Emotion Words Link Faces to Emotional Scenarios in Early Childhood

Abstract

Recent theories have suggested that emotion words may facilitate the development of emotion concepts. The present study investigates whether emotion words affect children’s performance on an emotion category learning task. Across two experiments, 72 3-year-old children (49 female) were asked to identify which emotional face best matched particular emotional scenarios during 9 pre-test and 9 post-test trials. The scenarios in the present studies aligned with emotions typically learned among older age groups (annoyed, disgusted, and nervous). Between pre-test and post-test, children participated in training in which a facial configuration (annoyed, disgusted, or nervous) was paired with an associated scenario while they heard the emotion labeled explicitly or heard irrelevant information (Experiment 1) or heard a broad emotion label versus irrelevant information (Experiment 2). Aside from the labels presented, all other information was kept the same across conditions, including the specific faces and scenarios heard during learning trials. In Experiment 1, children’s emotion understanding increased more from pre-test to post-test in the explicit label versus irrelevant condition (t(34)=2.26, p=.030, d=0.75), but in Experiment 2 the broad emotion labels did not provide an advantage over irrelevant information (t(34)=0.72, p=.474, d=0.24). These results suggest that emotion labels may be particularly helpful for young children learning about unfamiliar emotions, because specific labels may help children to aggregate disparate emotional information into meaningful categories.

Understanding emotional information (i.e., cues which suggest how someone may be feeling) is a crucial social skill. By making inferences about how others may be feeling, individuals can predict likely behaviors for others (e.g., Olson, Astington, & Harris, 1988) and adjust their behavior accordingly. Ultimately, this is beneficial for social and academic development (Cassidy, Parke, Butkovsky, & Braungart, 1992; Denham et al., 2012; Fabes, Eisenberg, Hanish, & Spinrad, 2001; Izard et al., 2001; Voltmer & Von Salisch, 2017). Despite the importance of this skill, learning to understand others’ emotions is complex and protracted. Learning about emotions involves integrating information across multiple sources, including another person’s facial expression, vocal tone, and the situation in which the emotion is embedded, and generalizing emotions to other appropriate circumstances. Early in development, infants begin to perceive and interpret emotional information (LaBabera, Izard, Vietze, & Parisi, 1976; Sorce, Emde, Campos, & Klinnert, 1985; Walker, 1982; Walle & Campos, 2012). Around 18 months children begin to produce some emotion words such as “happy” (Ridgeway, Waters, & Kuczaj, 1985), and by age 2 children can map some emotion labels to expressions and map expressions to emotional scenarios (Denham, 1986). However, other emotions (such as disgusted, annoyed, and nervous) are learned later, and are typically not produced by children until closer to 5 years of age (Ridgeway, Waters, & Kuczaj, 1985). Additionally, children’s emotion vocabulary rapidly increases and becomes more adult-like from age 4–11 (Baron-Cohen, Golan, Wheelwright, Granader, & Hil, 2010; Grosse, Streubel, Gunzenhauser, & Saalbach, 2021). Further, children’s ability to interpret facial expressions develops gradually across the first several years of life (Widen, 2013) and the ability to talk about emotions abstractly follows an even more protracted developmental timeline, plateauing around age 18 (Nook, Stavish, Sasse, Lambert, Mair, McLaughlin, & Sommerville, 2020). Thus, emotion understanding involves many components that children must learn to interpret and categorize throughout childhood. However, the mechanisms that contribute to this development are less understood. The present manuscript will focus on one potential mechanism, emotion labels, and how they may affect children’s emotion categorization.

Recent theoretical work suggests that language may be a key mechanism behind young children’s ability to learn about emotions (Barrett, 2017; Lindquist, MacCormack, & Shablack, 2015). Language has been shown to aid category learning in other domains. One potential reason for this is because words may serve as “invitations to form categories” (Brown, 1958) by highlighting commonalities among category members. Although broad labels may be less helpful for category formation because they can involve connections to too many concepts, specific labels may help children to aggregate disparate emotional information into a single emotion category. For example, prior research has shown that infants are capable of forming categories if category members are labeled, but not when another auditory stimulus (e.g., a tone) is paired with category members (Fulkerson & Waxman, 2007). Such evidence suggests that language may be an important aspect of category learning (Althaus & Mareschal, 2014; Levinson, 1997; Lupyan, Rakison, & McClelland, 2007). In particular, language may aid children in learning abstract concepts (Loewenstein & Gentner, 2005) such as emotion (Hoemann, Wu, LoBue, Oakes, Xu, & Barrett, 2020).

Emotions are abstract concepts. The same emotion can be displayed in a variety of ways and perhaps has no defining perceptual marker at all. That is, there is no concrete form for the features of any given emotion (e.g., people may or may not scowl when angry; Barrett, Adolphs, Marsella, Martinez, & Pollak, 2019), making it challenging for children to identify what determines emotion category membership across individuals and contexts. When learning about emotions children are tasked with integrating information from many sources and linking multiple diverse displays under the same emotion category and label (Lindquist & Gendron, 2013). If children learn about emotions similarly to how they learn other abstract categories, labels may help children to identify and form meaningful categories of emotion (Hoemann, Xu, & Barrett, 2019). This may be because specific emotion labels help children to group emotional information based on commonalities (e.g., hearing two people described as “afraid” may help children to link these instances into a single emotion category), or labels may link different aspects of emotion together, such as faces, contexts, or body posture, ultimately helping children to organize and develop emotion concepts (Shablack & Lindquist, 2019). Alternatively, it could be that emotion language draws children’s attention to emotional content more generally. In either case, recent theoretical work suggests that language may facilitate emotion category learning in a similar manner to how labels aid other forms of abstract category learning.

Some evidence exists to support this theoretical position. Research with adults has shown that across three distinct types of methodologies (semantic satiation, studying participants with semantic dementia, and absence of emotion words in the task), reduced access to emotion words results in disrupted processing of emotional faces (Gendron, Lindquist, Barsalou, & Barrett, 2012; Lindquist, Gendron, Barrett, & Dickerson, 2014; Nook, Lindquist, & Zaki, 2015). Additionally, the presence of labels has been shown to bias perceptual memory for emotional faces (Doyle & Lindquist, 2018) and facilitate adults’ categorical perception of chimpanzee facial expressions (Fugate, Gouzoules, & Barrett, 2020). Research with adults has also indicated that language aids emotion perception, and that this cannot be accounted for by evoking situational context, but rather a unique contribution of labels (Doyle, Gendron, & Lindquist, 2021). Further research suggests a relation between emotion understanding and language in childhood. For example, emotion understanding is correlated with children’s language abilities at 4 years of age (Cutting & Dunn, 1999), and language contributed more unique variance to children’s emotion understanding than their age, family background or false belief understanding. Emotion understanding and receptive language are also correlated among a broader age range of children, from 4 to 11 years (Pons, Lawson, Harris, & De Rosnay, 2003). Thus, emotion understanding and language appear to be related in both adults and children.

However, although previous research shows associations between language and emotion understanding in childhood, this research has been largely correlational. Therefore, it is unclear whether the development of emotion understanding is influenced by language, whether language development is influenced by emotion understanding, or whether a third variable (e.g., domain-general cognitive ability) may account for development in both language and emotion understanding. Recent theoretical work (e.g., Barrett, 2017) suggests that emotion words should facilitate the early development of emotion concepts beginning as early as infancy. Although infants are nonverbal, receptive language may help infants to identify similarities across various emotion categories even before infants can produce words (Shablack, Stein & Lindquist, 2020). However, evidence demonstrating that changes in labels influence perception or understanding of emotion are currently limited (but see Ruba, Meltzoff, & Repacholi, 2020). Some prior work has begun to examine how children learn emotion words (Shablack, Becker, & Lindquist, 2020), but less is known about how emotion words may or may not facilitate learning about emotion concepts.

The Present Study

Across two experiments, we examined whether the presence of emotion labels helps children to learn emotion categories. Three-year-old children participated in an emotion understanding generalization task. This age was selected because prior research has shown that emotion understanding at age three predicts emotion understanding at age six (Brown & Dunn, 1996), suggesting that three may be an important age at which to study early emotional development. Children were tested for their knowledge of the emotion categories annoyed, disgusted, and nervous. Previously, parents have reported that their 3-year-old children were unlikely to use these emotion words, as the percentages of children this age who were reported to produce these words were as follows: Annoyed: 23.3%, Disgusted: 18.4%, Nervous: 8.9% (Ridgeway, Waters, & Kuczaj, 1985). Additionally, a recent transcript analysis indicated that across nearly 2,000 naturalistic parent-child interactions with children ranging from 1 to 3 years of age, the words “annoyed”, “disgusted”, and “nervous” were produced 0 times by children (Ogren & Sandhofer, in press), lending further support to the notion that 3-year-old children are relatively unfamiliar with these specific emotion words. Thus, 3-year-olds represent the perfect age group for the present study, as they know enough language and have the patience to complete a task involving many brief vignettes, but do not yet know these specific emotion words.

At pre-test children heard brief vignettes in which a character experienced an emotion that is typically learned in later ages by children (annoyance, disgust, or nervousness) and were asked to select the face from an array that best matched how the character in the story would feel. Then, children observed learning trials where they heard the experimenter provide new vignettes and label the character’s emotional reaction using either an explicit emotion label (e.g., “feels disgusted”) or irrelevant information (e.g., “sits down”; Experiment 1), or a broad emotional label (e.g., “feels bad”) versus irrelevant information (Experiment 2). Across both experiments all children heard consistent language for each emotion, regardless of condition. Therefore, if children only needed consistent linguistic information paired with emotions across learning trials to perform well in the task, they should learn equally well across all conditions. Finally, children completed post-test trials which mirrored the pre-test trials but with new scenarios, requiring children to generalize emotional information to new faces and scenarios.

We conducted confirmatory analyses investigating whether changes in children’s performance differed by labeling condition. We hypothesized that children would perform better in the explicit emotion label condition than the irrelevant condition, but that the broad label condition would not differ from the irrelevant condition. This hypothesis is driven by theory which suggests that specific emotion words help children to group emotional information into meaningful categories (Barrett, 2017). However, if the label is too broad, too much information may be linked to it to be helpful for specific category formation, and therefore we reasoned that broad emotional labels may not offer the same advantages as they may serve only to draw children’s attention to emotional information.

Experiment 1

To determine whether explicit emotion labels may affect children’s emotion category learning, we examined whether presenting children with explicit emotion labels (vs irrelevant information) would affect understanding of emotion categories in a pre-test post-test design. We hypothesized that explicit emotion labels would help children to group relevant emotional information together into meaningful categories, and therefore that the performance of children in the explicit label condition would increase more than children in the irrelevant (control) condition.

Method

Participants

Thirty-six participants (25 female) ranging from 3.02 to 3.88 years of age (M_age=3.53 years, SD_age=0.22 years) participated in this study. Three additional children were excluded due to experimenter error (N=2) or failure to complete all trials (N=1). This sample size was selected by conducting a power analysis in G*power 3.1 (Faul, Erdfelder, Buchner, & Lang, 2009). Although no prior developmental studies have examined this specific research question, previous studies in the broader developmental literature assessing how information presented during training trials affects young children’s later learning in a 3- or 4-option forced choice paradigm have often found large effect sizes (e.g., Axelsson, Swinton, Winiger, & Horst, 2018; Kucker, McMurray, & Samuelson, 2020; Vlach & Sandhofer, 2011). Thus, using a large effect size (d=1.00), α=.05, and Power=0.8 to power for an independent samples t-test comparing difference scores across conditions, we determined that a necessary total sample size was 34. We added two additional participants who were already scheduled when the full sample size was reached. Participants were recruited from lists of birth records provided by the County and using the Children Helping Science website. Thus, participants were located across the United States. The racial breakdown of participants was as follows: White (N=25), Multiracial (N=7), Asian (N=2), African-American (N=1), Chose not to respond (N=1). Parents provided informed consent prior to data collection, and received a $5 Amazon gift card for participating. Data were collected from September 2020 through November 2020. This study received approval from the University of California, Los Angeles Institutional Review Board (IRB # 10–001578). This study was not preregistered.

Materials

Children first completed 9 pre-test trials (3 of each emotion), then 6 learning trials (2 of each emotion), and finally completed 9 post-test trials (3 of each emotion). Stimuli for these trials included brief vignettes, drawings related to the vignettes, and images of facial expressions.

Twenty-four vignettes were constructed to clearly convey annoyed, disgusted, and nervous scenarios (8 vignettes per emotion; Supplemental Material). Each vignette involved a female character engaged in a brief scenario that aligns with stereotypical experiences of the particular emotion. Annoyed vignettes all involved characters experiencing something irritating occurring repeatedly. Disgust vignettes all involved characters experiencing something gross or unpleasant. Nervous vignettes all involved characters experiencing something nerve wracking or uncertain. All vignettes involved distinct scenarios, but clearly adhered to these criteria. For example, a child listening to a nervous vignette might hear “Gabriella is going camping with her family. She will be sleeping in a tent outside. She has never done this before and isn’t sure if it will be fun or not.” Additionally, a drawing was included with each vignette (Supplemental Material). These drawings were provided as visual cues to help children remember the information from the vignette without providing clues to the emotion (e.g., For the vignette about camping, the drawing depicted a tent and campfire). These drawings were always paired with their corresponding vignette.

To create the facial expression stimuli, 11 undergraduate women were recruited. Only women were used as stimuli because mothers tend to contribute the most to childcare across early development (Fillo, Simpson, Rholes, & Kohn, 2015), making female faces typically more familiar to young children. Each undergraduate was recorded from the shoulders up as they stood in front of a solid white background. Undergraduates were asked to think about a time when they felt a target emotion (anger, annoyance, disgust, fear, happiness, nervousness, and sadness) and to convey this through their facial expression. The camera recorded these expressions, and the peak intensity of each emotional expression was extracted from these recordings. Images were edited such that the faces were all centered and approximately the same size.

Stimulus Validation

A Qualtrics survey was conducted to determine whether adults perceived the faces described above to belong to the intended emotional categories. Twenty-six adults (3 male) completed the survey, which required responses to every question. Each adult viewed 77 images (11 women, 7 emotions per woman) and were asked which emotion (afraid, angry, annoyed, disgusted, happy, nervous, or sad) best described each image. Adults were also asked to rate how confident they were for each response (Likert scale 1–5 from “not at all confident” to “very confident”). Based on these responses, the 9 women who had the highest agreement regarding their expression of the emotions were selected. On average for these images, the adult raters agreed on the intended emotion 84.1% of the time with average confidence ratings of 3.8. Agreement ratings by emotion were as follows: Afraid=82.1%, Angry=82.1%, Annoyed=77.7%, Disgusted=94.6%, Happy=95.3%, Nervous=53.1%, Sad=93.6%. Considering that each emotion had a 1 in 7 (14.3%) chance of being selected for any given image, these ratings are all substantially above chance levels. Confidence ratings by emotion were as follows: Afraid=3.4, Angry=3.8, Annoyed=3.9, Disgusted=4.0, Happy=4.2, Nervous=3.3, Sad=4.0. No single selected image was agreed upon by adults less than 50% of the time, and the intended emotion category was always the most frequently selected response for the selected stimuli. Even though the average adult responses for the nervous faces were lower than the other emotions, nervous was still correctly identified by the adults at levels substantially higher than chance, and for each nervous image the adults selected “nervous” as the correct response substantially more often than the next-closest response option. Thus, we determined that the selected images were viewed by adults as appropriate representations of the seven intended emotion categories.

Procedure

A live-action assessment of emotion category learning was administered entirely online via Zoom by an experimenter, with 9 pre-test trials followed by 6 learning trials, and finally 9 post-test trials. All experiments were conducted on a screen large enough for children to see the stimuli (i.e., a computer screen or tablet). No experimental sessions were conducted using a phone screen. During the task, parents were asked to stand behind their child and not to influence their child’s responses in any way. Participants were tested for their knowledge of three emotion categories which are typically challenging for children to recognize in this age range: Annoyed, Disgusted, and Nervous (Ridgeway, Waters, & Kuczaj, 1985).

Pre-Test.

During the pre-test, children heard 9 brief stories. For example, “Gabriella is going camping with her family. She will be sleeping in a tent outside. She has never done this before and isn’t sure if it will be fun or not.” The experimenter told the children each story in a neutral tone of voice without displaying any clear emotion through their facial expression. While the story was presented, the drawing aligned with that story was presented in the center of the screen to remind the children of the story events. In the case of the story above, the drawing depicted a tent and campfire. After completion of the story, the experimenter presented 4 emotional faces- one in each corner of the screen, with the drawing still present in the center. The experimenter then asked, “Which picture shows how Brittany feels?”, and children were tasked with pointing to their choice on the screen. The experimenter noted the child’s pointing response based on the video feed from Zoom. If the experimenter could not clearly identify where the child was pointing, they asked the parent to simply report the letter (A, B, C, or D) next to their child’s selected response. One of the pictures was the correct response (in this case, nervous), one was always a happy expression (in case children were not paying close attention and simply chose the most positive image), and the other two pictures represented familiar, but incorrect responses (e.g., sad and angry). Anger, sadness, and fear are typically known by 3-year-old children (Denham, 1986), and thus these negative emotions represented the familiar but incorrect choices. Children participated in nine pre-test trials (three for each target emotion), and their choices for each trial were recorded.

Learning Trials.

Children completed six learning trials, two for each emotion (annoyed, disgusted, and nervous). Children were randomly assigned to one of two learning trial conditions.

In the Explicit Label condition, children listened to six scenarios (see Supplemental Material) while a picture of the appropriate emotional face was displayed. Each scenario was followed by an explicit label for the emotion (e.g., “She feels disgusted”). The experimenter then asked the children to point to the girl who felt [annoyed/disgusted/nervous], and then asked the children to repeat the label by prompting “How does she feel?”. Asking the children to point to the face and repeat what they had heard during learning trials served the dual purposes of ensuring that children were paying attention and providing them with additional opportunities to learn the information.

The Irrelevant condition served as a control to determine whether simply having additional exposure to emotional faces paired with scenarios without emotion labeling boosted children’s performance at post-test. That is, comparing the explicit condition to the irrelevant condition was important to ensure that any differences from pre- to post-test were not driven simply by children seeing the correct faces paired with the appropriate scenarios during learning trials. In the Irrelevant condition, children heard the same scenarios and saw the same faces as the explicit label condition, but the information presented after the scenario was irrelevant to the character’s emotion (e.g., “She sits down”). After hearing the scenario, children in the irrelevant condition were also asked to point to the girl who [sneezes/sits down/wants a snack] and were asked to repeat the information by prompting “What did she do?”. Thus, both conditions involved the child pointing to the character on the screen and repeating what they had heard the character did or felt. The specific emotional faces presented during the learning trials were never shown during pre-test or post-test trials, ensuring that children would need to generalize learning trial information to new faces and new scenarios at post-test.

In both learning trial conditions, the information paired with the categories of emotional faces was consistent (e.g., disgusted faces were always paired with either “She feels disgusted” or “She sits down”; nervous faces were always paired with either “She feels nervous” or “She wants a snack”). All phrases presented during learning trial conditions were short, and phrase length did not notably differ between the two conditions. Table 1 indicates which information was paired with each emotion’s two learning trials by condition. All six learning trials were presented in succession, in a randomized order for each participant.

Table 1.

Information presented after the scenarios during learning trials for each condition and emotion in Experiment 1.

	Annoyed	Disgusted	Nervous
Explicit Label	“She feels annoyed”	“She feels disgusted”	“She feels nervous ”
Irrelevant	“She sneezes ”	“She sits down”	“She wants a snack”

Post-Test.

Immediately following the learning trials, children participated in nine post-test trials. The specific scenarios that were presented at pre-test vs learning vs post-test were randomized for each child. Each of the three emotion categories (annoyed, disgusted, nervous) was assessed in three different trials, for a total of nine post-test trials. These trials mirrored the pre-test trials using different scenarios, where children heard a brief scenario, were presented with four faces, and were asked to point to the face that showed how the character felt. All post-test trials involved new scenarios that the children had never heard before, requiring children to generalize any information that they learned. In all conditions, children’s responses to the nine trials were recorded. Study data and materials will be made available by the authors upon request.

Results

Descriptive Statistics

Each trial presented children with 4 faces to choose from, thus by chance children would make 2.25 out of 9 correct responses. We first examined only the pre-test responses to identify how challenging the task was for children prior to any learning trials. Results revealed that on average (across both conditions) children made 2.68 correct choices at pre-test (SD=1.21). Further, no children responded incorrectly to all 9 trials, and no child answered correctly on more than 6 out of 9 trials. Thus, we inferred that pairing annoyed, disgusted, and nervous scenarios with faces depicting that emotion was challenging for children. This aligns with previous research which suggests that 3-year-old children are generally unfamiliar with these emotion categories (Ridgeway, Waters, & Kuczaj, 1985).

Additionally, we investigated whether children’s performance at pre-test differed by the specific emotion. Results revealed that there was a significant effect of emotion at pre-test (F(2, 105)=26.84, p<.001, η_p²=0.34), with children’s average number of correct choices by emotion (out of 3 possible) as follows: Annoyed=0.64 (SD=0.87); Disgusted=1.67 (SD=0.96); Nervous=0.36 (SD=0.49). Thus, although the emotions were generally challenging for children at pre-test, performance was overall higher for disgusted scenarios compared to annoyed or nervous.

Effect of Emotion Labels

To determine whether the emotion labels presented during the learning trials affected children’s emotion categorization performance for annoyed, disgusted, and nervous scenarios and faces, we compared children’s performance across the two labeling conditions. Average change scores from pre-test to post-test by condition were as follows: Explicit Label=1.00 (SD=1.68); Irrelevant=−0.28 (SD=1.71). These values represent post-test scores minus pre-test scores, and therefore can be interpreted as, on average, how many more questions children answered correctly at post-test relative to pre-test. As Figure 1 shows, we conducted an independent samples t-test comparing the change scores (post-test – pre-test) across the two conditions (Explicit Label and Irrelevant). Children in the explicit label condition learned more from pre-test to post-test than children in the irrelevant condition (t(34)=2.26, p=.030, d=0.75, 95% CI=[0.13, 2.43]).

Figure 1.

Change in number of correct responses from pre-test to post-test for the explicit label and irrelevant condition in Experiment 1. Error bars indicate standard error.

We further examined whether performance in either condition was significantly different from what would be expected by chance (0 change from pre-test to post-test). One-sample t-test results revealed that the change score from pre-test to post-test was significantly different from chance in the Explicit Label condition (t(17)=2.53, p=.022, d=0.60, 95% CI=[0.16, 1.84]), but was not different from chance in the Irrelevant condition (t(17)=−0.69, p=.500, d=0.16, 95% CI=[−1.13, 0.57]).

Follow-up Analyses

Considering the significant difference between the two conditions, we followed up with additional post-hoc analyses to determine whether the effect of labeling may have been more pronounced for some children than others. First, we investigated whether the labels had a differential effect based on child age, as older 3-year-olds typically have higher vocabularies as well as more experience with emotions in the world around them relative to younger 3-year-olds. Overall, within this sample of children, age did not significantly predict children’s change score from pre- to post-test (B=.17, p=.905, 95% CI=[−2.66, 3.00]). Further, there was no significant interaction between children’s age and labeling condition for their pre-test to post-test change scores (B=−2.49, p=.094, 95% CI=[−5.42, 0.45]). Figures 2 depicts the relation between child age and their difference score in the explicit and irrelevant conditions.

Figure 2.

Change in number of correct responses from pre-test to post-test plotted with child age for the irrelevant (black) and explicit (blue) conditions in Experiment 1.

Next, because previous research has demonstrated that preschool-aged girls use more emotion language than boys (Kuebli, Butler, & Fivush, 1995), we examined whether there were any gender differences. Overall, average change scores for girls (M=0.56, SD=1.85) and boys (M=−0.09, SD=1.64) were comparable to one another (t(34)=1.00, p=.322, d=0.37, 95% CI=[−0.67, 1.97]). Additionally, there was no significant interaction between child gender and labeling condition for children’s change scores (F(1,32)=0.47, p=.693, η_p²=0.01).

We also assessed whether the effect of condition (explicit versus irrelevant) may have differed for the three separate emotions included in the present study. Results revealed no significant interaction for emotion by condition (F(2, 102)=1.83, p=.166, η_p²=0.04). Thus, the effect of explicit versus irrelevant information during the learning trials was not significantly different across the annoyed, disgusted, and nervous trials.

Interim Discussion

The results from Experiment 1 indicate that children learned significantly more about new emotions in the explicit label than in the irrelevant (control) condition. Further, there was no significant interaction between condition and children’s age (restricted to 3-year-old participants), gender, or the specific emotion on performance. However, it remains possible that children simply needed any information directing them toward emotional cues to help them learn in this task. That is, it may not have been the emotion words, specifically, that aided children’s performance, but rather any information indicating that emotional information should be attended to. To examine this possibility, we conducted Experiment 2.

Experiment 2

In Experiment 2, we examined whether broader emotional labels would similarly provide a benefit to children learning about emotions. Broad emotion words (e.g., “she feels good” or “she feels bad”) are typically familiar, known labels to young children, and may aid children’s performance in a similar way to explicit labels because they may draw children’s attention to feelings and emotional content. However, drawing children’s attention to emotional information may be insufficient to benefit learning. If this is the case, children may not see a benefit of broad labels as they do not help children to group emotional information more narrowly. To assess whether broad emotion labels may afford an advantage to children learning about new emotion categories, we conducted a study very similar to Experiment 1. Here, we examined whether presenting children with broad emotion labels (vs irrelevant information) would also help young children learning new emotion categories.

Participants

Thirty-six participants (24 female) ranging from 3.01 to 3.88 years of age (M_age=3.40 years, SD_age=0.26 years) participated in this study, exactly matching the sample size from Experiment 1. Four additional children were excluded due to experimenter error (N=2) or failure to complete all trials (N=2). Participants were recruited in the same manner as Experiment 1, using lists of birth records provided by the County, and using the Children Helping Science website. The racial breakdown of participants was as follows: White (N=20), Multiracial (N=14), Asian (N=1), African-American (N=1). Parents provided informed consent prior to data collection and received a $5 Amazon gift card for participating. Data were collected from November 2020 through February 2021. This study received approval from the University of California, Los Angeles Institutional Review Board (IRB # 10–001578).

Materials

The same materials were used in Experiment 2 as in Experiment 1. This included all vignettes, facial expression stimuli, and drawings.

Procedure

The procedure for Experiment 2 was also conducted entirely online and was identical to Experiment 1, with the exception of the learning trial conditions. Half of the children were still randomly assigned to an Irrelevant condition, which was identical to that used in Experiment 1. However, the other half of children were presented with Broad Labels. In the Broad Label condition, during the learning trials children heard a brief, nonspecific description of the emotion (e.g., “She feels bad”), also with a picture of the appropriate emotional face displayed, and were asked to point to the girl and to repeat the label. The information presented after each scenario in learning trials for Experiment 2 is depicted in Table 2. All other procedures operated in the same manner as Experiment 1.

Table 2.

Information presented after the scenarios during learning trials for each condition and emotion in Experiment 2.

	Annoyed	Disgusted	Nervous
Broad Label	“She doesn’t feel good”	“She feels bad”	“She doesn’t like this feeling”
Irrelevant	“She sneezes ”	“She sits down”	“She wants a snack”

Results

Descriptive Statistics

As with Experiment 1, we first examined how many correct responses children made at pre-test. On average (across both conditions) children made 2.57 correct choices at pre-test (SD=1.34). One child was incorrect on every trial at pre-test, however no children answered more than 5 out of 9 trials correct at pre-test. Thus, similar to Experiment 1, pre-test performance was close to chance and therefore the task was similarly challenging for children.

As with Experiment 1, we examined whether children’s performance at pre-test differed across the three emotions. Results revealed that there was a significant effect of emotion at pre-test (F(2, 105)=24.98, p<.001, η_p²=0.32). Children’s average performance by emotion (out of 3 possible) was as follows: Annoyed=0.56 (SD=0.65); Disgusted=1.56 (SD=0.88); Nervous=0.44 (SD=0.65). Thus, similar to Experiment 1, performance was overall higher for disgusted scenarios relative to annoyed or nervous scenarios at pre-test.

Effects of Emotion Labels

Average change score from pre-test to post-test by condition were as follows: Broad=0.11 (SD=1.45); Irrelevant=−0.22 (SD=1.31) (See Figure 3). An independent samples t-test was used to compare the change scores (post-test – pre-test) across the two conditions (Broad Label and Irrelevant). This analysis revealed that the change in responses from pre-test to post-test did not significantly differ by labeling condition (t(34)=0.72, p=.474, d=0.24, 95% CI=[−0.60, 1.27]). Thus, overall children’s learning from pre-test to post-test did not differ in the broad label compared to the irrelevant (control) condition.

Figure 3.

Change in number of correct responses from pre-test to post-test for the broad label and irrelevant condition in Experiment 2. Error bars indicate standard error.

As with Experiment 1, we assessed whether performance in either condition was significantly different from what would be expected by chance. One-sample t-test results revealed that the change score was not significantly different from chance in either the Broad Label condition (t(17)=0.33, p=.749, d=.08, 95% CI=[−0.61, 0.83]) or the Irrelevant condition (t(17)=−0.72, p=.481, d=0.17, 95% CI=[−0.87, 0.43]).

Additionally, we collapsed across Study 1 and Study 2 to examine whether the three conditions significantly differed from one another. The results of the one-way ANOVA revealed that children’s difference score from pre- to post-test did significantly differ across conditions (F(2, 69)=3.99, p=.023, η_p²=.104, 95% CI=[.28, 1.72]), with highest performance in the explicit label condition. However, it is important to note that a two-way ANOVA with Study and Condition as separate independent variables does not yield a significant interaction between Study and Condition (F(1,68)=1.68, p=.199, η_p²=.024), suggesting that the difference between conditions did not significantly change between Study 1 and Study 2.

Follow-up Analyses

To be consistent with Experiment 1, we followed up with additional analyses to determine whether the effect of broad versus irrelevant information may have been more prominent for older children with more language and emotion experience. In Experiment 2, within our sample’s children, age did not significantly predict child difference score from pre- to post-test (B=1.11, p=.218, 95% CI=[−0.69, 2.90]). Further, there was no significant interaction between children’s age and labeling condition for their pre-test to post-test change scores (B=−1.07, p=.619, 95% CI=[−5.40, 3.26]). Figures 4 depicts the relation between child age and difference score in the broad and irrelevant conditions for Experiment 2.

Figure 4.

Change in number of correct responses from pre-test to post-test plotted with child age for the irrelevant (black) and broad (orange) conditions in Experiment 2.

When examining gender differences, average change scores for girls (M=0.00, SD=1.18) and boys (M=−0.17, SD=1.75) did not significantly differ from one another (t(34)=0.34, p=.737, d=0.11, 95% CI=[−0.83, 1.17]). Further, there was no significant interaction between child gender and labeling condition for children’s change scores (F(1,32)=0.25, p=.621, η_p²=0.01). Similar to Experiment 1, we examined whether the conditions may have differentially affected the three emotions. Results revealed no significant interaction for emotion by condition (F(2, 102)=0.41, p=.664, η_p²=0.01). Thus, the effect of broad versus irrelevant information during the learning trials did not significantly change across the annoyed, disgusted, and nervous trials.

Interim Discussion

The results from Experiment 2 suggest that broad labels did not aid children in learning about emotions any more than irrelevant information, nor did they lead to children learning above chance levels. Considering the similar time frame and identical format of data collection as Experiment 1, any differences in the general pattern of results between the two experiments were unlikely to be due to cohort effects. Further, the highly similar values in the Irrelevant condition across Experiments 1 and 2 lend credibility to the findings and suggest that cohort effects appear unlikely to account for differences across the experiments. Thus, the benefit of labels for helping children learn the relation between emotional scenarios and facial expressions appear to be specific to explicit, non-broad emotion words.

General Discussion

Across two experiments, the present study investigated whether emotion language affected how 3-year-old children learned to associate facial expressions with emotional scenarios from typically later-learned emotion categories. Results of Experiment 1 revealed that children in the Explicit Label condition performed significantly better than children in the Irrelevant condition. Experiment 2 followed up by identifying that children in the Broad condition performed comparably to the Irrelevant condition. Additionally, there was no significant effect of child age or gender on performance, and neither age nor gender significantly interacted with the labeling condition within our sample. Overall, these results are important as they are some of the first to suggest that emotion words may be a helpful tool for children aggregating emotional information into meaningful categories.

As hypothesized, we observed that children learned the association between emotional scenarios and facial expressions better when hearing an explicit emotional label than when given irrelevant information. However, when children were provided with broad emotional information, their performance did not significantly differ from the irrelevant condition or from chance. It is important to note that in all conditions, children were always provided with consistent information across emotion types during the learning trials. That is, across both experiments all children hearing a “disgusted” learning trial scenario either heard that the character “feels disgusted”, “feels bad”, or “sits down”. Thus, if children only needed consistent linguistic information across learning trials to learn which face is associated with which type of scenario, they should have learned equally well across all three conditions. However, our findings suggest that consistent linguistic information was not enough in isolation to facilitate learning. Children performed significantly better when provided with explicit emotional labels than irrelevant information, and only the explicit label condition led to learning at above-chance rates from pre- to post-test. This suggests that emotion words, specifically, may play an important role in 3-year-olds’ learning about emotion categories, and that even broad emotion information (e.g., “she doesn’t feel good”) may not be as informative for young children.

Notably, broad emotion labels did not provide children with any advantage for learning emotional information over irrelevant information. Although these terms should have been sufficient to draw children’s attention to emotional cues, they did not help children learn relations between faces and emotional scenarios. One possibility is that the broad terms used in Experiment 2 are more general valence-based words that align with superordinate categories. Prior work has shown that superordinate categories are in general less effective at guiding categorization relative to more specific, basic-level categories (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976). Further, some work suggests that children learn words at the basic level before the superordinate level (Mervis & Crisafi, 1982) and that parents often use basic level words to help children learn superordinate categories (Callanan, 1985). In the present study, it could be argued that our terms of “annoyed”, “disgusted”, and “nervous” represent the basic level of organization, and therefore may simply be more effective at facilitating categorization relative to more broad, valence-based categories such as “good” and “bad”. If so, this would align with prior adult research indicating the benefit of basic-level emotion category terms (Doyle, Gendron, & Lindquist, 2021).

Another possible explanation for why the broad labels in Experiment 2 did not aid children’s learning about emotion categories is that children likely already have associations with broad terms like “feels bad” or “doesn’t feel good”. Because these terms are not new to children, they may have led children to attempt to focus more on familiar aspects of the stories and faces, thereby not helping the children to learn these new, more challenging emotion categories. That is, because children already had preexisting notions for what these broad emotion terms meant, and because these broad labels may already be linked to a large array of other information, this may have prevented them from learning the new association. This would align with previous research suggesting that infants’ ability to learn about abstract spatial relations is influenced by both the presence of novel words and infants’ existing relational vocabularies (Casasola & Bhagwat, 2007). Additionally, this may explain why previous research with infants has found that non-emotional labels (e.g., “toma”) can aid infant superordinate categorization of emotional stimuli (Ruba, Repacholi & Metlzoff, 2020), and why infants can map such nonsense labels to faces (Ruba, Harris, & Wilbourn, 2020). That is, if infants have less familiarity with specific emotion terms, any labels at all may be helpful for categorization. However, as children get older and more familiar with emotion terms, they make seek out these specific terms to aid their learning about emotion categories. Future research may wish to further examine how children interpret broad emotion terms such as “good” and “bad”, which may lend further insight into the added value gained by presenting children with specific emotion words instead of broad emotion terms.

Further, in thinking about the role of familiarity, it is important to consider two possible explanations for why children learned about these emotion categories in the explicit label condition. Although children learned the relations between faces and scenarios at significantly above-chance levels when presented with specific emotion words, it should be noted that these words were likely less familiar to children than the words presented in the other conditions (e.g., “feels bad” or “sits down”). Thus, children’s learning in the explicit label condition may have been due to label familiarity. If this is the case, we may expect that children would learn these relations equally well if presented with novel, nonsense labels (e.g., “she feels toma”), similar to what has been demonstrated with infants (Ruba, Repacholi, & Meltzoff, 2020). However, it is also likely that children have heard the specific emotion words “annoyed”, “disgusted”, and “nervous” before in their daily lives. Even if this previous exposure to these words was not sufficient for children to understand these categories, perhaps the present findings were enough to build on prior knowledge to help children establish which face belonged with which scenario. Future research may wish to build on the present findings by examining whether novel nonsense labels afford any advantage to 3-year-olds’ pairing of emotional faces with scenarios in order to disentangle these two possibilities.

These findings hold important implications for theoretical work as well as interventions. These results support theoretical work that highlights the importance of language for children’s developing understanding of emotions (e.g., Barrett, 2017; Lindquist, MacCormack, & Shablack, 2015). Specifically, by expanding beyond correlational research, this experimental design indicates that language (and more specifically emotion words) may facilitate children’s performance when learning about emotions. With our experimental manipulation we were able to demonstrate that, even in a short task with only 6 learning trials, children were more successful at learning about emotions when presented with specific emotion words than irrelevant information. These findings may also raise further questions for perspectives that emotions are unique, and that language is simply a communication device for labeling emotions. Such perspectives may hold that language is not crucial for developing an understanding of emotions (e.g., Ekman & Cordaro, 2011), or that perception of emotion is very similar cross-culturally (e.g., Ekman, 1999). However, our results suggest that language may in fact be important for emotion understanding and therefore that linguistic differences across cultures may lead to differences in emotion perception cross-culturally (Gendron, Roberson, van der Vyver, & Barrett, 2014). Ultimately, the present results align with shifting perspectives in other domains (e.g., number development) which emphasize the important role of language and labels in concept development (Mix, Sandhofer, & Baroody, 2005), but extend this framework to the abstract concepts of emotion.

Further, we propose that these findings may have important implications for interventions. Prior work has shown that emotion understanding interventions are promising (Sprung, Münch, Harris, Ebesutani, & Hofmann, 2015), but that there are many different approaches to such interventions. Our results suggest that perhaps targeting those around children (e.g., parents) and increasing their use of specific emotion words may help improve children’s developing understanding of emotion. Previous research has shown that increasing the use of emotion words among preschool teachers improves children’s emotion understanding (Grazzani & Ornaghi, 2011; Grazzani, Ornaghi, Agliati, & Brazzelli, 2016). However, the results of the present study suggest that even hearing emotion words from an unfamiliar experimenter benefitted children’s learning about emotions, suggesting that extending such interventions to others around the child, including parents, may be beneficial.

Interestingly, in both experiments we did not observe any significant interactions between children’s learning in the task and either the child’s gender or age among our sample of 3-year-olds. Some previous research has suggested that boys and girls differ in their understanding of emotion, typically with girls showing an advantage over boys (e.g., Denham et al., 2015; Dunn, Brown, Slomkowski, Tesla, & Youngblade, 1991; Ontai & Thompson, 2002). However, other studies have demonstrated no significant differences in children’s emotion understanding by gender (Dunn, Brown, & Beardsall, 1991; Grazzani, Ornaghi, Agliati, & Brazzelli, 2016). Our results align with those finding no gender differences, as we both observed no significant differences between boys and girls in their overall learning of emotions from pre-test to post-test, nor did we find any interaction between child gender and labeling condition in predicting learning about emotions. Thus, our results suggest that both boys and girls may see a similar benefit from hearing specific emotion words when learning about new emotions.

We also did not observe any significant differences in our sample between younger and older 3-year-olds in learning from pre-test to post-test, nor was there any interaction between children’s age and labels when predicting learning about emotions. However, although age may be an informative proxy for experience with emotional information, it is important to note that there is substantial variability in young children’s vocabulary (Fenson et al., 1994; Rowe, Raudenbush, & Goldin-Meadow, 2012) and vocabulary more than age may have a greater influence in learning from explicit labels, broad labels, and irrelevant information. Future research may wish to further investigate how children’s vocabulary contributes to learning in such paradigms. Additionally, our age range was relatively narrow (only 3-year-olds), and thus it is possible that age differences may be observed among a wider age group. These results also indicate that the effect of specific emotion labels on children’s learning about emotions held across both younger and older 3-year-olds, which aligns with previous work suggesting that labels are helpful for the simpler task of categorizing emotional facial expressions across a wide age range in early childhood (Russell & Widen, 2002). Thus, the present findings indicate that emotion labels may be helpful for children’s learning about new emotions regardless of gender or whether children are younger or older 3-year-olds. However, as these results as well as the gender findings were conducted post-hoc, it will be particularly important for future research to replicate these findings among a larger sample size.

Although these results provide important information about how emotion labels may help children to link later-learned emotion category events to facial expressions, there are still important questions that remain. Our study assessed children’s knowledge of annoyed, disgusted, and nervous scenarios, as these emotions are typically challenging for 3-year-olds to understand. However, it is unclear whether these results would extend to additional later-learned emotions (e.g., disappointed or embarrassed). It also remains to be determined how long these results would last, as our participants all took part in post-test trials immediately following learning. Future research may wish to examine whether hearing emotion labels afford a benefit to child emotion learning after a delay. Additionally, these results were specific to faces and scenarios, but emotions are fundamentally complex and involve many other components (e.g., vocal tone, body posture). Future research may wish to examine whether emotion words are similarly beneficial for helping children to link other components of emotions together. Another limitation includes the decision to include only images of women as stimuli. Although this was a deliberate decision, as women’s faces tend to be more familiar to young children, future research could examine whether these results generalize to images of men’s faces. Finally, the participants in the present study were all English-speaking participants from the United States. Thus, it will be important for future research to examine whether these results generalize to participants who speak other languages, which may draw different emotion category boundaries or have diverse ways of discussing broad versus specific emotional content. Further, it will be important to determine how these results may or may not generalize to other cultures, as previous research has shown that the same emotion category can be associated with different facial configurations across cultures (Elfenbein, Beaupré, Lévesque, & Hess, 2007), and interpretations of the same emotional faces can also vary (Gendron, Crivelli, & Barrett, 2018). Thus, identifying how emotion labels affect emotion understanding across languages and cultures, as well as replicating the present results among larger sample sizes, will be key for future research. To our knowledge, it is unclear whether this would influence only emotion category boundaries, or whether the underlying mechanism (i.e., the role of emotion labels) may change among a more diverse array of languages and cultures.

To conclude, across two experiments our results revealed that explicit emotion labels provided an advantage over irrelevant information for children learning about emotions, but that broad emotional information did not afford an advantage when compared to irrelevant information. That is, children’s performance on a task identifying the face associated with an emotional scenario significantly increased from pre-test to post-test when they were given explicit emotion labels, but not broad emotion information or irrelevant information. These results suggest a potentially important connection between emotion labels and children’s understanding of emotion categories, and also hold implications for theories of emotional development.