Keywords

1 Introduction

Constructing a compelling character with which people can empathize is an essential process in character design [1]. Successful characters allow audiences to put themselves in the character’s shoes, so to speak, and emotionally connect with the character. For this reason, a substantial body of research has focused on empathetic and sympathetic interaction between synthetic characters and users. For instance, research has shown that empathic emotions of the virtual character produced more positive feedback from people [2]. Also, another study set up a virtual classroom to teach children how to deal with bullying situations and showed that the children demonstrated empathy towards a victim character [3]. However, despite the increasing interest in virtual characters and how they can emotionally affect people, little attention has been paid to how the visual elements of a character influence people’s emotional responses. In this paper, the visual style of a character is examined to determine whether it affects viewer’s empathy and sympathy responses.

2 Character’s Visual Style

Characters are the direct agents with whom the audience communicates emotionally when watching an animation. In the book, The Illusion of Life, Thomas and Johnston discuss how designers should construct the characters carefully, considering all features a character has, from its costume, body proportions, facial features, to surrounding environment [4]. Some studies suggest that characters should be designed to look realistic [5, 6], whereas others suggest the opposite [7,8,9,10]. To investigate these contradictory ideas, we considered three visual styles of a character. The styles are based on the character hierarchy introduced by Bancroft in the book Creating Characters with Personality [11]. Bancroft’s character hierarchy includes six visual style types: iconic, simple (stylized), broad, comedy relief, lead and realistic (see Fig. 1). Each style varies in the degree of stylization, with iconic having the highest degree of stylization and realistic having the lowest degree of stylization. In detail, an iconic character is the simplest character and, therefore, very limited in expressions. A simple character is still stylized but can convey more expressions than an iconic character because it has more facial features. A broad character can express more than the first two, with its big eyes and a big mouth for extreme facial expressions. A comedy relief character is less stylized in facial features. It conveys emotions through subtle facial and body expressions. A lead character has realistic proportions and an expressive face. Lastly, a realistic character is the character with the most realism but still possessing some degree of caricature. The three characters’ visual styles selected for the present study were iconic, stylized and lead.

Fig. 1.
figure 1

taken from open access materials.

Examples of the character hierarchy with illustrations

Full size image

3 Prior Research on Virtual Characters and Audience Emotion

Prior studies have examined how virtual characters can be created to elicit emotion from the viewer. Paiva et al. highlighted the importance of the ‘proximity’ factor to evoke empathy in a role-playing game with 3D characters [12]. A few studies have shown that people feel empathetic towards the character if he/she is more similar to a human being [5, 6]. A study by Goetz et al. showed that the appearance of a robot influences subjects’ acceptance of and cooperation with the robot [13]. Ruttkay et al. argued that people may view realistic characters as more intelligent but may view non-humanlike characters as more appealing and entertaining [7]. McCloud, in his book Understanding Comics, claimed that iconic characters are more effective than realistic characters, as the audience’s involvement increases [8]. For that reason, iconic characters are often used commercially. Disney characters, with their iconized look and exaggerated expressions, engage the audience. People may prefer iconic agents because iconic agents are subject to fewer social norms [9]. A study by Adamo-Villani et al. investigated whether the visual style of signing avatars (realistic vs. stylized) affected viewers’ perception of the avatar’s appeal [10]. Results showed that the stylized signing avatar was perceived as more appealing than the realistic one. The ‘Uncanny Valley’ hypothesis, according to which people feel eerie and unpleasant when a high degree of realism (but not complete realism) in a character is reached [14], may explain why iconic characters can be more appealing and engaging than realistic characters.

4 Empathy and Sympathy

Many studies define empathy differently, but all definitions agree on the concept of sharing of ‘affect’. According to Eisenberg and Strayer, empathy is defined as “sharing the perceived emotion of another – ‘feeling with’ another” [15]. Often times, the terms, empathy and sympathy are used interchangeably, although they refer to different things. Sympathy, in contrast to empathy, is defined as “an emotional response stemming from another’s emotional state or condition that is not identical to the other’s emotion, but consists of feelings of sorrow or concern for another’s welfare [16]”. Empathy and sympathy are not mutually exclusive, but rather they are interconnected.

There are two aspects to empathy - mediation and its outcome. Mediation of empathy happens in two ways, via situations and via emotional expressions. Empathy is elicited via ‘situations’ when the observer interprets the emotion based on the situation another person is dealing with. Empathy is elicited via emotional expression when the observer concludes the emotional state based on the emotion the observed person is showing. These two ways of mediation produce the empathic outcome, which can be either cognitive or affective. A cognitive outcome means that the observer carries out an action to respond to empathy, for example, helping the target. An affective outcome is when the observer experiences an emotion because of the perception of the target.

In the study reported in this paper, in order to mediate via the situation, three scenarios were created, each of which could be easily understood by the audience. To mediate via emotional expressions, the audience needed to perceive the character’s emotions accurately. This emotion perception was achieved through the character’s body movements and facial expressions that were suitable to the story.

5 Assessing Empathic Response

Empathy can be divided into two categories: dispositional and situational. Dispositional empathy is a person’s general tendency to feel empathy and is used for psychological profiling. On the other hand, situational empathy is when a person empathizes in a given situation and is used to examine if a stimulus has an effect on people. Therefore, situational empathy is what the study reported in this paper aimed to assess.

There are several methods of measuring empathy. Hogan’s empathy (EM) and Mehrabian and Epstein’s questionnaire (QMEE) are commonly used to measure cognitive and emotional empathy [17]. However, these scales measure dispositional empathy and, hence, were not suitable for the study. It is difficult to measure situational empathy because empathy happens internally and therefore cannot be observed [18]. Biological measures are available, for example, through devices that read heart rate or breathing rate. However, due to the difficulty of analyzing such data, self-reporting methods are suitable for most studies measuring situational empathy [18]. For our study, a scale developed by Escalas and Stern was chosen to measure empathy and sympathy because of several similarities between their work and what this research aimed to measure [19]. Escalas and Stern investigated the effect of a classical drama advertisement and a vignette advertisement on viewers’ sympathy and empathy responses. The vignette ad was characterized by “multiple unconnected episodes (rather than a single unified plot), repetitive organization (rather than linear), and characters contained within each episode (rather than interacting with those in other episodes)” [19]. Those authors experimented with eight television (TV) ads, each one categorized either as a classical or a vignette ad. For the study, they created 10 survey questions and validated them by conducting two pretests with 147 participants. The first five questions tested whether subjects understood the feelings of the character and the situation happening in the TV ad, and therefore measured sympathy. The other five questions assessed whether the subjects felt like they were one of the characters or felt as though the events were happening to them, and therefore measured empathy. The underlying assumption of the survey questionnaire was that responses to the sympathy and empathy items are interconnected, not ‘mutually exclusive’. The authors asked the participants, ‘for the television commercial you just saw, please rate how descriptive each of the following statements is of how you personally reacted to this ad.’ Participants were given a seven-point scale ranging from not at all descriptive to very descriptive. For the statistical analysis, the means of the five items of sympathy and five items of empathy were calculated.

Escalas and Stern’s survey questionnaire was modified to fit the scope of this study. Questions were shortened to make them more easily readable, and a five-point scale ranging from strongly disagree to strongly agree was used. Table 1 presents the ten questions used in the study, where the first five questions assessed sympathy level and last five questions assessed empathy level. The fifth question said “last two videos” rather than all the videos because the first clip showed happy emotion, hence the character did not experience a problem.

Table 1. Modified sympathy/empathy scale used in the study
Full size table

6 Description of the Study

The objective of the study was to determine whether the degree of stylization of an animated character (e.g., lead, stylized, iconic) has an effect on viewers’ empathy and sympathy. The independent variable for the experiment was the degree of stylization in the character’s visual design. The dependent variables were the viewers’ empathy and sympathy ratings.

The hypotheses of the experiment were the following:

  • H0 = The degree of stylization of an animated 3D character has no effect on the audience empathy and sympathy levels.

  • Ha = The degree of stylization of an animated 3D character has an inverted U-shaped relation with the audience empathy and sympathy levels.

6.1 Stimuli

Three 3D animated characters were created with Autodesk Maya software. All three characters were of the same gender and approximately the same age. The only intended difference was the degree of stylization, which caused the proportion of the characters’ bodies and facial features to be different. Several features distinguished the three characters from each other. First of all, the body proportions were different, for instance, the head size compared to the body was the largest for the iconic character. The eyeballs were modeled and textured differently: The lead character used photo realistic textures on spheres, the stylized character used a simple color texture on spheres, and the iconic character’s eyes were modeled as solid-color flat disks. Facial features including eyebrow, ear, nose, lips, and hair were more detailed for the lead character and less defined for the iconic character. Textures for the lead character had the highest level of detail, as more detailed shades and wrinkles were added. For clothing, the colors and the designs were kept the same. Animation, rendering, compositing were also kept consistent across the characters. Figure 2 shows screenshots of the three characters.

Fig. 2.
figure 2

Neutral facial expression from the left, lead, stylized and iconic characters (face close-ups, top; full body shots, bottom)

Full size image

The three characters were rigged with the same number of joints and controllers, and key frame animations were created using the controllers of the lead character. By transferring the animations of the lead character to the stylized and iconic characters, we ensured that the three characters had the same movement. When transferring the animation, the rotational values of the joints were kept, but translations of the wrist and feet IK controllers were adjusted so that the movement was proportional to the character’s body. Some of the movements had to be edited slightly to avoid interpenetration issues caused by the differences in the character’s body proportions. Facial animations were created with blendshapes, which is a multiple-target morphing technique. In total, three blendshapes for the eyebrows, three blendshapes of the lips, and two blendshapes for the eyelids were produced. To have equivalent facial articulations among the characters, the amount of displacement of the eyebrows and lips for the emotions was proportional to the length of the character’s face. Figure 3 shows the blendshapes for lead, stylized and iconic characters.

Fig. 3.
figure 3

Blendshapes for lead (top), stylized (middle) and iconic (bottom) characters

Full size image

As previously discussed, in order to mediate empathy via emotion, the audience needs to recognize the emotions the characters are presenting. The emotions represented in the animations were happy, angry and sad, which are three of the six basic emotions identified by Ekman [20]. These three emotions were selected because they are easily recognizable. The selected emotions were expressed through characters facial expressions and mainly through body animation.

The scenario was also a significant factor within the experiment because participants needed to understand the situation in order for empathy to be mediated via situation. The situations represented in the stories were easy to understand and easy for subjects to relate to. Table 2 shows a brief summary of each story, and Fig. 4 shows nine frames extracted from the ‘sad animation’. The duration of each story ranged from 10 to 19 s. The order in which the stories were presented to the subjects was fixed (happy, angry and sad), whereas the order of presentation of the character’s visual style was randomized. The final animations were rendered at a resolution of 800 × 600 pixels, with a grey background and no sound. However, the screen resolution automatically changed if a subject decided to participate using a cellphone.

Table 2. Happy, angry and sad scenarios
Full size table
Fig. 4.
figure 4

Frames extracted from the “sad” animation, lead character (left), stylized character (middle), iconic character (right)

Full size image

6.2 Subjects

Seventy-one subjects, both undergraduate and graduate students participated in the study. All participants were students at Purdue University and were recruited via email. The age of the participants ranged from 18 to 45; 51% were male. A total of 48 participants indicated that they majored in Computer Graphics Technology (CGT); the others were non- CGT majors.

6.3 Procedure

Subjects were sent an email containing a brief summary of the research, an invitation to participate in the study, and the http address of the web survey. Participants completed the online survey using their own computers or smartphones, and the survey remained active for two weeks. In the survey, participants first indicated their age, gender, and college major, and then watched the three characters in random order. After viewing one emotion clip, subjects were asked to select what emotion the character was feeling; four choices were provided (happy, sad, angry and other, where option ‘other’ allowed users to type the emotion they thought the character was displaying in the clip). After viewing all three emotion clips for one character, the participants answered a questionnaire that measured their emotional response towards the character. After completing the questionnaire, they were asked to enter the age of the character and then proceed to the next set of clips. Participants could also leave comments about the experiment in an optional comment section.

7 Results

Out of the 71 people who participated in the study, only 61 responses were analyzed; the other responses were excluded due to the following reasons. Nine responses were excluded due to incomplete participation. One participant did not answer one question (e.g., the age of the iconic character), leaving a missing response to that question.

Identification of Character’s Emotion.

A total of 51 participants selected the correct emotions for all nine questions. The other ten subjects selected option ‘other’ and entered the following answers: joy, excitement, glee, intrigue, excitement, happiness and awe for happy animation; frustration, hurt, shock, disappointment, anger and rage for angry animation; and worried, depressed, shock, anxiety, shame, sadness, embarrassed for sad animation. According to the Feeling wheel [21], those emotional words can be categorized and narrowed down to be equivalent to happy, angry, and sad. Therefore, even though participants chose the option ‘other’, their answers were considered to be correct.

Empathy and Sympathy Levels.

The results were analyzed using a one-way ANOVA between character types and sympathy, and another one-way ANOVA between character types and empathy. These analyses showed no significant differences in subjects’ empathy and sympathy levels among the three characters’ visual types (sympathy: Mlead = 4.46, Mstylized = 4.53, Miconic = 4.45, F(2,180) < 1; empathy: Mlead = 2.64, Mstylized = 2.59, Miconic = 2.53, F(2,180) < 1). Also, a paired samples t-test was used to compare the responses between sympathy and empathy, and it was found that sympathy ratings were significantly higher than empathy ratings, t(182) = 21.17, p < 0.001).

For further analysis, a separate one-way ANOVA was used between empathy and gender (F(1,181) = 4.280, p = 0.040) and also between sympathy and gender (F(1,181) = 1.380, p = 0.242). Female participants tended to rate both sympathy, and empathy towards the characters higher than the male participants did. However, when analysis with a linear mixed-effects model for sympathy and empathy was calculated, the tendencies for gender differences were no longer evident in the statistics (Linear Mixed-Effects Model for sympathy and gender: F < 1 for empathy and F < 1 for gender).

In addition, from the same linear mixed-effects model, the effects of age, gender, college major of participants, order of characters, and characters visual type on participants’ empathy and sympathy levels were analyzed. Random effect of subjects and correlation among random errors were considered in fitting the model. Probability values assuming no effect for all the factors were obtained. For sympathy, all values were nonsignificant: gender (p = 0.365), college major (p = 0.986), character type (p = 0.219), character age (p = 0.975) and subject age (p = 0.122). Likewise, for empathy, all values were nonsignificant: gender (p = 0.444), college major (p = 0.220), character type (p = 0.974), character age (p = 0.483) and subject age (p = 0.733). Hence, we found no statistically significant effects in the study.

8 Discussion and Conclusion

The paper reported a study that investigated the effect of character’s visual style on audience empathy and sympathy. Findings from the statistical analyses, which included a one-way ANOVA model and a linear mixed model, revealed no significant difference in subjects’ empathy and sympathy levels among the three characters. This lack of effect is similar to the finding of Gruber, Aune, and Koutstaal for judgments of trust, who reported no significant difference in perceived trust of an automated decision aid as a function of whether the recommendations were made by a low or high anthropomorphic character [22]. The results of their study and ours suggest that any influence of character visual style, at least in the laboratory setting, is relatively weak. In addition to showing no main effect of character’s visual style, our study found no significant interaction between participants’ age, gender, college major, characters’ age and participants’ empathy and sympathy levels.

Although the study failed to reject the null hypothesis, it showed that the subjects’ sympathy level was significantly higher than the empathy level (p < 0.01). This difference could be due to two reasons. First, the short duration of the video might have prevented complete absorption needed for empathy [19]. Second, according to Escalas and Stern, it takes well-developed characters and a linear plot to evoke empathy [19]. This could explain why participants could understand the feelings and situation of the character (sympathy) but failed to completely identify themselves with the character.

The study had several limitations that future research can address. First, the characters were supposed to be the same age (to prevent age from being a confounding variable). However, the subjects’ ratings of the age differed across the three style types: A one-way ANOVA with character types as independent variables and the perceived character’s age as dependent variables showed that the lead and stylized characters were rated as much older than the iconic character (Mlead = 21.93 years, Mstylized = 20.70 years, Miconic = 12.71 years, F(2,179) = 50.151, p < 0.001). This one-way ANOVA shows a significant difference in the perception of the character’s age, but from the linear mixed-effects model, this difference did not affect the users’ ratings of sympathy and empathy. Second, it is possible that the visual differences between the lead and the stylized characters were not sufficiently distinctive. The differences could have been more exaggerated, for example, by making the lead character look closer to a realistic one, and by making the iconic character more abstract and less humanlike. In addition, for those subjects who took the survey on their cellphones, the differences between the characters’ styles might have been even less evident because of the lower screen resolution. Third, individual differences in sympathy and empathy levels exist. Some viewers may not reach the empathy stage due to the short runtime of the video clips, whereas others do. Lastly, for the optional comment section, two subjects reported perceiving the survey questions as unclear and too similar to each other. For instance, one participant commented, “Some questions are ambiguous, I think some of them just ask the same thing”. The fact that some users perceived the questions as being similar may have led them to give similar ratings rather than distinguish between the components being questioned.

To summarize, a convincing synthetic character plays a critical role in keeping the audience emotionally engaged throughout animation as it exhibits strong life-like presence. To enhance this experience, we empirically tested three visual styles of the characters and their influence on audience ratings of sympathy and empathy. Because creating more realistic characters is expensive in comparison to iconized characters, the lack of significant difference in sympathy and empathy ratings between character types suggests that the cost/benefit ratio for iconic characters may be best for situations in which the animator is seeking to elicit emotional responses from the viewer. However, the strongest result of the study was the somewhat surprising finding that the age of the iconic character was judged to be much younger than that of the lead and stylized characters from which it was derived. Whether that difference in perceived age is one that will interact with the age-appropriateness of scenarios is among the work that remains to be done to test a wider variety of visual types with more extended video clips.

According to the comments on the experiment, some of the survey questionnaires on empathy and sympathy were somewhat similar to each other. The empathy questions, ‘I experienced feeling as if the events were happening to me’ and ‘I felt as though the events in the videos were happening to me’ were ambiguous to participants as to how they are different from each other. The former question could be rephrased to ‘I experienced feeling because of the events’. Future research could address this issue by rephrasing the statements or better definitions and concrete examples.

In addition, future work could be performed with longer animation clips, richer scenarios, and characters that differ in visual style in a more salient way. Because research shows that characters’ movements may affect the believability of a character, the relationship between style of motion and empathy/sympathy responses needs to be investigated as well.