Abstract
Ruba and Repacholi (2020) review an important debate in the emotion development literature: whether infants can perceive and understand facial configurations as instances of discrete emotion categories. Consistent with a psychological constructionist account (Lindquist & Gendron, 2013; Shablack & Lindquist, 2019), they conclude that infants can perceive valence on faces, but argue the evidence is far from clear that infants perceive and understand discrete emotions. Ruba and Repacholi outline a novel developmental trajectory of emotion perception and understanding in which early emotion concept learning may be language-independent. In this comment, we argue that language may play a role in emotion concept acquisition even prior to children’s ability to produce emotion labels. We look forward to future research addressing this hypothesis.
The question of whether infants understand emotions in terms of discrete concepts (e.g., anger vs. fear) or broader valence-based categories (e.g., pleasant vs. unpleasant) is a topic of lively debate in emotion research. We commend Ruba and Repacholi (2020) on their comprehensive review of this complex literature; it will be an excellent resource for scientists working at the intersection of developmental, cognitive, and affective science. We agree with Ruba and Repacholi’s conclusion that the existing evidence is far from clear that infants perceive and understand facial muscle movements in terms of discrete categories. We also agree with them that language likely plays a role in the development of emotion category perception and understanding (see Lindquist & Gendron, 2013; Shablack & Lindquist, 2019). Nonetheless, what remains in question is how language does so; almost no research has explicitly tested these hypotheses in infants. Ruba and Repacholi outline clear developmental hypotheses that would guide this work. However, we believe their proposed developmental hypothesis underestimates the role of language in the learning of emotion concepts in preverbal infants in three important ways (see also Hoemann, Xu, & Barrett, 2019; Shablack & Lindquist, 2019).
First, Ruba and Repacholi (2020) may underestimate the role of language in emotion concept learning by assuming that infants need to be able to produce emotion labels for language to have a role in concept acquisition. They predict that infants begin differentiating broad valence-based emotion categories into discrete emotion concepts before they learn emotion labels. This prediction is based on evidence that infants exhibit some emotion understanding prior to producing emotion labels themselves (Ruba, Johnson, Harris, & Wilbourn, 2017), yet we argue that lack of emotion label production does not necessarily indicate that infants have not used language to construct their emotion concepts. Indeed, infants can understand words before they produce them (Bergelson & Swingley, 2012; Gershkoff-Stowe & Hahn, 2013; Plunkett, Hu, & Cohen, 2008). We hypothesize that the early learning environment exposes infants to facial muscle movements, vocalizations, emotional situations, and emotion words, and that they can use the co-occurrence of these components to construct emotion concepts, even if they cannot yet produce emotion labels themselves (which typically occurs around age 2; Widen, 2013).
Second, Ruba and Repacholi (2020) argue that infants do not require “explicit [teaching] via language” (p. 245) to begin to develop differentiated emotion concepts and that this prediction contrasts with the theory of constructed emotion’s (TCE) emphasis on “language-dependent learning mechanisms” (p. 245). However, from a TCE perspective, language does not only support learning in explicit learning contexts; it is an inextricable element in implicit statistical learning. The TCE hypothesizes that across the lifespan, people acquire and update emotion concepts via statistical learning; language facilitates this process when instances of emotion are paired with emotion words in one’s natural environment (Atzil, Gao, Fradkin, & Barrett, 2018; Atzil & Gendron, 2017; Barrett, 2017; Hoemann et al., 2019; Lindquist, MacCormack, & Shablack, 2015). Preliminary findings in cognitive development support such a role for language in statistical learning more generally (Batterink, Reber, & Paller, 2015; Xu, 2002; Xu, Cote, & Baker, 2005; Xu & Kushnir, 2013), even when labels are sparse (LaTourrette & Waxman, 2019). Children even use the broader syntactic context of spoken language to learn the meaning of difficult to acquire, abstract predicates such as emotions (Fisher, Gleitman, & Gleitman, 1991; Gleitman, 1990; Naigles, 1996; Shablack, Becker, & Lindquist, in press). Thus, even the syntactic context of spoken language itself may guide attention to statistical regularities of a specific emotion concept.
Finally, Ruba and Repacholi (2020) suggest that language may facilitate concept learning by directing attention to perceptual similarities among different exemplars of emotion categories. However, this assumes that there is sufficient statistical regularity across different instances of an emotion category for perceptual categories to develop. Indeed, a common assumption in the emotion literature is that the types of highly caricatured, stereotypical stimuli used in experiments reflect the statistical features infants experience in the real world (see Shablack & Lindquist, 2019). However, emotions are conceptual categories united by abstract function(s) (e.g., signaling displeasure at violations of expectations) and are highly perceptually diverse in terms of the sensory features that constitute them (e.g., people make different faces across different instances of anger; Barrett, Adolphs, Marsella, Martinez, & Pollak, 2019; Durán, Reisenzein, & Fernández-Dols, 2017; Fernández-Dols & Crivelli, 2013). Although, in principle, statistical learning could proceed on the basis of sensory information alone (e.g., learning to pair a facial expression with a behavior or situation), this type of learning is more difficult, if not impossible, when sensory input is highly variable. We propose that words play a more significant role than that proposed by Ruba and Repacholi, acting as “essence placeholders” that cohere together the variable features associated with an emotion category under a single linguistic label. Words thus provide an additional source of statistical regularity that does not otherwise exist in the sensory input. This hypothesis is consistent with evidence that infants use words to unite perceptually distinct objects as members of the same functional category (Casasola, 2005; Dewar & Xu, 2009; Yin & Csibra, 2015; Yoshida & Smith, 2005). Therefore, we hypothesize that words such as “anger” help infants unite diverse features (e.g., a variety of causes and consequences, facial configurations, physiological responses) as members of the same emotion concept.
In sum, we agree with Ruba and Repacholi (2020) that future research must address the role of language in emotion concept learning during infancy using both valid methods and clear developmental hypotheses. We look forward to research that assesses the role of words in statistical learning of emotion concepts using naturalistic facial muscle movements and both explicit and implicit learning environments. This type of research will finally begin to weigh in on the constructionist hypothesis that words serve as “essence placeholders” and a source of statistical regularity for emotion concept acquisition, even before children produce emotion labels. In the meantime, we are excited to see discussions that address the development of emotion concept understanding—and are hopeful that collaborations emerge amongst developmental, cognitive, and affective scientists.
