postgraduate and undergraduate populations of the University of Durham participated as subjects. Subjects were between the ages of 18 and 37 years and had normal or corrected to normal vision. They were paid for participating.
Ten monochrome photographs of famous personalities (see Appendix 2) were caricatured using Benson & Perrett's (1991b) continuous-tone caricature generator.
One hundred and eighty-six points were manually positioned on to each of the faces and joined to create a delineated representation containing 50 feature contours. Each of these features was represented by a fixed number of points, so that across faces there was conformity with respect to the number but not the positions of the points defining a particular feature. Each of the resultant delineated images were stored as an 186-point x/y coordinate database.
A caricature of a face was created by comparing its 186-point database to a norm database created by averaging the individual databases from a number of faces of the same sex, and exaggerating the differences between the two. The degree of exaggeration was calculated on a percentage scale; hence, a 50 per cent caricature was defined as one in which the differences between the norm and target face were exaggerated by a factor of 1.5.
In order to produce a continuous-tone image the caricatured and veridical images were divided into 340 triangular tessera. As an example, a triangulation might comprise the innermost points of the right and left eyebrows and the mid-point of the hair-line. The corresponding triangles in the veridical face and a caricature face were then compared, and the pixel intensity within each of the veridical triangles was mapped on to the corresponding caricature triangles. Where any of the caricatured triangles was larger than in the veridical image, 'stretching' of the spatial distribution occurred. Similarly, 'shrinking' of the spatial distribution occurred when one of the caricatured triangles was smaller.
Anti-caricature images were prepared using the same process, but instead of exaggerating the differences between the norm and target face the differences were reduced. Five levels of caricature were prepared for each face, 4-50, +25, 0, -25 and -50 per cent. Note that the terms -50 per cent caricature and a + 50 per cent anti-caricature are synonymous. Figure 4 shows an example of the stimuli used.
Monochrome transparencies were prepared of the 10 faces at each of the five levels of caricature. The images were photographed in such a way that the face filled the 36 x 24 mm frame. Transparencies were also prepared of the 10 personalities' names printed in uppercase Helvetica script (e.g. MARGARET THATCHER) and 10 unfamiliar names matched on forename and surname length.
Design and procedure
On each trial of the experiment a face was presented at one of the five levels of caricature (the prime), followed by an inter-stimulus interval and then a name (the target). Exactly the same presentation format used in Expt 1 was used here. Following a verbal signal from the experimenter the face prime was presented for 250 ms followed by an inter-stimulus interval of 250 ms, after which the target name was displayed for 2.5 s.
Three prime-target conditions, each with five sublevels of prime caricature were used in this experiment. Same: + 50, + 25, 0, -25 and 50 per cent; neutral: +50, +25, 0, - 25 and - 50 per cent; and unrelated: +50, +25, 0, -25 and -50 per cent. They were defined as follows:
Same. The prime face and target name were of the same person; e.g. a caricature of Margaret Thatcher's face followed by the target name MARGARET THATCHER.
Neutral. The target name was preceded by a caricature of an unfamiliar face; e.g. an unfamiliar caricature followed by the target name MARGARET THATCHER. The same unfamiliar neutral prime face caricatured at the five levels was used throughout.
Unrelated. The face prime and target name were of famous people who were not semantically related; e.g. a caricature of Anneka Rice's face followed by the target name MARGARET THATCHER.
Each of the 10 prime faces was used at all five levels of caricature in each of the same and unrelated prime type conditions. The same and unrelated conditions thus contained 50 prime-target experimental trials, and the neutral condition contained 25 prime-target pairs (five prime-target pairs at each of the five levels of caricature). Hence, there were 125 prime-target pairs to which the subjects were required to make a positive familiarity response. In addition, a further 125 prime-target pairs were added as 'No' response trials, where the 10 famous target names were replaced by 10 unfamiliar invented names matched for length; e.g. Ronald Reagan [right arrow] Andrew Waters. Hence, in total there were 250 prime-target pairs. Each subject saw all 250 of these prime-target pairs. Presentation of the stimulus pairs was pseudo-random with respect to prime type and familiarity of the target names. Each prime-target trial was separated from the next by an interval of approximately 3 s. The subjects' task was exactly the same as in Expt 1. Subjects were instructed to look at the prime but respond only to the target name by making a manual, button-press response to indicate whether the name was familiar or unfamiliar. Half of the subjects made a positive familiarity response with their right hands and half with their left. Reaction times were recorded from the onset of the target name and terminated by the subject's response.
Before starting the experiment, the subjects were presented with black and white slides of the 10 target names of familiar people and 10 target names of unfamiliar people written in uppercase Helvetica script (i.e. exactly as they were to appear in the main task itself). The names were pseudo-randomly presented with respect to familiarity and the subjects were required to make a familiarity decision to each name. The names were presented twice to ensure that the subjects were familiar with the target stimuli and practised in pressing the response keys. Following this, a practice block was run containing 10 of the stimulus pairs described above. The practice trials included examples from each of the three prime type conditions and included caricature, veridical and anti-caricatured face primes. Immediately after this practice block the main experimental trials were run.
Mean correct reaction times and mean error rates to familiar and unfamiliar target names are shown in Table 2. Subjects' response times to familiar target names preceded by the neutral primes (+ 50 [right arrow] -50 per cent) were pooled to give one overall mean for each subject. Error rates were low and will not be considered further.
Table 2. Mean correct reaction times (in ms) and mean error rates to familiar and unfamiliar target names preceded by same, neutral and unrelated primes, at the five levels of caricature: + 50, + 25, 0, -25 and -50 per cent. Maximum possible error rates (number of trials) for each of the prime type conditions are shown at the bottom of the table
Max. possible RTs Errors errors
Familiar targets Same (% caricature)
+50 499 0.2 10 +25 499 0.3 10 0 506 0.2 10 -25 511 0.2 10 -50 518 0.2 10 Neutral 561 0.2 25 Unrelated (% caricature) +50 570 0.2 10 +25 569 0.2 10 0 579 0.1 10 -25 577 0.4 10 -50 562 0.2 10 Unfamiliar targets 595 0.2 125
Analysis by subjects
A one-factor repeated measure ANOVA was carried out on the correct reaction time data to familiar target names. Prime type was the factor under investigation and had 11 levels (same: +50, +25, 0, -25 and -50 per cent; neutral (pooled); and unrelated: +50, +25, 0, -25 and -50 per cent; repeated measure). The results showed a significant effect of prime type (F(10, 110) = 14.673, p [less than] .001). Newman-Keuls tests ([Alpha] = 0.05) showed that responses to target names preceded by all of the same caricature primes (+50 [right arrow] -50 per cent) were faster than those to targets preceded by neutral or unrelated caricature primes (+50 [right arrow] -50 per cent). Responses to targets preceded by all of the unrelated caricature primes (+50 [right arrow] -50 per cent) did not differ from the responses to targets preceded by the neutral caricature primes. This overall result of facilitation from same primes without inhibition from unrelated primes is consistent with Posner & Snyder's (1975) criterion for automatic priming.
Newman-Keuls tests ([Alpha] = 0.05) on the five levels of same prime type did not reveal significant differences between any of the five levels. In short, whilst there was a clear facilitatory effect of prime type (same faster than neutral or unrelated), there was no significant effect of degree of caricature. However, inspection of Table 2 shows that reaction times in the same condition varied from 499 ms (+50 per cent caricature primes) to 518 ms (-50 per cent caricature primes) in the expected direction. Because the purpose of Expt 2 was to investigate effects of caricature on self-priming a t test was carried out as a planned comparison on the + 50 per cent caricatured and -50 per cent caricatured same primes in the same prime type. The results showed that target names preceded by the +50 per cent were responded to significantly faster than targets preceded by the -50 per cent same primes (t(11) = 2.21, p [less than] .05). Two further t test comparisons were carried out on the 0 and 50 per cent, and 0 and -50 per cent same prime types; neither of these two comparisons showed a significant difference.
Analysis by items
A one-factor repeated measure ANOVA by items was carried out on the data. Prime type was the factor under investigation and had 11 levels (same: +50, +25, 0, -25 and -50 per cent; neutral (pooled); and unrelated: +50, +25, 0, -25 and -50 per cent; repeated measure). The results showed a significant effect of prime type (F(10, 90) = 9.171, p [less than] .001). Newman-Keuls tests ([Alpha] = 0.05) carried out on the data showed exactly the same results as those found in the analysis by subjects, with overall faster reaction times on the same prime type condition, but no additional effect of caricature within this condition. Planned t test comparisons showed no significant difference in subjects' responses to target names preceded by the five levels of the same caricature primes (+50, +25, 0, -25 and -50 per cent) (t(9) [less than] 1.5).
The results showed suggestive but equivocal evidence of a caricature advantage. Newman-Keuls tests indicated that all five levels of caricature produced significant facilitation in the same prime condition, but none of the five levels differed significantly from each other. However, planned t test comparisons identified a significant difference between the -50 per cent caricature and +50 per cent caricature primes in the same prime condition, which reflected a distinct trend in the data towards increasing self-priming from the same -50 per cent primes through to the same +50 per cent. It was reasoned, therefore, that a clearer caricature effect might be found if a more sensitive design was used.
In addition, none of the five levels of unrelated prime type caricatures (+50, +25, 0, -25 and -50 per cent) produced inhibition. This result is consistent with the suggestion that the primes were being processed automatically (Posner & Snyder, 1975).
One factor that may have contributed to dampening the effect of caricature in Expt 2 was the large number of stimulus repeats. Subjects saw each of the 10 familiar faces in 20 of the prime type conditions (five same, five unrelated plus equivalent 'No' response trials; the neutral condition included an unfamiliar face prime). Hence, the subjects may have become very efficient at recognizing the identities of the face primes regardless of whether they were caricatures or anti-caricatures. Furthermore, subjects may then have been able to recognize some faces on the basis of a single salient feature. For example, under these circumstances the presence of long blonde hair may have been sufficient to lead to recognition of Anneka Rice without reference to the manipulated internal features of the face.
This observation highlights an important limitation of our continuous-tone caricatures. Benson & Perrett's (1991b) caricature process produces little or no change to a person's hair texture, hair tone and skin tone, etc. Hence, there were features of a face that may have provided cues to its identity that remained unchanged across the five levels of caricature used. Although obviously lacking the ecological validity of photographic quality stimuli, the caricatured line-drawings used by Rhodes et al. (1987) are not so susceptible to this problem, because these extraneous cues are not present.
In an attempt to combat any contaminating effects introduced by such factors the design was changed for Expt 3, so that each subject saw each name and face only once. A design without repeating stimuli required that the number of stimuli used should be increased (from 10 to 21). However, with the limited number of stimuli imposed by the need to use good full-face images, which are hard to find in press photographs of familiar people, it was not possible to include all 11 conditions used in Expt 2. The goal of the study was to determine whether a response to a name is speeded if it is preceded by a caricature of the same person's face as a prime, relative to the condition in which it is preceded by a veridical or anti-caricatured face prime. For this reason the neutral and unrelated conditions were dropped and comparisons were made across the +50, 0 and -50 per cent same prime type conditions only.
During the debriefing session after Expt 2 it became apparent that some of the subjects were aware that the stimuli were distorted in some way. Subjects mentioned that they were trying to work out 'what had been done to the stimuli' during the task. This factor may also have contributed to the results found. To prevent the subjects from studying the primes the prime-target stimulus onset asynchrony (SOA) was reduced in Expt 3. Unpublished work has found self priming for a 25 ms prime-target SOA (Calder, 1993). In Expt 3 we used a 50 ms SOA.
EXPERIMENT 3 Method
Twelve postgraduate members of the University of Durham participated as subjects. Subjects were between the ages of 21 and 45 years. All had normal or corrected to normal vision and were paid for participating. None of the subjects had taken part in Expts 1 or 2.
Twenty-one faces of famous personalities (including the 10 faces used in Expt 2; see Appendix 2) were caricatured at three levels of exaggeration (+50, 0 and -50 per cent), giving a total of 63 caricatures that were used as primes in the 'Yes' response trials. Targets in the 'Yes' response trials were the names of the 21 celebrities printed in uppercase Helvetica script. The primes for the 'No' response trials were composed of a number of faces of famous personalities caricatured at the same three levels of exaggeration (+50, 0 and -50 per cent) but judged by the experimenter to be relatively poor caricatures, on the basis that the original veridical images were of poor quality or the caricatured images contained glitches.
Design and procedure
The general design and presentation format were similar to those used in Expts 1 and 2. On each trial, following a warning signal from the experimenter, the prime was presented for 50 ms followed immediately by the target name, which remained in view for 2.5 s. Subjects were required to make a familiarity response to the target name as quickly and accurately as possible. Three prime type conditions were used; same +50, same 0 and same -50 per cent. They were defined as follows:
Same +50 per cent. A familiar target name was preceded by its corresponding caricatured (+50 per cent) face prime.
Same 0 per cent. A familiar target name was preceded by its corresponding veridical (0 per cent) face prime.
Same -50 per cent. A familiar target name was preceded by its corresponding anti-caricatured (-50 per cent) face prime.
The 21 familiar target names were split into three groups of seven, matched for mean familiarity from subjects' ratings. These three groups of seven faces were rotated across the prime type conditions so that each subject saw each prime face and each target name only once in one of the three prime type conditions. Hence, there were three experimental blocks of trials, and each block contained 21 face prime/name target pairs to which a positive familiarity response was required. In addition, a further 21 face prime/name target 'No' response trials were added to each block. The target names in the 'No' response trials were unfamiliar invented names, matched to the familiar target names for number of letters. The prime faces in the 'No' response trials were familiar faces caricatured at three levels of exaggeration (+50, 0 and -50 per cent). Hence, each of the three blocks contained a total of 42 face/name pairs. Presentation of the prime-target pairs was pseudo-random with respect to prime type and familiarity of the target names within each of the three blocks.
Each subject was presented with only one of these three blocks. First though, they were presented with 20 practice trials containing a veridical face prime and a name target; that is, the practice trials were made up of familiar and unfamiliar target names preceded by undistorted familiar faces. None of the names or faces used in the practice trials appeared in the main block of trials. These practice trials accustomed the subjects to the brief presentation of the face stimulus and gave them practice at pressing the response keys. The subjects were then presented with one of the three experimental blocks. This block started with a further six practice trials of the sort described above.
Table 3 shows the mean reaction times and error rates to target names preceded by same +50, same 0 and same -50 per cent prime types. Error rates were low and will not be considered further.
Table 3. Mean correct reaction times (in ms) and mean error rates to familiar and unfamiliar target names preceded by same face primes at three levels of caricature: +50, 0 and -50 per cent. Maximum possible error rates (number of trials) for each of the prime type conditions are shown at the bottom of the table
Familiar targets Unfamiliar targets Same +50% Same 0% Same -50%
RTs 586 621 635 724 Errors 0.3 0.1 0.4 0.8 Max. possible errors 7 7 7 21
The reaction time data were submitted to a one-factor ANOVA, where prime type (same +50, same 0 and same -50 per cent) was the factor under investigation. The results showed a significant effect of prime type (F(2, 22) = 8.485, p [less than] .005). Newman-Keuls tests ([Alpha] = 0.05) indicated that target names preceded by the +50 per cent same primes were responded to faster than those preceded by 0 and - 50 per cent same primes, which did not differ from each other.
The original hypothesis stated that caricatured faces should produce more self-priming than veridical faces or anti-caricatures. This was supported by the finding that a familiarity response to a target name is faster when it is preceded by a caricature of the same person's face, relative to the conditions in which a veridical or anti-caricature image precedes the target name.
These results are consistent with the idea that target names preceded by a caricature of their face are responded to fastest because caricatures of faces are more efficient at accessing their memory representations than veridical and anti-caricature representations. In essence, the results demonstrate a caricature advantage with continuous-tone caricatures.
Simulations with Burton et al.'s (1990) model of face recognition have indicated that for a self-priming task using face primes and name targets, the distinctiveness of the prime should interact with the amount of priming. This effect occurs in the model because a distinctive face prime produces a low level of inhibition within the FRU pool, allowing the prime's FRU, and consequently its PIN, rapidly to reach a high level of activation. In contrast, the model indicates that typical faces produce higher levels of FRU pool inhibition and low levels of activation of their FRUs and PINs. Given that a familiarity decision to a name is made at the level of the PINs, the model predicts that distinctive face primes should facilitate a familiarity decision to a target consisting of the same person's name more than will typical face primes. The results of Expt 1 were entirely consistent with this simulation.
Despite offering clear accounts of priming and distinctiveness effects, the IAC model does not present an immediately obvious account of the caricature advantage. However, Valentine has developed a multidimensional framework to account for 'front-end' processing (distinctiveness, race and inversion effects) that can readily accommodate caricature effects. In Expt 2 we used a hybrid of Valentine's MDFS and Burton et al.'s IAC model to generate the hypothesis that caricatures should show more self-priming than veridical or anti-caricatures in a self-priming task.
Within Valentine's MDFS framework caricatures have the same status as distinctive faces, therefore they can be thought of as 'distinctiveness-enhanced' representations of their faces. We reasoned that if this conception is correct, then we should be able to demonstrate that caricatured face primes produce more self-priming than veridical and anti-caricatured face primes. The results of Expt 2 showed no significant difference in the subjects' responses to target names preceded by a caricature (+25 and +50 per cent) of the same person's face, same veridical (0 per cent) or same anti-caricatured (-25 and -50 per cent) face primes. Nevertheless, a t test on the same +50 per cent and same -50 per cent primes indicated a significant effect of caricature, suggesting that such an effect would be worth pursuing with a more sensitive design.
Experiment 3 focused on a comparison between the conditions using the same person's face caricatured at +50, 0 and -50 per cent, with a minimal number of repetitions of each caricatured face. The results of Expt 3 demonstrated that target names preceded by the same face at +50 per cent caricature were recognized significantly faster than the condition in which the faces were caricatured by 0 and -50 per cent. Note that no effect of caricature had been found in Expt 2 for the unrelated condition. This result is consistent with the view that the primes were being processed automatically rather than strategically (Posner & Snyder, 1975); that is, facilitation from the same primes without inhibition from the unrelated primes. Experiment 3 did not include neutral and unrelated conditions due to the need to minimize stimulus repeats. However, this did not detract from our aim; to determine if caricatured representations of faces could access their PINs faster than veridical or anti-caricatured representations. Our results demonstrate that they can.
The experiments we have presented here are important for three reasons. Firstly, the results of Expt 1 are consistent with Burton et al.'s IAC accounts of self-priming and distinctiveness effects. Distinctiveness effects and self-priming are accounted for by different levels of the model, and it is impressive that the model is able to account for the interaction between these two factors found in Expt 1. Secondly, we have demonstrated that caricatured faces produce a similar pattern of effects to distinctive faces; that is, caricatured and distinctive face primes produce faster activation of their PINs than veridical and typical face primes. This result is consistent with the idea that caricatures can be envisaged as distinctiveness-enhanced faces in MDFS. Thirdly, we have demonstrated this caricature advantage with continuous-tone images of photographic quality.
Recently, Benson & Perrett (1991a) have investigated whether a caricature advantage on face recognition can be found with continuous-tone caricatures. Benson & Perrett (1991a) used a face-name matching task in which subjects were presented with a name followed by a face that could be a caricature, an anti-caricature or a veridical representation. The subjects' task was to decide whether the face and name matched. Benson & Perrett (1991a) found that subjects were fastest to make a correct mismatch decision when the face was a +16 per cent caricature; however, no effect of caricature was found for the match condition. This absence of an effect of caricature on the 'match' condition of Benson & Perrett's (1991a) study may be attributable to their design.
In Benson & Perrett's (1991a) caricature condition the subjects were asked to decide whether a caricatured face shared the same identity as the name that preceded it. However, caricatures are distorted representations of familiar faces. To that extent the subjects were being asked to match a name with a distorted representation of its corresponding face. Clearly the name Terry Wogan is a match for Terry Wogan's undistorted face, but it is less clear whether it is a match for a caricature of Terry Wogan. In other words, even though the caricatures can produce faster recognition, this benefit may have been counteracted by subjects' awareness that the representation was not entirely veridical. This is borne out by the observation that Benson & Perrett's (1991a) 'goodness of likeness' ratings indicated that overall, the caricatured and anti-caricatured images were often judged poorer likenesses of the persons they depicted than the veridical images.
The main difference between Benson & Perrett's (1991a) face - name matching task and our own face-name self-priming task was that in our task subjects were not required to respond to the face stimulus; we therefore have an indirect measure of caricature on recognition. Hence, in Expt 3 subjects responded most quickly to the +50 per cent same prime condition because there was no opportunity to study the prime and realize that it was a poor likeness of the person's face.
In summary, we have presented data predicted by Burton et al.'s IAC model of face recognition (Expt 1). However, because it was not intended to account for caricature effects the model does not present an immediately obvious interpretation of the results of Expts 2 and 3. We suggest that the results of all three experiments are best accommodated within a hybrid of Valentine's MDFS framework, which represents the best current account of 'front-end' processing, and Burton et al.'s IAC model, to date the best account of 'person priming'. In other words, a version of Burton et al.'s model in which the operation of the FRUs is seen in terms of an MDFS framework. This is not to suggest that the FRU concept is redundant, but that a point in an MDFS framework may provide a more useful metaphor for an FRU, than a single threshold unit. This standpoint has allowed us to explore the effects of caricature on recognition, and demonstrate that, as has been widely speculated, caricature effects can be seen as reflecting enhanced distinctiveness.
This work was supported by an MRC research studentship and an ESRC research grant (R0000234003). We thank the Press Association for permission to reproduce the photograph which forms the basis of Fig. 4, and Professor A. M. Burton and the British Psychological Society for permission to reproduce Figs 1 and 2. We also thank Duncan Rowland for preparing Fig. 4.
Benson, P. J. & Perrett, D. I. (1991a). Perception and recognition of photographic quality facial caricatures: Implications for the recognition of natural images. European Journal of Cognitive Psychology, 3, 105-135.
Benson, P. J. & Perrett, D. I. (1991b). Synthesising continuous-tone caricatures. Image and Vision Computing, 9, 123-129.
Benson, P. J. & Perrett, D. I. (1994). Visual processing of facial distinctivehess. Perception.
Brennan, S. E. (1985). The caricature generator: Dynamic exaggeration of faces by computer. Leonardo, 18, 170-178.
Bruce, V. & Valentine, T. (1985). Identity priming in the recognition of familiar faces. British Journal of Psychology, 76, 363-383.
Bruce, V. & Valentine, T. (1986). Semantic priming of familiar faces. Quarterly Journal of Experimental Psychology, 38A, 125-150.
Bruce, V. & Young, A. (1986). Understanding face recognition. British Journal of Psychology, 77, 305-327.
Burton, A. M., Bruce, V. & Johnston, R. A. (1990). Understanding face recognition with an interactive activation model. British Journal of Psychology, 81, 361-380.
Calder, A. J. (1993). Self priming in face recognition. Unpublished PhD Thesis, University of Durham.
de Haan, E. H. F., Bauer, R. M. & Greve, K. W. (1992a). Behavioural and physiological evidence for covert face recognition in a prosopagnosic patient. Cortex, 28, 77-95.
de Haan, E. H. F., Young, A. W. & Newcombe, F. (1992b). Neuropsychological impairment of face recognition units. Quarterly Journal of Experimental Psychology, 44A, 141-175.
Ellis, A. W., Young, A. W., Flude, B. M. & Hay, D.C. (1987). Repetition priming of face recognition. Quarterly Journal of Experimental Psychology, 39A, 193-210.
Perkins (1975). A definition of caricature, and caricature recognition. Studies in the Anthropology of Visual Communication, 2, 1-24.
Posner, M. I. & Snyder, C. R. R. (1975). Facilitation and inhibition in the processing of signals. In P.M. A. Rabbitt & S. Dornic (Eds), Attention and Performance V, pp. 669-682. London: Academic Press.
Rhodes, G., Brennan, S. E. & Carey, S. (1987). Identification and ratings of caricatures: Implications for mental representations of faces. Cognitive Psychology, 19, 473-497.
Rhodes, G. & McClean, I.G. (1990). Distinctiveness and expertise with homogeneous stimuli: Towards a model of configural coding. Perception, 19, 773-794.
Valentine, T. (1991a). Representation and process in face recognition. In R. J. Watt (Ed.), Pattern Recognition by Man and Machine, pp. 107-124. Basingstoke: Macmillan.
Valentine, T. (1991b). A unified account of the effects of distinctiveness, inversion, and race in face recognition. Quarterly Journal of Experimental Psychology, 43A, 161-204.
Valentine, T. & Bruce, V. (1986a). The effects of distinctiveness in recognising and classifying faces. Perception, 15, 525-535.
Valentine, T. & Bruce, V. (1986b). Recognising familiar faces: The role of distinctiveness and familiarity. Canadian Journal of Psychology, 40, 300-305.
Valentine, T. & Endo, M. (1992). Towards an exemplar model of face processing: The effects of race and distinctiveness. Quarterly Journal of Experimental Psychology, 44A, 671-703.
Valentine, T. & Ferrara, A. (1991). Typicality in categorisation, recognition and identification: Evidence from face recognition. British Journal of Psychology, 82, 87-102.
Young, A. W. & Bruce, V. (1991). Perceptual categories and the computation of 'grandmother'. European Journal of Cognitive Psychology, 3, 5-49.
Young, A. W., Flude, B. M., Hellawell, D. J. & Ellis, A. W. (1994). The nature of semantic priming effects in the recognition of familiar people. British Journal of Psychology, 85, 393-411.
The names of the 12 distinctive and 12 typical faces used in Expt 1 and mean ratings of distinctiveness and familiarity
Distinctive faces Dist. Fam. Typical faces Dist. Fam.
Kenneth Williams 6.4 6.3 Jason Donoven 1.8 6.7 Ken Dodd 6.0 5.6 David Steel 1.8 5.0 Rowan Atkinson 5.8 6.8 Philip Schofield 2.0 6.3 Les Dawson 5.7 6.1 Mel Gibson 2.2 5.1 Boy George 5.4 6.4 Michael Aspel 2.2 5.8 Bruce Forsyth 5.3 6.8 Emlyn Hughes 2.2 5.1 Mikhail Gorbachev 5.1 6.9 Roger Moore 2.3 5.1 Rod Stewart 5.1 5.9 Boris Becker 2.5 6.3 Telly Savalas 6.2 4.1 Hugh Laurie 2.9 6.1 Patrick Moore 5.0 5.9 Terry Wogan 2.6 7.0 Mick Jagger 4.8 6.1 Tom Cruise 2.6 6.0 Ronald Reagan 4.6 6.7 Jonathon Ross 2.6 6.8 Mean 5.5 6.1 Mean 2.3 6.0
The 10 faces caricatured at five levels of caricature (+50, +25, 0, -25 and -50 per cent) in Expt 2 and additional faces used in Expt 3
Experiment 2 faces Anneka Rice Margaret Thatcher Cyril Smith Bob Monkhouse Harold Macmillan John Cleese George Cole Stephen Fry Jonathon Ross Ken Dodd
Additional faces for Experiment 3 Harold Wilson Nicholas Parsons Winston Churchill Des O'Connor Cilla Black Steve McQueen Willie Carson Mikhail Gorbachev Philip Schofield Max Bygraves Gary Lineker
|Printer friendly Cite/link Email Feedback|
|Author:||Calder, Andrew J.; Young, Andrew W.; Benson, Philip J.; Perrett, David I.|
|Publication:||British Journal of Psychology|
|Date:||Feb 1, 1996|
|Previous Article:||Impulsivity and negative priming: evidence for diminished cognitive inhibition in impulsive children.|
|Next Article:||The Cognitive Neuropsychology of Schizophrenia.|