Do Preverbal Infants Understand Discrete Facial Expressions of Emotion?

Looking-Time Paradigms

Infant emotion discrimination has traditionally been measured with looking-time paradigms (for detailed discussion, see Oakes, 2010). In paired-preference paradigms, infants are shown two static facial expressions side by side. If infants look longer at one expression compared to the other, it is assumed that they (a) have a visual preference and (b) can discriminate the expressions. These looking-time differences are thought to reflect either a familiarity preference (e.g., in the case of happy expressions; Farroni, Menon, Rigato, & Johnson, 2007) or a novelty preference (e.g., in the case of fearful expressions; C. A. Nelson & Dolgin, 1985). However, if infants look equally long at both expressions, it is impossible to determine whether they (a) cannot discriminate the expressions or (b) do not have an expression preference.

Familiarization and habituation paradigms provide a more conclusive test of discrimination. In these paradigms, infants are repeatedly shown one expression (e.g., happiness). After a fixed number of trials (familiarization) or after infants’ looking time decreases to a certain criterion (habituation), infants are sequentially shown a novel expression (e.g., sadness) and a familiar expression (e.g., happiness). Infants provide evidence of discrimination if they look longer at the novel expression. Although these studies generally indicate that infants discriminate between various facial expressions, findings depend on the (a) stimuli (static vs. dynamic), (b) paradigm (paired-preference vs. habituation/familiarization), and (c) expression contrast (happiness–anger vs. happiness–fear).

Event-Related Potential (ERP) Paradigms

Researchers initially turned to ERP paradigms to examine why infants have visual preferences for some facial expressions. In these paradigms, infants observe multiple brief (< 1,000 ms) presentations of static facial expressions. ERPs are averaged from a continuous recording of electrical signals at the scalp, time-locked to the presentation of each expression. Infant studies have primarily focused on three ERP components: the N290, P400, and Nc (negative central). The N290 and P400 are thought to be “precursors” to the face-sensitive adult N170 (Rigato, Farroni, & Johnson, 2010), whereas the Nc is thought to relate to increased allocation of attention (de Haan, Johnson, & Halit, 2003).

The “Fear Bias”

One of the most consistent findings in both looking-time and ERP paradigms is heightened attention to fearful compared to happy expressions (for a review, see Leppänen & Nelson, 2012). Developmental researchers typically explain this “fear bias” in terms of the threat-signaling value of fearful expressions (e.g., Jessen & Grossmann, 2014), which might be adaptive for self-locomoting infants (Campos et al., 2000). Thus, the emergence of crawling is often used to explain why 7-month-olds, but not younger infants, attend more to fearful than happy expressions (e.g., Grossmann & Jessen, 2017; Jessen & Grossmann, 2016; Leppänen, Cataldo, Bosquet Enlow, & Nelson, 2018; but see Bayet et al., 2017; Heck, Hock, White, Jubran, & Bhatt, 2016, 2017). However, this is very much a classical explanation. In line with a constructivist view, there may be other explanations for these findings.

One possibility is that heightened attention to fearful expressions reflects a general negativity/threat bias (Vaish, Grossmann, & Woodward, 2008). This seems unlikely, however, given that 5- to 14-month-old infants do not have a visual preference for angry over happy expressions. Furthermore, even though 7-month-olds show larger Nc amplitudes in response to fearful than happy expressions, Nc differences are not found when anger is compared to happiness (Grossmann et al., 2007; Parker & Nelson, 2005). Thus, there may be something especially “attention-grabbing” about fearful expressions that is not shared with other negative, threat-related expressions like anger (although see Morales et al., 2017).

Another possibility is that low-level perceptual features of fearful expressions—such as wide eyes—elicit infant attention. Interestingly, however, 7-month-olds allocate more attention (i.e., longer looking times, larger Nc responses) to happy eyes than fearful eyes (Jessen & Grossmann, 2014, 2016; Krol et al., 2015). Also, in attentional disengagement tasks, 7-month-olds more slowly shift their attention to a peripheral target when presented with fearful expressions, compared to neutral expressions with wide, “fearful” eyes (Peltola, Leppänen, Mäki, & Hietanen, 2009). Thus, while wide eyes may elicit attention, this feature may not be the sole explanation for the fear bias.

A more likely explanation is that fear expressions are unfamiliar. Young infants are rarely exposed to these expressions (Malatesta & Haviland, 1982) and caregivers describe prototypical fear displays as “unnatural” or “uncharacteristic of their normal behavior” (Camras & Sachs, 1991; Rosen, Adamson, & Bakeman, 1992). Direct evidence also suggests that infants view fear expressions as “novel.” For example, in an attentional disengagement task, 7-month-olds were equally likely to fixate on nonemotional, “novel” expressions (i.e., lips closed, cheeks blown full of air, eyes open) compared to fearful expressions (Peltola, Leppänen, Palokangas, & Hietanen, 2008). In addition, the fear bias relates to positive maternal emotionality (de Haan, Belsky, Reid, Volein, & Johnson, 2004), suggesting that infants are attentive to expressions that are not typically encountered in their daily lives. Finally, the fear bias has been found to decline around 11 to 12 months of age (Peltola, Hietanen, Forssman, & Leppänen, 2013), presumably as infants gain more experience with fearful expressions (Xie et al., 2019).

Summary

These findings indicate that, by 5 months of age, infants can discriminate between happy and negative facial expressions (fear, anger, sadness). However, it is unclear whether infants at any age can discriminate different negative facial expressions at a behavioral or neural level. There is some, albeit limited, evidence for within-valence discrimination in the looking-time literature (e.g., Schwartz et al., 1985), but the ERP findings are inconsistent. Despite the lack of concrete evidence for within-valence discrimination, classical emotion theories have interpreted the discrimination literature as support for an early emerging preparedness for emotion understanding, particularly with regard to fearful expressions (e.g., Leppänen & Nelson, 2006). However, heightened attention to fear expressions does not necessarily mean that infants “understand” these expressions as “threatening.” Consistent with this interpretation, one study recently reported that the fear bias at 7 months does not correlate with emotion understanding at 48 months (Peltola, Yrttiaho, & Leppänen, 2018). In contrast, some constructionist theories have argued that infants discriminate facial expressions on the basis of isolated perceptual features, without understanding emotional meaning (Barrett, 2017; Lindquist & Gendron, 2013).

Given the reviewed literature, the interpretations made by both theories seem premature. There is currently no empirical metric to determine the nature of infants’ responses in looking-time and ERP tasks (Madole & Oakes, 1999). In other words, it is not possible to determine whether infants discriminate facial expressions based on (a) salient perceptual features alone (e.g., mouth shape), (b) affective meaning alone (i.e., the communicative signal of the expression), or (c) some combination of the two. Discrimination studies, as they are currently designed, are ultimately unable to provide meaningful insights into whether infants have discrete emotion concepts.

Between-Valence

There is some evidence that 7- to 10-month-olds can form a category of happiness (i.e., after habituation to happy faces) and differentiate this category from novel anger and fear expressions at test (A. Caron et al., 1988; Kestenbaum & Nelson, 1990; Ludemann, 1991; C. A. Nelson & Dolgin, 1985; C. A. Nelson et al., 1979; Safar & Moulson, 2017). It remains unclear, however, whether 3- to 12-month-olds can differentiate a category of happiness from novel sad expressions at test (A. Caron et al., 1988; Lee, Cheal, & Rutherford, 2015; Walker-Andrews et al., 2011). Moreover, some studies fail to find any evidence of happy categorization, even in infants as old as 11 months of age (Amso, Fitzgerald, Davidow, Gilhooly, & Tottenham, 2010; Phillips, Wagner, Fells, & Lynch, 1990; Schwartz et al., 1985; Serrano et al., 1995).

With respect to negative emotions, 4- to 12-month-olds can sometimes form a category of anger expressions and differentiate this category from happiness at test (R. Caron, Caron, & Myers, 1985; Lee et al., 2015; Serrano et al., 1995; but see Phillips et al., 1990; Schwartz et al., 1985). However, 6- to 11-month-olds do not seem to form a category of fearful expressions (Amso et al., 2010; Ludemann & Nelson, 1988; C. A. Nelson & Dolgin, 1985; C. A. Nelson et al., 1979; Safar & Moulson, 2017; but see Cong et al., 2018) or sad expressions (Lee et al., 2015; Walker-Andrews et al., 2011) when presented with happy expressions at test.

In these studies, it is unclear whether infants’ categories are based on salient perceptual features (e.g., teeth) alone or affective meaning. To test this question, infants have been presented with happy expressions that vary either in intensity or amount of teeth. Infants at 5 to 12 months of age can form a happy category even when the expressions vary in intensity during habituation (i.e., small, closed-mouth smiles and big, toothy smiles; Bornstein & Arterberry, 2003; Kotsoni, de Haan, & Johnson, 2001; Lee et al., 2015; Ludemann & Nelson, 1988; but see Cong et al., 2018; Phillips et al., 1990). In contrast, when salient facial features vary systematically between the habituation and test trials, infants use those features as the basis for categorization. Specifically, after habituation to nontoothy happy expressions, 4- to 7-month-olds show heightened attention to a novel model expressing toothy happiness (R. Caron et al., 1985). However, when the amount of teeth is held constant from habituation to test, 7-month-olds provide evidence of categorization (Kestenbaum & Nelson, 1990). Although this sensitivity seems to decrease over the first year of life, it is still evident around 10 months of age (R. Caron et al., 1985; Ludemann, 1991). It is unknown whether infants older than 10 months continue to be influenced by these perceptual cues.

Within-Valence

Only a handful of studies have used negative facial expressions during both habituation and test. These studies indicate that 4- to 18-month-olds can form a category of anger (i.e., after habituation to anger expressions) and differentiate this category from novel sad, fear, and disgust expressions (Ruba, Johnson, Harris, & Wilbourn, 2017; Schwartz et al., 1985; Serrano et al., 1992). Moreover, 10- and 18-month-olds can form a category of disgust and differentiate this category from anger expressions at test (Ruba et al., 2017). Findings are mixed as to whether 4- to 6-month-olds can form a category of sadness or fear during habituation (Schwartz et al., 1985; Serrano et al., 1992). Given that infants’ categorization abilities are tenuous before 7 months of age (e.g., A. Caron et al., 1988; R. Caron et al., 1982), it is possible that only older infants can form these within-valence categories. On the other hand, in a recent paired-preference study, 5-month-olds were sensitive to the categorical boundary between sadness and disgust, as well as sadness and anger (White et al., 2019). In the same study, however, 5- and 9-month-olds were not sensitive to the category boundary between anger and disgust.

Summary

The categorization literature provides some evidence that, by 7 months of age, infants can form (a) a category for happiness and differentiate this category from (some) negative expressions, and (b) a category for anger and differentiate this category from happy expressions. There is also emerging evidence that infants can form categories of discrete negative expressions and differentiate these categories from other negative expressions. From a classical perspective, these findings could be used to argue that infants “understand” discrete facial expressions (e.g., Walker-Andrews, 1997). If infants can perceive that multiple people are displaying the same emotion, then these categories might be conceptual (i.e., based on affective meaning). On the other hand, constructionist theories would argue that these categories are still perceptual in nature. Infants may attend to a shared facial feature across models (e.g., scrunched noses on disgust expressions) as the basis for these categories. To date, however, there have been few systematic efforts to manipulate or control for salient facial features in categorization tasks.

Similar to the discrimination literature, it is difficult to discern the nature of infants’ responses in categorization tasks. If infants understand the affective meaning of discrete facial expressions, then they would likely draw on this information, even if they are still influenced by salient perceptual features. In fact, even though adults have discrete emotion concepts, their emotion categorization is still influenced by facial features (e.g., presence of teeth; Ruba, Wilbourn, Ulrich, & Harris, 2018). Thus, the existent categorization literature also cannot answer the question of whether infants have discrete emotion concepts.

Between-Valence

Multiple studies have confirmed that 5- to 12-month-olds can match happy and sad vocalizations/tones to their respective facial expressions (Flom & Whiteley, 2014; Phillips et al., 1990; Walker, 1982; but see Soken & Pick, 1999). Younger (3.5-month-old) infants can also form these matches, albeit only when the expressions are posed by their mothers (Kahana-Kalman & Walker-Andrews, 2001). Infants between 5 and 7 months of age can also match happy and angry vocalizations to facial expressions (Grossmann, Striano, & Friederici, 2006; Soken & Pick, 1992, 1999; Vaillant-Molina, Bahrick, & Flom, 2013; Walker, 1982; Walker-Andrews, 1986; but see Palama et al., 2018). More recently, however, Ogren, Burling, and Johnson (2018) reported that 9-month-olds did not form intermodal matches for happy, sad, and angry expressions (when paired with a neutral expression). Unlike previous research, this study controlled for baseline expression preferences, thereby providing a more stringent test of intermodal matching.

Within-Valence

Only two studies have tested whether infants can form intermodal matches when two negative facial expressions are presented. Phillips et al. (1990) found that 7-month-olds did not match loud and quiet tones to anger and sad faces, respectively. However, 7-month-olds formed intermodal matches for angry and sad faces when the vocal expressions contained human speech (Soken & Pick, 1999). Thus, successful intermodal matching might depend on using ecologically valid auditory cues. Further, although infants might match negative emotions across the dimension of arousal (anger is high arousal, sadness is low arousal; Russell, 1980), it is unknown whether they can match negative emotions within the dimension of arousal (e.g., anger vs. fear).

Summary

In summary, 5- to 12-month-olds can match positive and negative faces to vocalizations. However, because these studies have yet to compare expressions within one dimension of valence and arousal, it is unknown whether these responses are based on discrete emotions (e.g., happy vs. sad). Regardless, from a classical standpoint, intermodal matching is assumed to go beyond simple expression discrimination, instead signifying “emotion recognition” (Walker-Andrews, 1997). More specifically, infants are thought to recognize the common affective information communicated across modalities.

However, a leaner interpretation cannot be discounted. Infants might have simply learned that certain facial expressions (e.g., a smile) and certain vocal expressions (e.g., laughter) co-occur in their social environment. Consequently, infants could make intermodal matches without understanding the affective meaning of the expressions. In support of this, Grossmann et al. (2006) found that 7-month-old infants showed larger Nc and Pc (positive component) amplitudes to congruent facial–vocal expression pairs, compared to incongruent pairs. As previously mentioned, the Nc is thought to reflect heightened visual attention (de Haan et al., 2003), while the Pc is thought to reflect memory for familiar items (C. A. Nelson, Thomas, de Haan, & Wewerka, 1998). This finding suggests that infants may have been relying on their memory for the learned associations between facial and vocal expressions.

Even if intermodal matching recruits emotion concept knowledge, the current findings do not refute constructionist emotion views. For instance, infants’ ability to match positive and negative facial expressions to vocalizations is consistent with the hypothesis that infants have valence- and arousal-based emotion concepts. However, given that discrete emotional expressions are thought to occur with statistical irregularity (Barrett, 2017), preverbal infants should have more difficulty forming intermodal matches for emotions within a dimension of valence and arousal (e.g., fear vs. anger). Because these emotion contrasts have yet to be studied, the current intermodal matching literature cannot address this claim.

Between-Valence

Two studies suggest that by late in the first year of life, infants match positive facial expressions with positive events. Hepach and Westermann (2013) reported that 10- and 14-month-olds expected an emoter to express happiness, rather than anger, when patting a stuffed animal. Similarly, Skerry and Spelke (2014) found that 8- and 10-month-olds expected an agent to express happiness, rather than sadness, after completing a goal. In this study, the agent was a circle that expressed positive affect by smiling, giggling, and bouncing, or negative affect by frowning, crying, and slowly rocking side to side. Thus, it is possible that infants’ responses were driven by the vocal and/or movement cues, rather than by the agent’s “facial expressions.”

Infants do not seem to match negative emotions to negative events until the second year of life. For instance, neither 8- nor 10-month-olds in Skerry and Spelke’s research (2014) expected an agent to express sadness, rather than happiness, after failing to complete a goal. Moreover, in Hepach and Westermann’s study (2013), 14-month-olds, but not 10-month-olds, expected an emoter to express anger, rather than happiness, when hitting a stuffed animal. Consistent with these findings, Reschke et al. (2017) reported that 12-month-olds expected an emoter to express (a) sadness or anger, rather than happiness, after fighting over a toy, and (b) happiness, rather than anger, after receiving a toy. Infants did not expect an emoter to express sadness, rather than happiness, after someone broke the emoter’s toy. Finally, Chiarella and Poulin-Dubois (2013) reported that 18-month-olds, but not 15-month-olds, expected an emoter to express (a) sadness, rather than happiness, after an object was taken away, and (b) happiness, rather than sadness, after receiving a desired object.

Within-Valence

To date, only two studies have examined whether infants match different negative facial expressions to different negative events. Reschke et al. (2017) reported that 12-month-olds did not expect an emoter to express (a) sadness, rather than anger, after another person broke the emoter’s toy, or (b) anger, rather than sadness, after fighting over a toy with another person. In contrast, Ruba, Meltzoff, and Repacholi (2019) found that 14- and 18-month-olds expected an emoter to express (a) anger, but not disgust or fear, after failing to achieve a goal, and (b) disgust, but not anger or fear, after tasting a novel food. Infants at this age did not expect an emoter to express fear, rather than anger or disgust, after encountering a novel object.

Summary

This relatively new body of research suggests that infants can match (a) positive emotions to positive events late in the first year of life, and (b) negative emotions to negative events in the second year of life. In the second year of life, infants are also beginning to match different negative facial expressions to specific negative events. A classical interpretation of these findings is that infants “understand” something about the causes of discrete emotions. However, similar to the intermodal matching literature, it could be argued that infants are simply remembering specific event–emotion associations encountered in their daily lives, without understanding the emotions or the causal link between emotions and events. We hypothesize that infants’ event–emotion matching reflects some, albeit limited, degree of emotion understanding. For instance, infants might link emotional expressions with specific classes of events (e.g., goal achievement vs. goal failure; Meltzoff, 1995; Woodward, 1998) as opposed to specific events experienced or observed by the infant (e.g., obtaining a desired stuffed toy). ³

However, the extent to which infants can form event–emotion matches based on discrete emotions remains unclear. To date, only one study has tested emotions within a dimension of valence and arousal (Ruba et al., 2019). Importantly, this study found that preverbal infants were able to match some negative emotional expressions to specific events. This suggests that discrete event–emotion matching may be possible before infants learn emotion labels. In other words, infants may be able to detect the regularities between facial expressions and eliciting events, without needing to be explicitly taught these associations via language. Although the findings of Ruba et al. (2019) challenge the theory of constructed emotion (Barrett, 2017), it is worth noting that this study tested bimodal expressions (i.e., face and voice).

Between-Valence

Most studies have compared a happy or neutral expression to a negative expression (for a table of studies, see Vaish et al., 2008). Few differences emerge in 10- to 18-month-old’s responses to objects that have been the target of a happy versus a neutral expression (e.g., Hornik, Risenhoover, & Gunnar, 1987; Mumme & Fernald, 2003; Mumme, Fernald, & Herrera, 1996; Repacholi, 2009). In contrast, numerous studies have indicated that 10- to 24-month-olds approach and/or touch an object if the emoter expresses happiness or neutral affect, but avoid the object (delayed and/or reduced object contact) if the emoter expresses fear (e.g., Kim & Kwak, 2011; Kim, Walden, & Knieps, 2010; although see Leventon & Bauer, 2013). Similar findings have been obtained for 11- to 18-month-olds with happy/neutral versus disgust expressions (e.g., Carver & Vaccaro, 2007; Chiarella & Poulin-Dubois, 2018; Flom & Johnson, 2011; although see Schieler, Koenig, & Buttelmann, 2018). In addition, 15- and 18-month-olds are less likely to imitate a model’s actions that have been the target of sad or angry expressions, compared to happy or neutral expressions (e.g., Patzwald, Curley, Hauf, & Elsner, 2018; Repacholi, Meltzoff, Spiewak Toub, & Ruba, 2016). Taken together, these findings suggest that, by 10 to 12 months of age, infants understand something about the functional behavioral responses specified by positive and negative facial expressions.

Most of this research has focused on infants 10 months of age and older, and it is unclear whether younger infants also engage in social referencing. A few studies have failed to find evidence that 6- to 9-month-olds regulate their behavior in response to an adult’s happy, fearful, and disgust expressions (Slaughter & McConnell, 2003; Walden & Baxter, 1989; Walden & Ogan, 1988). In contrast, Vaillant-Molina and Bahrick (2012) reported that 5.5-month-olds preferentially touched a toy that had been linked with a happy expression compared to a fearful expression. Infants in this study were habituated to the emotion–object pairings before the behavioral response period, and this increased exposure may have helped infants encode these pairings. In addition, several ERP studies have found that infants as young as 3 months show increased Nc activity to pictures of objects previously paired with fear or disgust expressions, compared to neutral or happy expressions (e.g., Carver & Vaccaro, 2007; Hoehl & Striano, 2010; Hoehl, Wiese, & Striano, 2008; but see Aktar et al., 2016; Leventon & Bauer, 2013). These ERP findings suggest that infants are more attentive to objects that are linked to negative facial expressions. In summary, younger infants may engage in social referencing when the tasks are more developmentally appropriate.

Within-Valence

To date, few social referencing studies have examined infants’ responses to within-valence emotions. In a classic study, 12-month-olds were more likely to cross a visual cliff when their mothers posed sadness compared to anger and fear expressions (Sorce et al., 1985). Similarly, Martin and colleagues (Martin, Maza, McGrath, & Phelps, 2014; Martin, Witherington, & Edwards, 2008) reported that 16- to 18-month-olds touched target objects for shorter durations in response to anger and fear expressions compared to sadness. One explanation for these findings is that—in the context of an ambiguous object/event—high arousal, negative emotions (anger, fear) communicate threat and danger (Shariff & Tracy, 2011), and avoidance is an appropriate response to both emotions (Walle & Campos, 2012). These studies do, however, provide evidence for arousal-based behavioral responses (sadness vs. anger/fear).

One limitation of social referencing studies is the use of a relatively limited behavioral coding system (i.e., behaviors coded as either approach or avoidance). Recently, Walle, Reschke, Camras, and Campos (2017) designed a coding system focused on the “goal” of infants’ behavioral response (e.g., prosocial responding, relaxed play, social avoidance). Infants (16-, 19-, and 24-month-olds) saw an emoter displaying multimodal (face, voice, posture, gesture) expressions of sadness, anger, fear, or disgust in response to two events. Compared to the other three negative emotions, 24-month-olds showed greater avoidance of the emoter when she displayed anger. In addition, 19-month-olds (but not 16- or 24-month-olds) demonstrated more “information seeking” (i.e., alternating their gaze between the object and emoter) in response to disgust than anger. However, infants’ “information seeking” did not differ in response to fear compared to disgust or anger. Thus, evidence for differential responding to negative emotions was less clear-cut at this age.

Summary

By 10 to 12 months of age, infants can use another person’s positive and negative emotional expressions to regulate their own behavior. According to a classical view of emotion, this suggests that infants understand the meaning of these emotional expressions. For instance, infants may understand that positive emotions communicate safety and approach, while negative emotions signify danger and avoidance (Shariff & Tracy, 2011).

However, an alternative interpretation is that adults’ emotional expressions directly modify infants’ own felt emotions and subsequent behavior. In this interpretation, infants need not understand the emotional expression as a meaningful signal in order to regulate their behavior. Consistent with this contagion hypothesis, some evidence suggests that infants display more negative affect in response to an emoter’s fearful expressions, and more positive affect when the emoter expresses happiness (e.g., Hirshberg & Svejda, 1990; Mumme et al., 1996). However, other studies have failed to find differences in infants’ affect, particularly when the emoter expresses anger or disgust (e.g., Hertenstein & Campos, 2004; Repacholi, 2009; Repacholi, Meltzoff, Hennings, & Ruba, 2016).

Further evidence against the contagion hypothesis comes from social referencing studies that have manipulated stimulus ambiguity, emoter competence, and emoter attention. For instance, when the experimental stimuli are low in ambiguity, adults’ emotional expressions have little or no impact on infants’ behavior (e.g., Kim & Kwak, 2011; Tamis-LeMonda et al., 2008). Behavioral regulation is also less likely if the emoter is “incompetent” (e.g., Stenberg, 2012, 2013). Finally, infants are less likely to regulate their behavior if the emoter is not visually attending when the infant has access to the object (Botto & Rochat, 2018; Repacholi, Meltzoff, & Olsen, 2008; Repacholi, Meltzoff, Rowe, & Spiewak Toub, 2014). These modulations suggest that infants’ behavioral regulation cannot be reduced to emotional contagion. From a contagion perspective, the adult’s expression directly modifies the infant’s own affective state (e.g., a fearful expression causes the infant to become scared, which in turn inhibits their object exploration). If infants are “catching” adults’ emotion via contagion, then they should regulate their behavior regardless of these manipulated task features. However, this is not the case.

Even if social referencing reflects true understanding of emotions, the findings are not inconsistent with the constructionist view that infants have valence- and arousal-based emotion concepts (e.g., Barrett, 2017; Widen, 2013). For instance, Walle et al. (2017) found that 24-month-olds differentially responded to different negative emotions, but at this age, infants are quite verbal. Specifically, emotion labels (e.g., “mad,” “angry”) are beginning to emerge in infants’ productive vocabularies (Ridgeway, Waters, & Kuczaj, 1985). Although no language data were reported in this study, it could be argued that infants’ newly acquired emotion language facilitated their understanding of these discrete emotions. Another potential issue is the use of multimodal expressions. Some studies have suggested that the vocal expression, rather than the facial expression alone, drives social referencing (Kim et al., 2010; Mumme et al., 1996; Vaillant-Molina & Bahrick, 2012; Vaish & Striano, 2004). Thus, even if preverbal infants were to show distinct behavioral responses to discrete negative emotions, it would be useful to determine which expressive modality primarily influences infants’ behavior.

Language and Emotion Categorization in Children and Adults

Research with older, verbal children and adults, however, does suggest that language constructs emotion categories. First, emotion words influence how facial expressions are encoded and remembered (e.g., Brooks et al., 2017; Doyle & Lindquist, 2018; Fugate, Gendron, Nakashima, & Barrett, 2018). For example, adults remember facial expressions as “angrier” or “happier” depending on whether the expressions were paired with the word “angry” or “happy” (Halberstadt & Niedenthal, 2001). In addition, the inclusion of emotion labels in emotion categorization tasks improves children’s and adults’ performance (e.g., Camras & Allison, 1985; Carroll & Russell, 1996; N. L. Nelson et al., 2018; N. L. Nelson & Russell, 2016; Nook, Lindquist, & Zaki, 2015). For example, when asked to sort facial expressions into different categories, (a) 2- to 7-year-olds are more accurate when the categories are specified by an emotion label (Russell & Widen, 2002; Widen & Russell, 2004), and (b) adults are more accurate after reading instructions that include specific emotion labels (i.e., “you will sort anger and disgust expressions”; Ruba et al., 2018). In contrast, reduced accessibility to emotion labels leads to slower and less accurate facial expression categorization in adults (Gendron, Lindquist, Barsalou, & Barrett, 2012; Lindquist, Barrett, Bliss-Moreau, & Russell, 2006; Lindquist, Gendron, Barrett, & Dickerson, 2014). Taken together, these studies suggest that language fundamentally impacts emotion categorization in children and adults.

However, one clear limitation to this research is that children and adults have considerable experience with facial expressions and emotion labels. In particular, emotion labels and concepts are always implicitly available in participants’ minds (Lindquist & Gendron, 2013), and participants may draw on this knowledge during the testing session (N. L. Nelson et al., 2018; Ruba et al., 2018). To address this problem, some studies have examined adults with various neurological deficiencies (e.g., Lindquist et al., 2014; Nook et al., 2015) or presented healthy adults with unfamiliar, nonhuman faces (e.g., Doyle & Lindquist, 2018; Fugate, Gouzoules, & Barrett, 2010). Although insightful, these studies cannot address the question of how language constructs emotion categorization with human facial expressions in typically developing populations. A potential solution is to study preverbal infants. To date, no published work has examined how language influences infants’ emotion categories.

Language and Object Categorization in Infancy

Nevertheless, over two decades of research has documented how labels influence object categorization in infancy (for a review, see Ferguson & Waxman, 2016). In a seminal study, Waxman and Markow (1995) familiarized 13-month-olds with four objects from either a basic-level category (e.g., cars) or a superordinate category (e.g., vehicles, including cars and airplanes). An experimenter presented and labeled each object with either a noun (“look, a car”) or no-noun (“look what’s here”). In the subsequent test phase, infants were shown two new objects: a novel object from the familiarized category (e.g., car) and a novel object from an unfamiliar category (e.g., horse). Infants formed a basic-level category (cars) regardless of whether a noun or no-noun was presented during familiarization. However, infants only formed a superordinate category (vehicles) when a noun was presented. Similar facilitative labeling effects have subsequently been found with basic-level categories (e.g., LaTourrette & Waxman, 2019), novel objects (e.g., Fulkerson & Haaf, 2006), and other object properties (e.g., spatial relationships; Casasola, Bhagwat, & Burke, 2009).

Waxman and Markow (1995) argue that labels are “invitations” to form categories. In fact, research has found that labels are unique in their ability to facilitate categorization, compared to other sounds, such as instrumental music (Roberts & Jacob, 1991), nonlinguistic tones (e.g., Althaus & Westermann, 2016), and nonsensical/backwards human speech (e.g., Ferry, Hespos, & Waxman, 2013). Infants at 12 months of age are also unable to form categories when inconsistent labels are used (i.e., each object is given a different label; Waxman & Braun, 2005). These findings suggest that labels do not facilitate categorization simply by heightening infants’ attention to objects (Waxman, 1999). Instead, labels appear to facilitate category formation by highlighting commonalities between objects (for alternative explanations, see Ferguson & Waxman, 2016). Recent eye tracking and EEG research has confirmed that, for 12-month-olds, labels (a) direct visual attention to perceptual commonalities (Althaus & Plunkett, 2016), and (b) increase neural activity over the visual cortex (Gliga, Volein, & Csibra, 2010). This suggests that labels impact infants’ visual processing of objects. With evidence from a connectionist model, Westermann and Mareschal (2014) further hypothesize that labels modify visual perception, so that objects from the same-labeled category are perceived as more similar to one another. This is congruent with findings that labels influence facial expression perception in adults (e.g., Brooks et al., 2017; Fugate et al., 2018). Thus, while similar processes between language and categorization are evident at multiple stages of development, it remains to be seen whether language also influences emotion categorization in infancy.

A New Hypothesis

Currently, a comprehensive developmental hypothesis describing infants’ understanding of emotional facial expressions does not exist. Based on the existent literature, we outline the following proposal. At birth, infants’ visual systems are likely not sufficiently mature to discriminate facial expressions (C. A. Nelson, 1987). However, by around 5 months of age, infants should be able to visually discriminate between all pairs of “basic” facial expressions, including within-valence pairs (e.g., Schwartz et al., 1985). Differences in neural activity to different facial expressions may also emerge at this time. With increased cognitive maturation at around 7 months of age, infants begin to form perceptual categories for these facial expressions. By the end of the first year of life, infants should be able to form perceptual categories for all pairs of “basic” facial expressions, including within-valence pairs (e.g., Ruba et al., 2017). Thus, we argue that infants develop the requisite perceptual and cognitive skills needed to discriminate and categorize “basic” facial expressions in the first year of life. However, the ability to discriminate and categorize these expressions (even at a neural level) does not require infants to attribute any affective meaning to these displays. We argue that discrimination and categorization tasks likely test perceptual abilities, rather than emotion concepts. Similarly, in the first year of life, infants should form intermodal matches between facial and vocal expressions, both across- and within-valence. These tasks likely test infants’ ability to detect regularities between facial and vocal expressions encountered in their environment, and also need not reflect any conceptual understanding of emotions.

In regard to infants’ emotion concepts, we argue that infants’ conceptual understanding of emotional facial expressions is initially broad, based on valence and arousal. This is congruent with constructionist theories (Barrett, 2017; Lindquist & Gendron, 2013) and research findings with preschoolers (Widen, 2013). It is unclear, however, when this broad understanding first emerges: it may be innately specified or gradually learned in the first year of life through observation of emotions in the infants’ environment. In contrast to constructionist theories—specifically, the theory of constructed emotion—we predict that in the second year of life, but before emotion labels are learned, this broad conceptual understanding of facial expressions gradually becomes more refined. Specifically, we argue that the acquisition of emotion labels is not necessary for infants to begin forming event–emotion matches for within-valence emotions (Ruba et al., 2019; Wu, Muentener, & Schulz, 2017). Further, it is possible that infants may begin to show differential functional responses to some within-valence emotions before emotion labels are learned (Walle et al., 2017).

Thus, in the second year of life, infants may understand something about the causes and functional behavioral responses for discrete “basic” emotions. Infants could learn about these components of emotion understanding through observation of emotions in their environment, without needing to be explicitly taught this information via language. However, it is very unlikely that preverbal infants have robust or fully formed discrete emotion concepts at this age. Further, this emerging and rudimentary understanding of discrete emotions is likely influenced by language. In other words, while emotion language may not be necessary for infants to discriminate, categorize, and match facial expressions to voices and events, or respond to others’ emotions, language may still play a constructive role in all of these abilities (e.g., language may change how emotions are categorized; Plunkett, Hu, & Cohen, 2008).

Overall, this hypothesis is unique from current proposals in that it (a) clearly differentiates between individual components of infants’ emotion understanding, while (b) emphasizing developmental change over the first 2 years of life. However, far more work is needed to empirically confirm this developmental sequence. Critically, future research should focus on within-valence emotion contrasts and infants at multiple ages. Currently, discrimination, categorization, and intermodal matching are primarily studied in the first year of life, while event–emotion matching and social referencing are studied in the second year of life. For this reason, it is unclear how these abilities emerge and change over time, particularly as older infants begin to learn emotion labels. In fact, no studies have examined how language influences emotion concept development in infancy, even though emotion labels (e.g., “happy”) begin to appear in infants’ productive vocabularies late in the second year of life (Bretherton, Fritz, Zahn-Waxler, & Ridgeway, 1986; Ridgeway et al., 1985). Finally, given that emotion concepts undergo developmental change throughout infancy and early childhood, it is important to document the learning mechanisms that account for these changes. While constructionist theories have largely advocated for language-dependent learning mechanisms, it is likely that language-independent learning mechanisms (e.g., observational and statistical learning; Plate, Wood, Woodard, & Pollak, 2019) also play an important role in emotion concept development.

Future research with preverbal infants has the potential to dramatically influence our understanding of human emotions. Infants are an ideal test case to isolate the relative roles of evolution, language, and social experience in the development of emotion understanding. Historically, developmental psychologists have not designed studies to test emotion theories, and affective scientists have not fully integrated infant studies into these theories. However, by connecting these two disciplines, researchers can turn towards collaborative projects to answer fundamental questions about the nature of human emotions.