Toddler Word Learning is Robust to Changes in Emotional Context

Abstract

Word learning is a crucial aspect of early social and cognitive development, and previous research indicates that children’s word learning is influenced by the context in which the word is spoken. However, the role of emotions as contextual cues to word learning remains less clear. The present study investigated word learning among 2.5-year-old children in angry, happy, sad, and variable emotional contexts. Fifty-six children (30 female; Mean age=2.49 years) participated in a novel noun generalization task in which children observed an experimenter labeling objects in either a consistently angry, consistently happy, consistently sad, or variable (one exemplar per emotion) context. Children were then asked to identify the label-object association. Results revealed that children’s performance was above chance levels for all four conditions (all t’s>3.68, all p’s<.01), but performance did not significantly differ by condition (F(3,52)=0.51, p=.677). These results provide valuable information regarding potential boundaries for when contextual information may versus may not influence children’s word learning.

Emotion, including emotional information presented in speech, is a rich cue that helps infants and children to make sense of the world around them and influences their behavior (e.g., Repacholi & Gopnik, 1997; Repacholi, Meltzoff, Rowe, & Toub, 2014; Walle, Reschke, Camras, & Campos, 2017). However, the role of emotion in speech as a cue to the meaning of words remains understudied. Emotional information is helpful for disambiguating word meaning (e.g., Berman, Graham, Callaway, & Chamberes, 2013), but it may also provide a general contextual cue for word learning. That is, the consistency and variability with which emotions are used to label certain words may provide children with additional cues that aid word learning. From a young age, infants show a preference for happy vocal tones (Corbeil et al., 2013), but sadness in speech is typically slow (Hou et al., 2015) and less interesting to infants (Walker-Andrews & Grolnick, 1983). Thus, it is possible that different emotional information conveyed in speech provides crucial contextual information (i.e., changes to the learning environment) which influences children’s word learning. To address this, the present study investigated whether the emotional context in which words are presented impacts word learning.

Why may emotional context be important for word learning? Broadly speaking, contextual information is important for children’s learning because it provides cues that facilitate category formation, and ultimately facilitate word learning. Context is a broad term, generally referring to the circumstances in which word labeling occurs that are not necessary for understanding the label itself. For example, if a child hears the word “bottle” used at home, at the park, and in the car, these changes in location represent different contexts for word learning. Similarly, if the child’s mother, father, and nanny all label the word “bottle”, the child has heard this label in a variety of contexts (in this case, the people labeling the object). Thus, when thinking about word learning in children’s daily lives, the context refers to all parts of the situation in which the learning event took place. Prior research indicates that children are more likely to aggregate information into one category when there is greater similarity in their memories of the category instances, even when that information is not diagnostic of the learned category (Sandhofer & Schonberg, 2020). In this way, aspects of the broader context (e.g., location, person speaking) can influence word learning. Redundant cues that are correlated with the learned information (i.e., contextual cues) can, therefore, facilitate category learning. Even seemingly “irrelevant” contextual information can help children to form categories because it can highlight relevant similarities among category exemplars (e.g., Dueker & Needham, 2005). That is, seeing flamingos in the zoo but peacocks on a farm may help children to learn the difference between these two categories of birds, even though the context (zoo vs farm) is not necessary for understanding the category differences. Similarly, 4- to 5-year-olds can use context (e.g., screen color)-predictor contingencies to learn patterns (Sloutsky & Fisher, 2008), and older infants can learn dimension-context contingencies with sufficient repetition (Sloutsky & Robinson, 2013). Thus, visual pattern learning and categorization studies suggest that even simple contextual manipulations can provide children with cues to facilitate learning.

Variability in contextual information can also be beneficial for children’s learning. This is because children often initially learn words in a manner that is bound by context (e.g., Hoff, 2013). For example, Perry and colleagues (2014) found that 16-month-olds showed better understanding for naming of non-solids when seated in a high-chair, the typical context in which they experience non-solids (namely, foods). Because children typically interact with non-solids while in a high-chair, they could learn to name non-solids in this context but struggled to do so in non-high-chair contexts. That is, infants’ ability to learn words for non-solids appears to be contextually bound. However, learning words in a variety of contexts allows children to generalize words to new contexts in which they are encountered. Supporting this concept, when 3- to 4-year-old children learned novel word-object pairings with various cloth backgrounds, children learned the word substantially better when the words were learned on a variety of cloth contexts relative to learning in one context and being tested in another (Vlach & Sandhofer, 2011). This mirrors adult literature, which indicates that learning information in multiple different rooms can aid recall in a novel room (Smith et al., 1978). Taken together, these results suggest that learning information in a variety of contexts can support learning more broadly.

Given that many of the previous studies have focused on the impact of largely artificial cues on word learning (e.g., screen color, cloth background), it is important to identify what types of naturalistic contexts may (or may not) operate in a similar fashion. For example, the specific location of an object can serve as a redundant, correlated cue, providing children with additional memory supports to facilitate learning the word that refers to that object (Samuelson et al., 2011). Similarly, speaker variability can influence infants’ word learning (Rost & McMurray, 2009; 2010). Thus, multiple types of context can provide helpful cues for children’s word learning. However, it is important to note that previous work involving the role of contextual cues in children’s learning have often involved different social contexts. For example, some studies have tasked children with observing stimuli on a screen (e.g., Rost & McMurray, 2009, 2010; Sloutsky & Fisher, 2008; Sloutsky & Robinson, 2013), while others have involved an experimenter directly communicating with a child (e.g., Perry et al., 2014, Vlach & Sandhofer, 2011). Given the similarities and differences in social contexts for these studies, examining how changes in social cues may impact children’s learning remains an important question.

Much remains unknown regarding how more complex social contexts affect word learning. In particular, the role of emotion as a potential contextual cue to word learning warrants further examination, as typically developing children hear new words through speech, which also carries emotional tones. A variety of emotions may also become more prevalent in children’s environments in the toddler years (Campos et al., 2000) a time when many new words are being learned. Toddlers also understand some emotion labels and can map faces to emotional situations (Denham, 1986), indicating that children are sensitive to emotional information at this age. Thus, investigating emotion as a potential context for word learning offers an interesting and important opportunity to examine whether social contexts may influence word learning in a similar manner to what has been demonstrated for other contexts (e.g., background cloth, room in which information is presented).

Emotions involve many distinct components (e.g., facial expression, body posture, tone of voice), but there is reason to expect that in particular emotional cues presented in speech should affect word learning. From birth, infant-directed speech is common in young children’s environments, and this speech is associated with smiling, happy facial expressions and upbeat tones of voice (Kim & Johnson, 2013). Infants prefer infant-directed speech (Fernald, 1985), and greater exposure to infant-directed speech is related to larger expressive vocabularies in children (Weisleder & Fernald, 2013). However, it has been proposed that as children get older and begin to crawl and walk, exposure to new expressions, such as anger, from caregivers may increase (Campos et al., 2000). According to this perspective, mobile children may then begin to hear words spoken in a variety of tones of voice. Some words may be spoken most often in only one emotional tone of voice (e.g., “broken” in a sad tone or “ice cream” in a happy tone). Other words may be spoken in highly variable emotional tones (e.g., “Let’s put your shoe on so we can go to the playground!”, “Do NOT throw your shoe at me!”, “Your favorite shoe is lost”). Thus, it is possible that hearing this emotional information in speech may serve as contextual cues that impact children’s ability to learn words. That is, this emotional information may not be diagnostic of the word that is learned, but may provide more general contextual information to help facilitate word learning.

In particular, the role of angry, happy, and sad emotional contexts are of interest, as previous research with 2- to 5-year-olds has shown that these are the first emotion categories learned by children (Widen & Russell, 2003). Happy facial expressions are commonly displayed to children, and even infants have strong preferences for happy faces (Kim & Johnson, 2013) and voices (Corbeil et al., 2013). Angry facial and vocal expressions may be attention-grabbing and become more common as children become mobile (Campos et al., 2000), but also may be aversive. Sad vocal tones are typically slow (Hou et al., 2015) and therefore sad expressions may be less interesting (Soken & Pick, 1999; Walker-Andrews & Grolnick, 1983). Given the differences in these social signals, it is important to consider the broader contextual information that these emotional facial and vocal cues may provide to young word learners.

There is evidence that infants and young children attend to emotional information during word learning. From as early as 7.5 months, infants are more sensitive to phonological differences in words when they hear them in a variety of emotional tones than when emotional variability is low (Singh, 2008). Additionally, 4- and 5-year-old children can use emotional tones to identify the referent of novel words when observing stimuli on a screen (Berman et al., 2013). For example, when hearing a label in a sad tone of voice, children were more likely to map that label to the image of a broken object than a non-broken object. Thus, children appear to notice emotional cues in language, and recent theoretical work suggests that learning emotion and language may be tightly linked (Barrett, 2017). However, what remains unclear is whether hearing words labeled using multiple categories of emotional tones of voice (e.g., in an angry and happy and sad tone) differentially impacts children’s ability to learn those words. That is, we expect that in general children can learn from hearing words in emotional tones, but are children more or less likely to learn words if they hear the words in particular emotional tones, or across a variety of emotional tones?

It is important to note that emotions are complex, involving a multitude of features including tones of voice, facial expressions, body posture, emotion words, as well as situational and relational cues. Yet, whether emotions may serve as contexts for word learning in similar manners to other contextual cues remains understudied. As a first step toward examining this question, we operationalized emotional context as the emotional tone of voice and facial expression heard and seen by the child. These two features were selected because novel words are heard through voices, making vocal tone highly relevant, and a large body of prior research with infants and young children examines their understanding of emotional facial expressions (e.g., Denham, 1986; Ruba & Repacholi, 2020). Thus, for the purposes of the present study, emotional context refers specifically to emotional tone of voice and facial expression presented across auditory and visual modalities.

The Present Study

Do emotional cues serve as contexts for word learning? Does any particular emotional context provide a word learning advantage over other emotions? To answer these questions, we assessed the performance of 2.5-year-old children in a version of the novel noun generalization task. Children watched as novel objects were labeled in either an angry, happy, sad, or variable (1 exemplar labeled in each emotional tone) context. We investigated whether performance differed by emotional context. As children are frequently exposed to happy tones through infant-directed speech (Kim & Johnson, 2013), and happy tones are typically used with children during word-learning studies to maintain engagement and interest in the task, the happy condition served as our baseline condition (i.e., this represents how children typically hear words spoken in these sorts of tasks as well as during everyday interactions through infant-directed speech). Although adults perceive neutral information as neither positive nor negative, children may not perceive neutral information as truly neutral, but instead as slightly negative. For example, a previous study with 6-to 17-year-olds found that neutral stimuli were consistently rated as negative across this age group (Tottenham, Phuong, Flannery, Gabard-Durnam, & Goff, 2013), and a study with infants found that they may even potentially be aversive in the still face paradigm (Mesman, van Ijzendoorn, & Bakermans-Kranenburg, 2009; but see Repacholi, Meltzoff, Toub, & Ruba, 2016 and Ruba & Repacholi, 2020). Thus, although the age of participants in the present study was between these two age groups, a “neutral” condition was not included as a baseline condition in the present study due to the possibility that a true “neutral” labeling of the stimuli may not be perceived as such by the toddler participants.

In the present study, we did not want children to interact directly with an angry adult in case this might cause some children distress. Thus, we implemented a design where children observed two adults interacting with one another and labeling objects while certain emotions were expressed. This is similar to the overhearing procedure used by Repacholi and Meltzoff (2007) in their emotional eavesdropping study. In our design, the emotional experimenter never interacted directly with the child, but rather a neutral experimenter asked the children for a response on test trials. We then examined whether emotional contextual variation influenced children’s performance. This paradigm is useful for two main reasons: 1) It allows for a more robust assessment of children’s learning, as they must overhear the object labeled by one experimenter but respond appropriately when later asked to identify that label by a different experimenter in a different tone (i.e., context) from learning trials, and 2) Children do not hear the labels directly, and therefore should not feel as if they are the target of any particular emotional display. Previous studies using overhearing paradigms with 2-year-old children have demonstrated that children can learn novel words in such social contexts (e.g., Akhtar, 2005).

Because variable contexts have been shown to aid word learning in previous research (Perry et al., 2010; Vlach & Sandhofer, 2011), we hypothesized that children would perform better in the variable context condition (i.e., hearing the same word in multiple different emotional tones). That is, perhaps children would have improved performance on variable trials because hearing the same word in a variety of emotional tones may make the label more generalizable to the child (i.e., less context-dependent), thereby allowing for more robust learning and extension to a new context (i.e., a new speaker and tone in the test trial). We also hypothesized that children would perform well in the happy condition due to infant preference for happy vocal tones (Corbeil et al., 2013) and previous research suggesting that infant-directed speech (which tends to be happy in tone) benefits word segmentation by as early as 9 months (Schreiner & Mani, 2017). Given that the negative valence of the angry condition may have been aversive and thereby hindered children’s performance, but the high arousal may have aided children’s attention in the task, we did not have a specific hypothesis for children’s performance in the angry condition. However, we expected that children may perform differently in the sad condition, as we speculated that sadness may be less engaging for children (Soken & Pick, 1999) and therefore may not aid word learning.

Method

Participants

Fifty-six children (30 female) ranging from 2.02 years to 2.99 years of age (M_age=2.49 years, SD_age=0.30 years) participated in this study. Five additional children were excluded due to failure to complete all eight trials (N=4) or experimenter error (N=1). A power analysis assessing the effect for a cross-condition effect including four conditions was conducted using G*Power 3.1 (Faul et al., 2009) based on the effect size (η²_p=.336) of changing context reported in a word learning study (Vlach & Sandhofer, 2011). This analysis revealed that the total number of participants necessary to achieve power of 0.8 was 28. However, the previous study on which we based our power analysis had a notably high effect size. Thus, to ensure adequate power in our study we doubled this to a larger sample size of 56. Participants were recruited from local preschools and from lists of birth records. Race/ethnicity and SES information was not collected from participants but were likely representative of Los Angeles County in which the data were collected. Race/ethnicity of the experimenters were as follows: White (N=7), Multiracial (N=2), Asian (N=2), Latina (N=1), Middle Eastern (N=1), Not Provided (N=1). This study was approved by the university Institutional Review Board (#10-001578). Parents provided written informed consent prior to any data collection.

Materials

Novel objects were created for the present study to control for children’s prior experience with the objects and labels. That is, because different children have varying degrees of familiarity with objects and labels in the real world, we created novel objects and novel labels so that all children would have no prior experience with any of them. Objects were created by gluing and painting combinations of supplies from craft stores and hardware stores. All exemplars from the same category (i.e., ascribed the same label) had the same shape but differed in color. Categories of novel objects were randomly assigned a novel label (e.g., “wug”). All children saw the same eight novel object categories and heard the same eight novel labels, but the label-object pairings and order of presentation were randomly assigned.

Procedure

Prior to the experiment, children were randomly assigned to one of four emotional conditions: Angry, Happy, Sad, or Variable. Children were then introduced to two experimenters, one of whom was the emotional experimenter who provided the object labels, and the other was the neutral experimenter who interacted directly with the child. Using these two experimenters ensured that any emotions conveyed were never directed at the child. That is, when the emotional experimenter was speaking in an angry tone of voice, the child was simply overhearing the angry emotional tone which was directed at the neutral experimenter. Overhearing paradigms have been previously used with young children (e.g., Akhtar, 2005) as have eavesdropping paradigms with emotional content (Repacholi & Meltzoff, 2007). In addition, having the emotional experimenter label the object before presenting it to the neutral experimenter, who then presented it to the child introduced substantial temporal lag between when children initially heard the label and when they physically held the object. Ultimately, this gap in time between hearing the label and holding the object allowed for a more robust assessment of child word learning. The experiment included eight trials, each trial with a learning, distractor, and test phase.

Learning Phase.

For each trial, the emotional experimenter pulled novel objects out of a box and showed them to the neutral experimenter. The emotional experimenter presented the neutral experimenter with three separate exemplars from the novel object category, one at a time. The emotional experimenter showed the novel object to the neutral experimenter, labeled it two times (e.g., “See the wug? It’s a wug.”), then handed it to the neutral experimenter. The neutral experimenter took the object, looked at it, then handed it to the child, telling them that they could play with it. Children could play with each object for 10s before the neutral experimenter retrieved the object. This procedure was repeated for the second and third exemplar objects belonging to the same category (e.g., two wugs of different colors). After the child had seen all three instances of a given category, the emotional experimenter presented a distractor object. This was a novel object that did not belong to the learned category. The emotional experimenter did not provide a label for this object and presented it in a neutral tone of voice (e.g., “Look at this. Do you see it?”). The neutral experimenter then took this object and allowed the child to play with it in the same manner as the exemplar objects. The distractor object was introduced for two reasons: First, it gave children a distraction from the exemplar objects, and therefore an opportunity to forget the category label-object pairing. That is, if children advanced immediately from the learning phase to test trials, they would simply need to select the object that they had seen only seconds before. By incorporating a distractor object, children did not immediately advance to the test trials after seeing the third object exemplar, providing time for forgetting and something else to focus their attention on between learning and test. Second, it gave children an opportunity to hear the emotional experimenter in a neutral tone of voice. Although children may not have perceived this to be truly “neutral”, this provided a contrast to the emotional tones presented during exemplar object labeling (e.g., to demonstrate that she isn’t just always a “sad person”). This contrast between emotional tones during learning trials and neutral tones during distractor trials may have also served to further emphasize the presence of emotion during labeling. That is, the emotional information presented during learning trials may have become more salient by virtue of seeing the emotional experimenter occasionally speaking in a less emotional tone.

Test Phase.

Immediately following the distractor object for a given trial, children participated in the test phase. This procedure wherein the relevant test trial shortly follows learning is common in work with young children (e.g., Vlach & Sandhofer, 2011), as memory constraints limit young children’s ability to recall novel words learned over longer intervals.

Here, the neutral experimenter simultaneously presented four objects to the child (e.g., “You can play with these”). Children could play with these objects for 30s. Of these objects, one belonged to the learned category (the target object). That is, it matched the category shape, but was of a new color. The other three objects were all entirely novel shapes, but each matched the color of one of the exemplar objects that had been presented previously. After the children played and familiarized themselves with the objects, the experimenter took the objects back, lined the four objects up, then presented them to the child again and asked for the category object (e.g., “Can you hand me the wug?”) in a neutral tone of voice. Asking the test question in a neutral tone ensured that for children in all conditions, the test trials were conducted in a different emotional context from the learning trials, even if children did not perceive the emotion to be truly “neutral”. Thus, to answer the question correctly, children needed to recognize which of the now-familiar objects matched the previously heard label in a new context (i.e., new speaker and new emotional tone) and provide that object to the experimenter. The emotional experimenter looked away and the neutral experimenter looked straight ahead with the objects out of her reach as the child made their selection to reduce the risk of either experimenter biasing the child’s choice. When the child handed the neutral experimenter one of the objects, this experimenter recorded the child’s choice on a data entry form, and the next trial immediately began. This process continued for eight total trials. Examples of the objects presented to children are presented in Figure 1.

Figure 1.

Examples of stimuli used in learning, distractor, and test trials.

The procedure for the experiment lasted about 15 minutes and was identical across conditions except for the vocal tone and facial expression of the emotional experimenter during labeling. In the angry condition, the emotional experimenter labeled all novel objects in a terse, clipped tone of voice accompanied by a compressed mouth and furrowed eyebrows. In the happy condition, all novel objects were labeled in an upbeat tone of voice with a smile, akin to the tone commonly used in infant-directed speech. In the sad condition, all novel objects were labeled with a downcast, whiny tone of voice presented at a slow pace with a frown and raised inner eyebrows. In the variable condition, each of the three exemplars within a category was labeled once with each emotion (angry, happy, and sad), providing the children with a variety of emotions for each category. The tones of voice and facial expressions used by the emotional experimenter in the learning trials was based off those conveyed in the puppet vignettes used in the Affective Knowledge Test (Denham, 1986). Each experimental session was recorded, allowing for validation that the emotional experimenter conveyed facial and vocal expressions in a manner consistent with the intended emotional categories.

Emotion Validation

To ensure that the experimenters presented the intended emotions on the correct trials for participants, a Qualtrics survey was created. Five adults (all female) who were unfamiliar with the study and hypotheses viewed brief clips from the experiment in which the emotional experimenter labeled each novel object. The adults were then asked to identify whether the emotion being conveyed by the experimenter in each labeling instance was anger, happiness, or sadness. On average, the adult raters correctly identified the intended emotion 98.8% of the time (Separated by condition: Anger=99.0%, Happiness=99.8%, Sadness=99.8%, Variable=97.6%). This forced-choice selection of emotions has been commonly used to validate emotional stimuli in the past (e.g., Ogren & Johnson, 2020). Based on these results, we concluded that the experimenters conveyed the correct emotions for each condition and were unlikely to present emotional information from an incorrect condition (e.g., experimenters were unlikely to convey anger when the child was in the sad condition) throughout the experiment. However, it is important to note that the forced-choice paradigm may have given the adults hints as to how the emotions were intended to be interpreted (i.e., either as happy, sad or angry), and this may have influenced participant responses (Barrett, Adolphs, Marsella, Martinex, & Pollak, 2019).

Results

To determine whether the emotional context influenced children’s ability to learn novel label-object pairings, we compared children’s performance across the 4 emotional conditions. Average number of correct responses per condition out of 8 possible were as follows: Angry=4.50 (SD=2.53); Happy=4.57 (SD=2.03); Sad=4.67 (SD=2.53); Variable=5.46 (SD=1.90) (see Figure 2). A one-way Analysis of Variance revealed that the number of correct responses did not significantly differ by condition (F(3,52)=0.51, p=.677, η²_p=.029). An additional ANCOVA analysis was conducted with age included as a covariate. These results revealed a very similar pattern of findings, with the number of correct responses not significantly differing by condition when child age was accounted for (F(3,51)=0.54, p=.656, η²_p=.031). Thus, the effect of condition was not significantly impacted by whether children were younger or older 2-year-olds, and thus age was not included as a variable in subsequent analyses.

Figure 2.

Average number of correct responses on the novel noun generalization task separated by emotional condition. Error bars indicate standard error.

Further, a Bayesian analysis investigating the effect of condition on number of correct responses was conducted using JASP (JASP Team, 2020). Bayesian analyses can be useful to complement standard null hypothesis testing, as these analyses can allow for comparisons of relative evidence for the null and alternative hypotheses (Quintana & Williams, 2018). That is, while standard null hypothesis testing can only tell you if the alternative hypothesis of an effect is significant or not, Bayesian analyses can provide valuable information regarding whether there is evidence for the alternative hypothesis, the null hypothesis, or neither. Thus, following up with a Bayesian analysis allows for further insight into whether the previous null result was driven by evidence for the null hypothesis, or lack of sufficient evidence in either direction. Results revealed BF₁₀=0.16, suggesting substantial evidence for the null hypothesis (Jeffreys, 1971). Thus, evidence from the present study suggests that child performance did not differ across emotional conditions.

Given the lack of significant difference among the four emotional conditions, we performed follow-up analyses to identify whether performance in each condition was significantly above chance. Chance performance was 2 correct responses (i.e., 1 in 4 chance of selecting correctly across 8 trials). We corrected for multiple comparisons using a Bonferroni correction, yielding a new alpha level of .0125. Results revealed that children’s performance was above chance for each of the four conditions (Angry: t(13)=3.69, p=.003, d=0.986; Happy: t(13)=4.75, p<.001, d=1.268; Sad: t(14)=4.09, p=.001, d=1.056; Variable: t(12)=6.58, p<.001, d=1.824), and Bayesian analyses revealed that evidence for children’s above-chance performance was strong in the angry condition (BF₁₀=16.77), very strong in the happy (BF₁₀=89.54) and sad conditions (BF₁₀=35.49) and decisive in the variable condition (BF₁₀=929.45).

Discussion

The present study assessed whether 2.5-year-olds’ word learning was differentially impacted by the emotional context in which the word was presented. Results revealed that there were no significant differences in word learning across the angry, happy, sad, and variable conditions. Further, a Bayesian analysis revealed substantial evidence for the null hypothesis of no difference among the conditions. Additionally, results revealed that children’s performance was significantly above what would be expected by chance in each of the four conditions. These results suggest that under tested circumstances more complex social information, such as angry, happy, and sad emotions, may differ from previously tested contexts in how it relates to children’s word learning. This provides potentially valuable boundary conditions for what types of contexts may and may not influence word learning.

Contrary to our hypothesis, we did not observe an advantage for word-object recognition in the emotionally variable condition. This result runs counter to previous findings that variability in other contextual domains can benefit young children’s word learning (e.g., Perry et al., 2010; Vlach & Sandhofer, 2011). Thus, it appears that an advantage for context variability did not extend to emotional contexts in our study. One reason why this may be the case is that emotional contexts are much more abstract (Barrett, 2017) than previously tested contexts. That is, not all instances of one emotion category look alike, as the same emotion includes a range of facial expressions, vocal tones, and body postures. In the present study, experimenters attempted to maintain as much consistency as possible for each emotion, and their emotional displays were consistently recognized as the intended emotion by adult raters. Nonetheless, some variability within each emotion category was inevitable, and previous research suggests that children incorporate high within-category variability into a single concept less readily than adults (Rhodes & Brickman, 2010). Thus, in the context of the present overhearing design, children may have had trouble interpreting discrete emotion categories in the experimenter’s tone (cf., Morton & Trehub, 2003) thus lending no advantage to the emotionally variable context. This may partially explain why children’s performance was comparable across all conditions, although this seems unlikely given that infants begin to discriminate emotional faces and voices within the first year of life (Flom & Bahrick, 2007). It is also possible that emotional contexts may impact children’s word learning at older ages, as emotion understanding is still undergoing substantial development in our tested age group (Denham, 1986) and continues to develop into middle childhood (Pons, Lawson, Harris, & De Rosnay, 2003).

Additionally, we speculate that the broader context of the experimental set-up may have influenced children’s performance. For the present study, children were introduced to two unfamiliar experimenters, one of whom they then observed labeling novel objects. However, without any prior exposure to this experimenter, children may have found her emotional displays more challenging to categorize and less socially relevant than the displays of someone with whom they are more familiar. Even very young infants can match emotional information (faces and voices) when the stimulus is their own mother, but they cannot match this same information when it is displayed by an unfamiliar woman (Kahana-Kalman & Walker-Andrews, 2001). Thus, the lack of familiarity with both experimenters may have played a role in children’s consistent performance across emotional conditions, as they were less familiar with this particular individual’s pattern of emotional responses. However, this also seems unlikely given previous research indicating that infants and young children can discriminate and label various emotional faces of unfamiliar adults (Ruba & Repacholi, 2020, Widen, 2013). Nonetheless, future research may wish to investigate whether emotional information presented by a familiar individual (e.g., a parent) would differentially affect children’s ability to specifically learn words from various emotional contexts, as compared to previous tasks of social referencing and emotion perception.

Further, the experimental set-up (although common for word-learning studies, and beneficial for clearly isolating emotional contexts) may have seemed unusual to the children. That is, although children completed the tasks, they may have found it mildly confusing or unfamiliar for the emotional experimenter to repeatedly label the objects in certain emotional tones without a broader reason for doing so, potentially deemphasizing the personal significance of the word learning opportunity to the child. Emotions are complex and involve several features beyond tone of voice and facial expression. Typically, emotions involve some relevant relational communication. That is, the emotional cues are something of personal significance related to the individual’s goals in the environment (e.g., being sad that something has broken or happy after receiving a gift), and prior research has shown that experiencing emotions can influence categorization (Niedenthal, Halberstadt, & Innes-Ker, 1999). In the present study, children were observing an experimenter expressing anger, happiness, and sadness, but without any apparent relational significance for these expressions to the experimenter or to the child. Future research may wish to embed controlled emotional contexts into more socially meaningful situations (e.g., shared book reading) to assess whether contextual information impacts word learning in more realistic, commonly encountered circumstances where the relational significance of the emotion may be clearer to the child.

One particularly interesting finding was that children in the sad condition performed comparably to the other three conditions. That is, hearing words consistently labeled in a sad tone of voice did not provide any disadvantage for children’s word learning relative to other emotional conditions. Given that sadness is a low-arousal negative emotion, we questioned whether children would attend to sadness as much as the higher-arousal happiness and anger. Additionally, we had hypothesized that sadness may be less interesting for young children than anger or happiness (Soken & Pick, 1999). However, our results suggest that children’s word learning in the present study was not negatively impacted by expressions of sadness. This may be due to lack of familiarity with the experimenter, as described above. In contrast, it may be that children simply are not as impacted by emotional cues as previously tested cues in word-object recognition tasks.

The present study is a first step in asking how emotional context influences word learning and represents an important starting point for beginning to understand whether these more complex social cues operate similarly or differently from other contextual cues (e.g., background color). However, there are still several important questions within this area of research that remain to be examined. First, in our study all children were exposed to three learning trials for each novel word, but it is possible that some emotions may be more salient than others and thus may require fewer trials for learning. Because we kept the number of trials consistent for each condition, we could not examine whether learning is faster for some emotions than others. Future research may wish to examine this possibility. Second, although children may not perceive neutral facial expressions and tones of voice to be truly neutral in the same way that adults would, it may be interesting to examine whether a neutral context would result in different word learning patterns relative to the more emotional contexts examined in the present study. Third, although prior research has shown that children tend to learn angry, happy, and sad emotion categories first (Russell & Widen, 2003), other research indicates that even infants are sensitive to many discrete emotions (Walle, Reschke, Camras, & Campos, 2017). Thus, future research examining how word learning is affected by emotions that tend to direct attention toward a particular referent, such as fear and disgust (Knothe & Walle, 2018), would be highly valuable.

Additionally, in our study we were interested in whether variability in the specific categories of emotional tones used to label objects would serve as contextual cues to impact children’s learning. However, this does not address the idea of variability within a particular emotion category. That is, would children learn words differently if they heard the same object labeled in very similar sad tones, or across more variable instances of sad tones? Future research may wish to examine the acoustic properties of emotional tones to identify whether consistency vs variability for the tones conveyed for a certain emotion would influence children’s word learning.

Another important finding from the current experiment was that the 2.5-year-old participants performed above chance in all 4 conditions. This is worthy of note given that this was an overhearing study, in which the experimenter labeling the object never talked directly to the child during labeling, and the child had to remember the label over the course of the observed interaction and their interaction with the neutral experimenter. Previous research has demonstrated that 2-year-old children can learn in overhearing contexts (Akhtar, 2005; Martinez-Sussmann et al., 2011). However, in some cases, such as when there is insufficient engagement in the task, children do not learn in overhearing contexts (O’Doherty et al., 2011). Given that performance in each of our conditions was substantially above chance levels, we theorize that perhaps multiple different emotional word learning contexts may be sufficient to promote word learning in toddlers. This may be supported by research indicating that prosody facilitates memory recall in a speaker’s first language (Schmidt, Perez, Cilibrasi, & Tsimpli, 2020). By providing emphatic facial expressions and tones of voice, children may have been motivated to pay attention to and remember the naming events, thereby learning from them in all conditions. How children learn novel words in more emotional versus less emotional (e.g., neutral) contexts remains an open question for future research.

Conclusion

In conclusion, the present study adds to our knowledge of children’s word-object recognition by identifying that children learn novel words equally well in angry, happy, sad, and variable overhearing contexts. These findings provide important boundary conditions for when and how context influences word learning. That is, under tested circumstances variability in social contexts, such as emotion, may not influence word learning in the same manner as previously tested contexts. Ultimately, these results extend our knowledge of word learning in early childhood and indicate that children’s word learning may be robust to changes across happy, sad, and angry emotional contexts.

Highlights.

• The role of emotions as contextual cues to word learning is unclear

• Using a novel noun generalization task to test 2.5-year-olds’ word learning, we found no significant difference between four emotional context conditions

• Suggests that the tested emotional contexts may differ from previously tested contextual manipulations with regards to their role in word learning

Data Availability:

Data will be made available upon request