Emotion Words in Early Childhood: A Language Transcript Analysis

Abstract

Children learn the abstract, challenging categories of emotions from young ages, and it has recently been suggested that language (and more specifically emotion words) may aid this learning. To examine the language that young children hear and produce as they’re learning emotion categories, the present study examined nearly 2,000 transcripts from 179 children ranging from 15- to 47-months from the Child Language Data Exchange System (CHILDES). Results provide key descriptive, developmental, and predictive information regarding child emotion language production, including the finding that child emotion word production was predicted by mothers’ emotion word production (β=.21, p<.001), but not by child or mother language complexity (β=.01, p=.690; β=.00, p=.872). Frequency of specific emotion words are presented, as are developmental trends in early emotion language production and input. These results improve the understanding of children’s daily emotional language environments and may inform theories of emotional development.

Understanding emotions is an essential aspect of healthy social cognitive development (Hesse & Cicchetti, 1982; Olson, Astington, & Harris, 1988), and knowledge of emotion words plays an important role in developing knowledge of these crucial social categories. Not only are children with a larger emotion vocabulary more capable of talking about their own and others’ emotions, but a recent theory suggests that emotion words help to foster children’s learning of emotion categories (Barrett, 2017). Considering the importance of emotion words for children’s developing emotional competence and broader social cognition, it is critical to better understand what emotion words children hear in their natural environment as well as what they produce. To address this, the present study uses naturalistic corpus data to characterize the emotion words in children’s early language environments.

Emotion words are challenging for children to learn because emotion categories are abstract. That is, although we treat emotions categorically, they are truly continuous gradients. There is not one distinct facial configuration that maps on to any given emotion, and the same emotion category involves diverse displays. In fact, emotion categories are expressed with substantial variability (Barrett, Adolphs, Marsella, Martinez, & Pollak, 2019) and involve facial movements, tone of voice, body posture, and situation-specific information. Thus, children are faced with a challenging problem of integrating information across multiple domains and identifying how and where to draw category boundaries between various emotions (Hoemann, Xu, & Barrett, 2019). It has been proposed that learning emotion categories is similar to learning other types of categories or concepts (e.g., Rakison & Oakes, 2003), in that emotion categories are constructed from simpler precursors via learning and experience. In particular, children’s emotion category learning may be aided by language (Ruba, Meltzoff, & Repacholi, 2020), because it has been previously shown to influence category development for other abstract categories such as relations (Loewenstein & Gentner, 2005). This link between language and cognition is thought to exist because words invite children to form categories (Brown, 1958), and may highlight commonalities among category members.

Further, it has been suggested that learning emotion words may be particularly helpful, as they provide a “powerful tool” for understanding emotions (Kopp, 1989). This may be because emotion words allow children to link a variety of emotional displays and events together via one label (Lindquist & Gendron, 2013). For example the label “afraid” may be used to describe a child who is scared of the dark, a teenager demonstrating stage fright, and an adult running away from a dog. Thus, the label may help children to join these diverse experiences into one category of “afraid”. In this way, emotion words may help children to organize and develop emotion categories (Shablack & Lindquist, 2019). In other domains, the specific input that infants are exposed to is important for directing attention and learning (Hurley & Oakes, 2015; Smith, Jayaraman, Clerkin, & Yu, 2018). Thus, if emotion category learning operates in a similar manner to learning other types of categories, it is pivotal to identify the emotion words that children hear as input. Ultimately, identifying this input will lend valuable insights into how the categories are learned in the first place. Thus, language (and more specifically emotion words) may help children to attend to and learn the abstract categories of emotion.

Although learning emotion categories appears challenging, children typically begin to talk about emotions from a young age. By around 18 months of age, children begin to produce emotion words such as “happy” (Ridgeway, Waters, & Kuczaj, 1985), and by age 2 children begin to map emotion words to stereotypical facial expressions and to scenarios (Denham, 1986). However, individual differences are marked in both children’s production of emotion words (Denham, 1986; Dunn, Brown, & Beardsall, 1991; Pons & Harris, 2019) as well as in their understanding of emotion categories (Cutting & Dunn, 1999; Pons, Lawson, Harris, & de Rosnay, 2003). Further, stability in the trajectories of individual differences in emotion understanding has been observed across the preschool years (Brown & Dunn, 1996) and middle childhood (Pons & Harris, 2005), with the same children tending to perform well or poorly across developmental time. Therefore, it is important to understand the mechanisms that underlie children’s early development of emotion categories and use of emotion words.

To understand children’s use of emotion words, it is essential to first understand the emotional language environment in which children are embedded. Identifying the emotion words that children produce and the emotion words they hear in their early environments is vital for understanding how children learn emotion categories. This base knowledge will provide a foundation on which more nuanced questions about emotion category learning can be answered and may offer key insights into the information that children have access to and use to learn emotion categories. However, much remains unknown regarding what specific emotion words are present in children’s natural environments (Hoemann, Xu, & Barrett, 2019), including the frequency and type of emotion words produced.

Additionally, it is important to understand what aspects of the linguistic environment may influence children’s production of emotion words. Specifically, it is important to understand whether exposure to language in general or exposure to emotion words in particular is beneficial for children’s ability to learn emotion categories and to talk about emotions themselves. It may be the case that hearing more emotion words helps children to learn these words and to conceptually organize emotion categories (Shablack & Lindquist, 2019). In contrast, exposure to more general language may be beneficial, as it may provide children with more information which they may then extend to emotion categories. That is, children with higher general vocabularies may have more language at their disposal to think about and process the world around them, thereby allowing them to more readily interpret and categorize emotional content. For example, a child who sees their sibling throwing a tantrum in the store might hear their parent say “Oh no! This reminds me of how you felt when you had to eat broccoli with dinner last night”. If the child has enough general language skill to interpret this sentence, they may be able to similarly draw connections between emotional events even without the presence of specific emotion words.

Previous research has shown that the relation between age and a more mature conceptualization of emotions was mediated by general verbal abilities, but not by emotion language specifically across the age range of 6- to 25-years (Nook, Sasse, Lambert, McLaughlin, & Somerville, 2017). Further, young children with impairments in general language ability show deficits in emotion identification and attribution, even though they are capable of discriminating emotional stimuli (Rieffe & Wiefferink, 2017). Thus, it is possible that exposure to and knowledge of more language in general may facilitate children’s learning about emotions and their propensity to talk about emotional content. However, it is also possible that hearing emotion-specific language drives children to learn emotion categories and talk about them, as mothers’ elaborate discussions of emotions when talking about past events predicted children’s later emotion understanding (Laible, 2004). Thus, examining the language that children hear in their natural environment in terms of both emotion words and general language is of great importance.

Some studies have begun to answer questions regarding children’s early emotion word production. Ridgeway, Waters, and Kuczaj (1985) identified norms for when young children begin to understand and produce emotion words, and others have investigated the development of emotion language production from age 2 to 5 among a sample of five children (Wellman, Harris, Banerjee, & Sinclair, 1995). Additional research has examined various aspects of children’s early emotion language, including when children produce positive versus negative emotion words (Lagattuta & Wellman, 2002), how parent-and child reminiscing about past negative events relates to child socioemotional development (Laible, 2011), how parent mental-state talk relates to child emotion understanding (Tompkins, Beningo, Lee, & Wright, 2018), how mother and sibling references to feeling states predict a child’s later talk about feeling states (Dunn, Bretherton, & Munn, 1987), and how mother and father discussions about emotions during a picture-book task relate to child emotion understanding (LaBounty, Wellman, Olson, Lagattuta, & Liu, 2008). Indeed, emotion language is a broad term, that encompasses many key factors which may influence children’s ability to learn about emotions, including vocal tone, grammar, and narrative structure (e.g., Majid, 2012; Ochs & Schieffelin, 1989). Although the information presented from previous studies is key for understanding the type of emotion language children tend to hear, much remains unknown regarding children’s early emotion language environments. In particular, the specific emotion words (as opposed to broader mental-state talk) that children are exposed to and produce naturalistically, as well as the frequency with which such words are naturally produced, remains poorly understood. Thus, the present study will focus on emotion words in children’s early language environments.

Although the findings from previous research have been valuable for identifying developmental trajectories, some have included extensive transcript analyses from sample sizes as small as five or six children (e.g., Lagattuta & Wellman, 2002; Wellman et al., 1995), or relied on parent reports for children’s emotion vocabulary (e.g., Ridgeway, Waters, & Kuczaj, 1985). Thus, understanding what emotion words children naturally produce among larger samples of children is vital. Echoing this concern, Bretherton, Fritz, Zahn-Waxler, and Ridgeway (1986) previously expressed a need for further research into children’s production of emotion words, and more recently Hoemann, Xu, and Barrett (2019) have expressed concerns that there is little evidence regarding how frequently parents label emotions when their young children are present. Thus, it is crucial for additional research to identify key aspects of the emotion words in children’s environments using a large, natural dataset. The present study provides a novel contribution to the literature by identifying basic characteristics of young children’s emotion lexicon, including the frequency of naturalistic emotion word production among young children and their parents, the types of emotion words spoken, and factors that predict individual differences in a large, naturalistic sample of child emotion word production. Additional information regarding specifically what emotion words children produce and hear in their natural environment will be critical for furthering our knowledge of children’s emotional development.

The Present Study

In the present study, we addressed three key topics regarding children’s naturalistic use of emotion words: 1) Descriptive information regarding the frequency and type of emotion words that young children produce and hear, 2) Developmental trends in early emotion word production and exposure, and 3) Factors that predict children’s production of emotion words. To do so, we analyzed transcripts from 15- to 47-month-old children in the Child Language Data Exchange System (CHILDES; MacWhinney, 2000) database. We coded parent and child production of emotion words and reported descriptive statistics, including the frequency and type of emotion word production by parents and children. Further, we explored developmental trends by looking at how emotion word production for children and parents changed with age, and how our naturalistic data compared to existing developmental norms. Finally, we investigated factors that predicted children’s production of emotion words. Specifically, we assessed the effect of child age, parent and child general language complexity, parent emotion word production, and parent emotion word priming. Our analyses were largely exploratory, with the aim of gaining further insight into the emotion words in children’s natural environments without specific predictions about exact frequencies or trends. Although existing research has addressed some of these factors separately in smaller samples, the present study is the first, to our knowledge, to address all of these important aspects of children’s emotion word exposure and production together in one large naturalistic sample. Examining all of these pivotal aspects of the emotion words in young children’s early language environments together will provide novel, comprehensive insight into how young children begin to talk about emotions.

Method

Participants

The transcripts of 15- to 47-month old children interacting with their parents were accessed from the CHILDES database (MacWhinney, 2000), a publicly available data set of conversational interactions with children. Transcripts in this database were collected from a variety of researchers interested in various aspects of language. Thus, none of the transcripts were collected with the intent of investigating emotion words, and therefore represent naturalistic depictions of emotion word production among families. The age range was selected because children over 15 months but under 4 years of age are typically verbal and demonstrate substantial variability in emotion word production (Ridgeway, Waters, & Kuczaj, 1985) as well as linguistic complexity (Fenson et al., 1994; Hoff, 2013). The final data set for the present study included 179 children with an average age of 29.2 months (SD=7.4). The researchers did not collect any new data for this project and did not have access to any personally identifiable information regarding the participants (including informed consent documents), and thus this study was deemed exempt by the IRB at the University of California, Los Angeles, project title “CHILDES Investigations”, and adheres to the American Psychological Association ethical standards.

Transcript Selection

Transcripts included in the present study were selected based on several key criteria. All transcripts were in English and belonged to the English- North America collection of the CHILDES database. The target child of the transcript was always within the age range of 15- to 47-months and from a non-clinical sample. As mothers tend to contribute the most to childcare across early child development (Fillo, Simpson, Rholes, & Kohn, 2015), we included only transcripts in which mothers were directly interacting with the target child for the present analysis. Some transcripts included other adults (e.g., fathers, grandmothers, experimenters) or siblings, however these other adults contributed speech to these transcripts at low frequencies. Thus, only interactions between mothers and target children were analyzed. Further, to assess naturalistic conversations between mothers and children, we removed transcripts where an experimenter provided an explicit task to complete during the session, such as having the child complete a delayed gratification task or the parent read an assigned book. The rationale behind excluding these transcripts was that they would unnaturally constrain the parent-child conversation, and may result in certain words or conversations that the parent and child may not have had unprompted (e.g., If an assigned book includes many emotional events, this may lead to more emotional discussions than the parent and child would have naturally had). However, if transcripts involved parents and children choosing to read books during free play sessions, this was included. The determining exclusion factor was whether the particular book or task was explicitly assigned by an outside experimenter.

Transcripts from the same child at multiple ages were included, although some children only provided a transcript at one time point. This decision was made to maximize the number of transcripts included, and to allow analyses to include trends from children who grew and developed over time. This resulted in a final sample of 1,987 transcripts from 179 children in 17 corpora which were accessed in December 2019 (Bates, Bretherton, & Snyder, 1988; Bliss, 1988; Braunwald, 1997; Feldman & Menn, 2003; Gleason, 1980; Higginson, 1985; Kuczaj, 1986; Keefe, Feldman, & Holland, 1989; MacWhinney, 2000; MacWhinney & Snow, 1990; McCune, 1995; Morissett, Barnard, & Booth, 1995; Nelson, 1989; Post, 1994; Song, Demuth, Evans, & Shattuck-Hufnagel, 2013; Sachs, 1983; Suppes, 1974). The final sample size was determined based on including all available transcripts that adhered to our selection criteria.

The transcripts included in the present analyses were recorded from a variety of settings. Most of the transcripts involved free play sessions between mothers and children, but many occurred during other settings (e.g., meal times, bed time). Specifically, of the 17 corpora, 11 involved exclusively free play/toy play sessions, 1 involved bedtime and nap time sessions, 1 involved meal times, and the remaining 4 involved both free play and meal or snack time sessions. Further, most of the transcripts were recorded in the child’s home (13 corpora), but some also involved semi-naturalistic parent-child interactions in a physical laboratory setting (2 corpora), and some involved sessions both at home and in the lab (2 corpora). Including recordings from a variety of settings and contexts allows us to gain a more well-rounded view of the emotion words that children produce and hear during daily interactions. Additionally, the number of transcripts provided by each child was variable, with some corpora including 1 transcript per child, and other corpora including a maximum of 171 transcripts per child. On average, corpora included 37.3 transcripts per child (SD=47.87). Although the length of the transcript (in minutes) was not always provided in CHILDES, the majority of the audio and video recordings with reported time duration ranged from 15–45 minutes in length.

Emotion Talk

To assess emotion talk, we counted the number of times emotion words were produced by mothers and children in each transcript. To determine which words to include as emotion words we began by compiling a list based on two previous studies of emotion words (Baron-Cohen, Golan, Wheelwright, Granader, & Hill, 2010; Wellman, Harris, Banerjee, & Sinclair, 1995), lending over 300 potential emotion words in total. However, some of these words can be used to convey an emotion but are more frequently used to describe a non-emotion (e.g., “Absorbed” or “Broken”) or are used to convey emotions in the United Kingdom but may not be expected to frequently convey emotion among a North American sample (e.g., “Merry”). To account for this and ensure that our study investigated words primarily used to convey emotions among our sample, we conducted a survey with undergraduate students.

Fourteen undergraduate students (12 female), all between the ages of 18 and 24 participated in this survey. Each undergraduate viewed all 351 words and was asked to answer two questions for each word: Whether or not each was an emotion word (yes or no), and how frequently that word is used to convey an emotion (1=Always emotion word, 3=equally often emotion vs non-emotion, 5=Never emotion word). Based on these responses, words were removed from our final emotion word list if they were identified as “yes” to being an emotion word by 50% or fewer of the undergraduate participants. This resulted in a final list of 141 emotion words used in the present study. These 141 words were identified as “yes” to being emotion words by an average of 73.1% (SD=12.2) of participants. The average frequency of use as an emotion word score for the final word list was 2.5 (SD=0.5), indicating that they were perceived as more often used to indicate emotions than non-emotions. The final list of words can be found in the Appendix.

The childes-db r package version 0.1.2 (Sanchez et al., 2018) was used to identify emotion words within our selected transcripts. Using the get_types function, we searched for these specific 141 emotion words and counted how many times each emotion word from our list was produced by the mother or child within each transcript. As we were interested in children’s exposure to and production of emotion words, broadly construed, we included these emotion words regardless of whether they were used in noun or verb form as both forms are still used to convey emotional meaning. We then summed across all emotion words to calculate the total number of emotion words produced by each speaker within each transcript.

General Language Complexity

To calculate the general language complexity of each speaker within each transcript, we also used the childes-db r package to determine each speaker’s mean length of utterance (MLU). This value indicates the average number of morphemes (the smallest meaningful unit of language) for each speaker’s utterances (Brown, 1973), and is frequently used as a measure of the complexity of productive language (Gavin & Giles, 1996). MLU is highly correlated with other measures of language complexity, including the number of different words spoken (Dethorne, Johnson, & Loeb, 2005) as well as the number of different word roots, the Reynell Developmental Language Scales, and the Early Language Inventory (Bornstein & Haynes, 1998). Importantly, MLU can be reliably calculated with as few as 100 utterances (Gavin & Giles, 1996), and in contrast to other measures of language complexity (e.g., number of different words, word roots), MLU is not substantially impacted by transcript/sample length. As the transcripts included in our analyses from the CHILDES database varied notably in length and MLU correlates highly with other measures, MLU was selected as the best indicator of language complexity.

Results

Descriptive Information

First, we report basic descriptive characteristics of the emotion words produced in the naturalistic transcripts. To present a comprehensive report of the emotion words in children’s environment, we include descriptive statistics, the number of emotion words produced by mothers and children, the number of emotion word types versus tokens spoken, and gender differences.

Descriptive statistics.

We first report the number of emotion words, emotion word frequency, and language complexity, as these are key basic characteristics for understanding the emotion words in children’s natural environments and the information that they may be using to learn emotion categories. The average number of emotion words produced by children per transcript was 0.84 (SD=2.07). However, 1,396 of the 1,987 transcripts had 0 emotion words produced by a child, while the maximum was 29 emotion words produced by a child in one transcript. Further, when an emotion word was produced, the same word was repeated by the child multiple times within the same transcript on 338 occasions. The average number of emotion words produced per transcript by mothers was 2.01 (SD=4.36). 1,230 transcripts had 0 emotion words produced by the mother, and the maximum number of emotion words produced in one transcript by mothers was 34. Further, when an emotion word was produced by mothers, the same word was repeated multiple times within the same transcript on 750 occasions. On average, children produced 180.45 (SD=210.47) utterances per transcript, indicating that children were producing approximately one emotion word per 215.87 utterances. Mothers on average produced 210.26 (SD=294.68) utterances per transcript, indicating approximately one emotion word per 104.58 utterances. The average child MLU across transcripts was 3.71 (SD=1.94), while the average mother MLU was 5.35 (SD=1.62).

Considering the high number of transcripts in which mothers and children did not produce any emotion words, we followed up on this to examine differences between transcripts with and without emotion words. We began by identifying all transcripts in which neither speaker produced any emotion words, which resulted in 1,018 total. The remaining 969 transcripts were coded as having at least one emotion word produced by one of the speakers. Overall, the average child age for transcripts with no emotion words was 27.64 months (SD=6.92), while transcripts with at least one emotion word came from slightly older children (t(1985)=9.58, p<.001) at an average of 30.77 months (SD=7.62). It is interesting to note that there is substantial variability in age among both groups, as indicated by the reported standard deviations. Additionally, transcripts with at least one emotion word also had mothers with higher MLU (M=5.70, SD=1.40) compared to those with no emotion words (M=5.02, SD=1.75; t(1985)=9.62, p<.001). However, there was no difference in child MLU between the transcripts with at least one emotion word (M=3.72, SD=1.73) and those with no emotion words (M=3.70, SD=2.13; t(1985)=0.17, p=.866). Thus, transcripts with at least one emotion word were more likely to occur among older children and those where mothers, specifically, use more complex language.

Emotion word count.

To understand how children learn about emotion categories, it is important to not only know how frequently emotion words are spoken in general, but also which emotion words children hear and produce. To address this, we investigated the specific emotion words that were produced by both mothers and children. Of the 141 possible emotion words searched for in the transcripts, only 79 were spoken by a mother or target child. Children produced 47 unique emotion words and mothers produced 77. Figures depicting the frequencies separated by specific emotion words are presented separately for the target child (Figure 1) and mother (Figure 2). As the figures depict, some emotion words such as happy, love, scared, sad, angry, mad, afraid, and glad are produced relatively frequently by both speakers. However, other words such as hate and surprised are spoken frequently by one speaker but not the other. Additionally, for both mothers and children there were a few emotion words that were spoken with high frequency, followed by a sharp drop off with a high number of words spoken at very low frequencies. This pattern of results aligns with Zipf’s law (see Piantadosi, 2014), indicating that the reported results are likely reflective of natural language production.

Figure 1.

Frequency of emotion words spoken by the target child across all transcripts

Figure 2.

Frequency of emotion words spoken by the mother across all transcripts

Types vs tokens.

In addition to understanding the specific emotion words being produced, the diversity of emotion words spoken by mothers and children is important to explore in order to understand how children are making sense of emotional information. As described above, children produced an average of 0.84 emotion words per transcript (1,678 emotion words across all transcripts) and mothers produced an average of 2.01 words per transcript (4,009 emotion words across all transcripts). However, these counts allow for individual speakers to be providing the same emotion word more than once within a given transcript. Thus, in addition to looking at the total number of emotion words spoken (tokens) we also investigated the number of unique emotion words produced by each speaker within each transcript (types). Results revealed that, across all transcripts, children produced 899 emotion word types (an average of 0.45 per transcript), indicating that the remaining 779 emotion words produced involved children repeating words they had already spoken earlier in the same transcript. Across all transcripts, mothers produced 2,021 emotion word types (an average of 1.01 per transcript), indicating that the remaining 1,988 emotion words produced involved mothers repeating words that they had already spoken earlier in the same transcript. Thus, both mothers and children repeated emotion words often, but mothers still produced substantially more emotion word types than children on average.

Gender difference.

We also analyzed for differences by child gender, as previous research has shown that parents tend to talk about emotions more with their preschool-aged daughters than sons (Denham, Bassett, & Wyatt, 2010) and that preschool-aged girls use more emotion language than boys (Kuebli, Butler, & Fivush, 1995). Thus, in our large sample of transcripts we explored whether these patterns emerged in naturalistic speech. Of the 1,987 transcripts included, the majority (1,360) had a girl target child. Of the remaining transcripts, 538 had a boy target child and 89 did not specify the gender of the child in the transcript. Although there were substantially more girls than boys in the transcripts, we conducted some basic analyses to assess gender differences. Results revealed that the average age of boys (31.79 months) was significantly higher than that of girls (27.91 months; t(1896)=−10.51, p<.001, d=0.53). Due to this difference in age, additional gender differences were assessed covarying for child age. Further analyses revealed that, accounting for child age, mother MLU was significantly higher for girls than for boys (B=−.05, p=.022), and child MLU was significantly higher for girls than for boys (B=−.23, p<.001). Additionally, when accounting for child age, mothers used more emotion words with boys than with girls (B=.06, p=.009) but boys and girls did not differ in their use of emotion words (B=.02, p=.339), although these results should be interpreted with some caution given the uneven gender breakdown.

Developmental Trends

In addition to characterizing the emotion words in the environment of young children, it is also essential to understand how this changes with age. To assess this, we investigated the effect of child age on both mother and child emotion word production. We also compared these naturalistic corpus data to an open child vocabulary repository of existing developmental norms wherein parents reported the age at which their child is capable of producing certain words (Frank, Braginsky, Yurovsky, & Marchman, 2017). This comparison to existing norms allows for a more comprehensive understanding of the development of emotion word production in young children.

Age effects.

How do the emotion words in children’s environment change as children get older and begin to gradually understand more complex emotion categories? To investigate this developmental change, we analyzed how children’s emotion talk and the emotion words they hear from their mothers change with age. To do so, we binned transcripts into multiple age groups: Infants under 2 (N=547), 2-year-olds (N=1,017), and 3-year-olds (N=423). We then conducted two separate, cross-sectional one-way ANOVAs (one for the emotion words produced by children and one for mothers) comparing emotion words produced across these 3 age groups. To control for potential changes in overall number of words spoken, we assessed whether the proportion of emotion words produced out of total words produced changed by age group. Results revealed that the proportion of emotion words produced significantly changed by age bin for children (F(2, 1984)=10.68, p<.001, η_p²=.011) and mothers (F(2, 1984)=3.20, p=.041, η_p²=.003), with the proportion of emotion words produced increasing with age as depicted in Figures 3 and 4. However, as these analyses included some children at multiple time points, we followed up with multilevel model analyses investigating the effect of age as a continuous variable on emotion word production. These results mirrored the previous results, indicating that emotion words produced were significantly predicted by age for children (β=.05, p<.001) and mothers (β=.02, p=.043) Thus, as children get older they are both producing more emotion words themselves, and hearing more emotion words produced by their mothers.

Figure 3.

Proportion of emotion words produced by children in speech (out of total words spoken) by child age

Figure 4.

Proportion of emotion words produced by mothers in speech (out of total words spoken) by child age

Comparison to Existing Norms.

Previous work has aimed to characterize changes in children’s emotion word production across development by asking parents to report when their child produces various emotion words. To more comprehensively understand the development of children’s emotion word production, we compared children’s production of common emotion words to existing norms in order to assess how closely aligned parent reports were with children’s natural production. Specifically, we compared the ages at which children in our naturalistic sample began producing common emotion words relative to norms provided by WordBank (Frank, Braginsky, Yurovsky, & Marchman, 2017), an open child vocabulary repository. Vocabulary norms presented in WordBank represent the ages at which parents report that their child can produce particular words. Here, we compare this with children’s actual propensity to produce these words in their natural environment. We investigated the emotion words “happy”, “sad”, “mad”, and “scared”, as these are all included in the WordBank repository and are among the 6 most commonly produced emotion words in our sample. Wordbank data suggest that children at the 50^th percentile can produce these words at approximately the following ages: Happy=23 months; Sad=27 months; Mad=28 months; Scared=26 months. For comparison, the mean age of children from our transcript analysis who produced each of the following words was as follows: Happy=31.18 months (SD=7.44); Sad=31.76 months (SD=6.79); Mad=35.48 months (SD=6.12); Scared=33.67 months (SD=6.81). Additionally, the 25^th percentile for children’s age in our sample who produced each emotion word were as follows: Happy=24.95 months (SD=7.44); Sad=27.86 months (SD=6.79); Mad=30.53 months (SD=6.12); Scared=28.78 months (SD=6.81), although the youngest children to produce each emotion word in our naturalistic sample were below 17 months of age with the exception of “mad” (mad=24.76 months). Thus, the same general pattern of which words emerged at which ages was represented in the naturalistic transcript data (although children in the transcripts produced “scared” at later ages on average than “sad”). These data suggest that either parents may not be reliable reporters of their child’s emotion word production or (perhaps more likely) that there may be a notable lag between when children are capable of producing these emotion words, per parent report, and when they actually do so consistently in naturalistic speech. See Figure 5 for the total frequencies with which these four words reported in previous norms were produced per transcript among particular age groups. In line with the WordBank data, this figure demonstrates the later development of children’s production of the word “mad” relative to “happy”, “sad”, and “scared”. However, it is worth noting that children in our transcript analysis produced the synonyms “mad” and “angry” at comparable frequencies (see Figure 1), despite vocabulary norms frequently only including the word “mad”. Thus, examining children’s production of both synonyms may be an important direction for future research in order to best capture children’s production and understanding of this emotion concept.

Figure 5.

Total frequencies with which children produced the emotion words “happy”, “sad”, “mad”, and “scared” per transcript across age groups beginning at 15–21 months and ending at 39–47 months.

In addition to comparing child emotion word production to existing norms, we looked at mother production of emotion words. In particular, we were interested in whether the emotion words more commonly produced by mothers in our transcripts aligned with the emergence of emotion word production over development from existing norms. WordBank data suggest that “happy” is the earliest learned emotion word. Across the age range of 16- to 30-months, WordBank data shows “sad” and “scared” emerging next, with “mad” emerging last. Our transcript data presented in Figure 2 show how often these words are produced by mothers. Importantly, “happy” is stated substantially more often by mothers than any of the other three emotion words, which aligns with WordBank data showing that children learn this word earliest. Additionally, mothers produce the word “mad” less often than any of the other three words. This aligns with existing WordBank data and suggests that because this word is less frequent in children’s input, this may explain why it is produced later in development than the other three emotion words.

Factors Predicting Child Emotion Talk

To better understand the substantial individual differences in children’s production of emotion words (and thereby their understanding of emotions more broadly), we investigated what linguistic factors would (or would not) predict young children’s production of emotion words. We assessed the role of multiple linguistic factors using multilevel models and descriptive reporting.

Statistical analyses.

We assessed whether multiple aspects of the emotion words in children’s environment were related to their emotion word production. Data for each transcript included child age, the MLU for mother and child, as well as a count of emotion words produced by the mother. A three-level multilevel linear mixed model was used to analyze the impact of these predictor variables and to control for the substantial variability in our outcome variable (child emotion word production) accounted for at each of these levels (Corpus: B=.37, SD=.61; Child: B=.19, SD=.44). That is, the multilevel model was used to statistically account for the clustered nature of the data at the level of the child and the corpus (e.g., some children and some corpora contributed more transcripts to the analysis than others).

Predicting child emotion talk.

We used our multilevel model to assess how well each of the following factors predicted child emotion word production: Child age, child MLU, mother MLU, and mother emotion word production. When looking at each variable over and above the effect of the other variables in the model, results revealed that child age (β=.05, p<.001) and mother emotion talk (β=.21, p<.001) significantly predicted child emotion talk, but child MLU (β=.01, p=.690) and mother MLU (β=.00, p=.872) did not.

Priming.

An additional factor that may influence children’s production of emotion words is priming. That is, children may be more likely to produce emotion words that they have recently heard spoken. Imitation can be complex to characterize, as children can imitate exactly what an adult said, can imitate only part of what was previously said, or can expand on a parent’s imitation (Snow, 1981). Additionally, previous research has shown that imitation is relatively common in infancy, bur rare in older children (Seitz & Stewart, 1975). For our analyses, we investigated whether children’s production of emotion words, specifically, was related to priming by their mother. In other words, how often did children produce an emotion word when their mother had spoken that same emotion word earlier in the same transcript (regardless of how much earlier)? We examined priming within each transcript because this allowed us to ensure that the mother produced the word and it was heard by the child that same day typically within 45 minutes, although the specific amount of time between when the child heard the word and when the child produced it may have varied. To address this, we identified the emotion word types that children produced and identified how frequently those emotion words had been produced by the mother earlier in the same transcript. Results revealed that of the 899 emotion word types produced by children, 502 (55.84%) were words that had been primed by mothers. Thus, mothers’ production of a word does strongly relate to child production of the same specific word. Young children who naturalistically produce emotion words, therefore, are frequently, but not always, repeating a word that they had recently heard from their mother.

Discussion

In this study, we examined naturalistic emotion words both in children’s input and in their language production from late infancy through early childhood using a large sample of corpora from the CHILDES database (MacWhinney, 2000). Our results provide key descriptive information regarding the frequency and type of emotion words produced by parents and children, developmental trends, and linguistic factors that predict children’s naturalistic production of emotion words among a large sample of English-speaking children in North America. Notably, the general pattern of when emotion words emerged aligned with existing norms, providing further confidence in these results across multiple sources (i.e., parent report and naturalistic child language production). Further, results revealed that parent emotion word production, but not parent or child MLU, predicted child emotion word production. Addressing all of these aspects of the emotion words in the natural early language environment helps us to better understand the specific emotion words that most children are hearing and saying in early development, and this information may be useful for understanding emotional development more broadly. Investigating these frequencies and patterns of emotion words in children’s early language environment among a large sample across a broad age range in early development is necessary for more holistically understanding this aspect of early social development.

It should be noted that mothers produced more than twice as many emotion words per transcript as children produced, but there was substantial variability in the number of emotion words produced across transcripts by both mothers and children. Additionally, over 50% of transcripts did not involve a child producing any emotion words, and over 50% of transcripts did not involve a mother producing any emotion words. Thus, emotion words may not be produced very frequently in natural mother-child conversations, but when they are produced they are produced by mothers substantially more often than children. This suggests, not surprisingly, that when mothers do produce emotion words, this may be an important source of information for children’s developing emotion vocabulary. The descriptive data regarding specific emotion words produced by both speakers provides valuable information regarding how hearing emotion words at higher relative frequencies (e.g., happy) may allow children to consistently learn these emotion categories more readily and at younger ages than others, and why many other emotion categories which are spoken in their natural environments at lower frequencies (e.g., disappointed and calm) may take more time for children to learn. Why children produce some emotion words (e.g. hate) at higher frequencies compared to other emotion words relative to their mothers remains an important area for future research.

However, it is important to note that in all of the included transcripts, parents were aware that they were being recorded. Thus, it is possible that parents may have adjusted their behavior based on this fact, and therefore the present data may have underestimated the number of negative emotion words used by parents. Considering that many of the recordings occurred within the home, and the high number of utterances in most transcripts, it is likely that parents eventually acclimated to the presence of the recording device and therefore that the transcripts represent naturalistic data. However, the possibility that parents adjusted their behavior due to the recording device and minimized their use of negative emotion words should not be discounted.

An interesting pattern revealed in our results was that mothers’ emotion word production, but not mother or child MLU, predicted child emotion word production. This result aligns with previous research finding that maternal mental state language and general conversations about emotions are beneficial for children’s early emotional development (Dunn et al., 1987; Taumoepeau & Ruffman, 2006, 2008; Tompkins et al., 2018), but is novel in that it extends these findings to specifically the number of emotion words produced by the mother. While this likely indicates that children are learning emotion words from their mothers, this result is in contrast to prior research showing that general language abilities are important for developing more complex conceptualizations of emotions (Nook, Sasse, Lambert, McLaughlin, & Somerville, 2017). However, it is important to note that this previous study was conducted with 6- to 25-year-old participants, while our study was focused on children under the age of 4. Thus, we believe it is possible that early in development, hearing emotion words, specifically, may be helpful for children to gain this vocabulary and to begin talking about and conceptualizing emotions. After this foundation is built, general language abilities may be useful for older children to build on existing emotion categories. Exploring this possibility remains for future research.

Results from the present analysis also revealed notable differences between young boys and girls. Specifically, we found greater language complexity (as measured by MLU) among girls than boys, which is in line with previously published literature (e.g., Bornstein & Haynes, 1998; Fenson et al., 1994). Further, we found that mother MLU was higher when they were with daughters than with sons, which aligns with prior research suggesting that parents adjust their language to be more complex as their child’s language abilities are higher (van Dijk et al., 2013). However, an important limitation in these data is that they included substantially more girls than boys and on average older boys than girls, and thus these gender differences should be interpreted with caution. Additionally, we found no significant gender differences in terms of emotion word production from children, and that mothers provided emotion words more frequently to their sons than daughters. This appears in contrast to prior literature suggesting that girls tend to outperform boys in early emotion understanding and perception tasks (e.g., Brown & Dunn, 1996; Denham et al., 2015; Ontai & Thompson, 2002), and a recent meta-analysis identifying minimal differences in parenting of sons versus daughters (Endendijk, Groeneveld, Bakermans-Kranenburg, & Mesman, 2016). We posit that perhaps girls have improved language abilities (as indicated by higher MLU) and thus do not need as many emotion words in their input as their male counterparts in order to achieve the same level of emotion word production themselves. However, we acknowledge that this is speculative, and future research would be necessary to explore this possibility. In particular, additional research examining further detail about parent use of emotion words with boys and girls, such as use of negative versus positive emotion words (e.g., Lagattuta & Wellman, 2002) among a large naturalistic sample with a more even gender breakdown would be highly valuable.

Importantly, the results from the present study inform theories of emotional development. Specifically, two prominent theories differ in how they perceive language relating to emotional development. Basic Emotions Theory (Ekman, 1992) suggests that emotions have a strong evolutionary and biological basis, and therefore that the ability to interpret emotional information is strongly influenced by evolutionary history (Ekman, 1999). This theory holds that language is a communication device that is used to label emotions, but that it is not crucial for learning to interpret and understand emotions (Ekman & Cordaro, 2011). In contrast, the Theory of Constructed Emotion (Barrett, 2017) proposes that emotion categories are learned through contexts and experiences and that language (and emotion words, in particular) may facilitate the process of learning emotion categories (Hoemann, Xu, & Barrett, 2019). This is because due to the abstract nature of emotion categories, language may provide structure for cognitive representations of emotion categories, as language does for other abstract categories such as spatial relations (Loewenstein & Gentner, 2005). Our results indicate that mother’s production of emotion words significantly predicts children’s production of emotion words, which suggests that, relatively speaking, hearing emotion words more frequently may aid children in learning to talk about emotions and perhaps ultimately construct emotion categories. However, the present analyses were all conducted by comparing data within a single time point. Thus, future research may wish to use a lagged model to examine how mother’s emotion word use predicts her child’s emotion word use at a later time point. Importantly, the present results also provide key descriptive information regarding what emotion words young children commonly hear and produce, how this changes across development, and what linguistic factors do and do not predict children’s emotion word production. This information may provide a critical base for future research wishing examine support for whether prevalent theories (e.g., Basic Emotions Theory or the Theory of Constructed Emotion) explain children’s emotion category development. By more clearly characterizing the emotion words in children’s early language environments, the present study opens the door for future research to assess how children’s early language environments may or may not influence their developing understanding of emotions.

Although the results of the present study may provide a crucial base for studying children’s emotion word production (and perhaps emotion language more broadly), there are a few limitations that warrant discussion. First, we specifically examined emotion words in the present study. However, we did not investigate broader terms such as “good” and “bad”, which may sometimes be used to convey emotional meaning, but other times may not. Further, we were interested in how often emotion words were stated in total, but did not disentangle when an emotion word was used in its original form or negated (e.g., “glad” and “not glad” would both have been coded as 1 emotion word). Future research may wish to examine the frequency with which children use these broader emotion terms or negations to discuss emotional states. Additionally, we searched specifically for the 141 emotion words on our list, but did not include these words in other forms (e.g., we searched for “happy” but not “happier” or “happiest”). A cursory search for some of the most commonly spoken emotion words in other forms (specifically “happier”, “happiest”, “sadder”, “saddest”, “angrier”, and “angriest”) indicated that these words were infrequent in our transcripts (across all six words, these were only produced 13 times in total, 11 times by mothers). Nonetheless, it would be interesting and important for future research to examine how emotion words use varies across grammatical forms, and in particular how this may change across child development, potentially among older age groups. Additionally, although the present study focused specifically on emotion words, other aspects of emotion language such as grammar (Majid, 2012), narrative structure (Ochs & Schieffelin, 1989), syntax (Shablack, Becker, & Lindquist, 2020), and cause and effect discussions (Russell, 1990; Widen & Russell, 2010) warrant investigation. Further, the scope of the present study involved examining linguistic factors that predict children’s use of emotion words, but other cognitive factors (such as children’s abstract thought or discussions about abstract concepts) may also be important to examine in future research (e.g., Nook et al., 2017).

The results presented here describe how emotion words are used in children’s early language environments, which likely contributes to children’s understanding of emotion categories. For example, children who know certain emotion words (e.g., “fear”) categorize emotional faces differently from children who do not know these words (Widen & Russell, 2008). Therefore, children’s use of emotion words likely plays an important role in their emotion understanding development. However, future experimental research is necessary to clarify this relationship and to fully understand how emotion words relate to children’s emotion understanding. For example, researchers may wish to examine how the presence vs absence of emotion words influences children’s ability to draw connections between facial expressions and emotional scenarios with which they are not yet familiar. Additionally, studies asking young children to explain their reasoning about how characters would feel in certain scenarios when they either do or do not know the relevant emotion word would provide clarification regarding how exactly emotion words may relate to emotion understanding development.

In addition to clarifying the relation between emotion word production and emotion understanding, there are several other important areas where future research should build on the present findings. First, in an effort to improve children’s use of emotion words, it may be important to create interventions to increase emotional language use by parents as has previously been successful in preschool settings (Grazzani & Ornaghi, 2011; Grazzani, Ornaghi, Agliati, & Brazzelli, 2016). Knowing the typical frequency of emotion word production and the types of emotion words commonly produced during these interactions, as presented here, may aid in the creation of such interventions. Second, future research may also wish to examine the emotion words that children hear in a broader context than direct conversations with their mother (e.g., in overheard speech, when talking to other adults) in order to further inform such potential interventions. Third, it should be noted that the current study analyzed data from exclusively English-speaking participants. Whether or not there are differences in this pattern of results across a more diverse array of languages is an important question that may also be addressed in future research.

Conclusion

In conclusion, results from the present study provide valuable, novel insight into the emotion words in the natural environment of young children using a large sample of mother-child interaction transcripts. Altogether, the present study clarifies the nature of emotion words in children’s early language environment, how the language environment changes across development, and what factors influence children’s emotion word production. Learning to understand emotions represents a challenging problem, as children must integrate information across a variety of domains. Therefore, identifying what emotion words children are actually exposed to on a daily basis as they are developing these categories is a pivotal first step to understanding how these categories are formed. Some previous work has begun to investigate the emotion words in children’s early language environments (Frank, Braginsky, Yurovsky, & Marchman, 2017; Ridgeway, Waters, and Kuczaj, 1985; Wellman, Harris, Banerjee, & Sinclair, 1995), but the present study extends these findings to a large, naturalistic sample of children’s actual emotion word input and production. Thus, these findings will provide a crucial base from which future research can examine more fine-grained questions regarding emotion words early in development. Ultimately, the information from the present study may be particularly useful when considering theories (e.g., Barrett, 2017) that suggest an important role of emotion words in the development of early emotion understanding, and thereby in early social cognition.

Highlights.

• Analyzed nearly 2,000 transcripts for emotion language of 15- to 47-month-olds

• Assessed emotion language that children heard and produced in CHILDES database

• Frequency of specific emotion words and developmental trends are presented

• Child emotion language was strongly predicted by age and mother emotion language