How to Define Emotions Scientifically

Griffiths on Why Emotion Is Not a Natural Kind

Philosophers are responsible for bringing the language of natural kinds to bear on the question of whether emotion is a theoretically homogenous category for purposes of scientific extrapolation. Rorty (1980, p. 1), for instance, argued that “[e]motions do not form a natural class.” This idea has been most forcefully articulated by Griffiths (1997, 2004) in recent years. According to Griffiths, emotion is not a natural kind because it contains instances of at least three different types: basic emotions, higher cognitive emotions, and socially sustained pretenses. Let us consider them in turn.

According to Ekman (1999, p. 46), the most prominent defender of basic emotion theory, “emotions evolved for their adaptive value in dealing with fundamental life tasks” such as avoiding predators, progressing towards goals, coping with losses, finding mates, fighting, and so on (Ekman, 1999; Tooby & Cosmides, 2000). Paradigmatic examples of basic emotions comprise fear, anger, disgust, happiness, sadness, and surprise. Each of these basic emotions, Ekman (1999) emphasized, forms a family. Instances of irritation, rage, and fury, for instance, all belong to the anger family.

Basic emotions are defined by markers such as distinctive universal signals, distinctive physiology, autonomic appraisal tuned to distinctive universals in antecedent events, distinctive developmental appearance, distinctive thoughts, memories and images, distinctive subjective experience, quick onset, brief duration, unbidden occurrence, and presence in other primates (Ekman, 1999).

Griffiths argued that this account does not fit emotions such as envy, moral indignation, resentment and others, which seem to lack quick duration, are not present in other primates, and require “responding in a more cognitively complex way to more highly analyzed information” (2004, p. 236) than basic emotions do. Griffiths proposed to call such emotions higher cognitive emotions.

Finally, Ekman’s account does not fit emotions that “involve an internalized cultural model of appropriate behaviour.” This is the case for emotions such as “going postal,” which follow a script “derived from real or fictional incidents that are culturally salient” (Griffiths, 2004, p. 236). He proposed to call such emotions socially sustained pretenses.

Griffiths’ key point is that higher cognitive emotions and socially sustained pretenses differ from basic emotions so significantly that they cannot be comprised within one and the same scientific theory of emotions. ² Thus, emotion is not a natural kind. Griffiths’ argument has been criticized by a number of philosophers (Charland, 2002; Nussbaum, 2001; Prinz, 2004a, 2004b, 2004c), but their objections have, in my opinion, already been successfully rebutted by Griffiths himself (Griffiths, 2004).

In what follows, I want to develop a theme hinted at, but not fully developed, by Griffiths (1997, 2004), namely that what is true of emotion “is probably true of many specific emotion categories, such as anger and love” (Griffiths, 2004, p. 233). To be clear, the thesis that anger, love, fear, disgust, etcetera are not natural kinds does not follow from the thesis that emotion is not a natural kind. It may well be that, even though emotion is not sufficiently homogeneous to allow for extrapolation of discoveries about emotion to the whole emotion category, the traditional categories subordinate to it are sufficiently homogenous for scientific extrapolation. In this case, the heterogeneity of the emotion category would result from the fact that there are important differences between subordinate categories such as, say, anger and love—instances of anger are not sufficiently similar to instances of love to belong to one and the same scientific theory.

Barrett on Why Discrete Emotions Are Not Natural Kinds

My view is that subordinate categories such as anger, love, fear, disgust and so on are themselves not natural kinds. This thesis does entail that emotion is not a natural kind. If categories subordinate to emotions such as love and anger are each not sufficiently homogenous for scientific extrapolation, emotion will also not be sufficiently homogenous for scientific extrapolation. The heterogeneity of emotion in this case would result from the fact there are important differences both within and between subordinate categories—instances of anger are not sufficiently similar to other instances of anger or to instances of love to belong to one and the same scientific theory.

I begin from Barrett’s (2006) attack of what she calls the natural kind view of discrete emotions such as anger, fear, shame, and so on. Barrett focused her critique primarily on basic emotion theory and appraisal theory. “Unlike the basic-emotion approach,” Barrett (2006, p. 30) writes, “the appraisal approach does not assume … that objects or situations trigger prescribed emotional responses in an unmediated or reflexive way … [positing instead] that cognitive processes mediate emotion elicitation.” Generally speaking, appraisal models give pride of place to the complex evaluation of stimuli that initiates the cascade of emotional responses (see Scherer, Schorr, & Johnstone, 2001).

What basic emotion theory and appraisal theory have in common, Barrett argued, is that they are both “natural kind models”:

Natural kind models of emotion not only assume that there are distinct profiles of responses to characterize each kind of emotion, but they also assume that these responses are caused by distinct emotion mechanisms. The causal mechanism for anger is presumed responsible for the coordinated package or correlated set of features that constitute an anger response. (Barrett, 2006, p. 31)

More precisely, “natural kind models” take traditional discrete emotion categories such as anger, fear, disgust, etcetera to be associated with an emotion-specific response profile and/or an emotion-specific mechanism causally responsible for the response profile. A response profile, in turn, may comprise subjective experience, facial and vocal signals, peripheral nervous system responses, voluntary behaviors, and neural circuitry. These are the most common components that have appeared in proposed definitions of discrete emotions over the history of affective science.

Barrett’s (2006) key point is that there is no one-to-one correspondence between anger, fear, shame, etcetera and any response profile and/or emotion-specific mechanism. On the one hand, neither subjective experience, nor facial and vocal signals, nor peripheral nervous system responses, nor voluntary behaviors, nor neural circuitry appear to be distinctive of anger, or fear, or disgust and so on in the sense that they apply to all and only instances of what we call in English “anger,” “fear,” “shame,” etcetera. Whatever signature may be proposed for, say, anger at the phenomenological, physiological, expressive, behavioral, or neural level, we seem to be able to find instances of anger without that signature.

Furthermore, if we collect all instances of what we call in English “anger,” “fear,” “shame,” etcetera, we will not find among such instances any specific set of components that correlate to a high degree. If so, not only will there be no bodily signatures for anger, fear, shame, etcetera at the level of any single component, but neither will there be any signatures at the level of any single set of correlating components.

I have argued elsewhere that Barrett’s (2006) empirical data on lack of componential specificity and on lack of high correlation among components are convincing with respect to traditional emotion categories such as anger, fear, shame, etcetera (Scarantino, 2009; Scarantino & Griffiths, 2011). Before considering some possible responses to such data, I discuss a central disagreement with Barrett’s analysis of the failures of basic emotion theory and appraisal theory.

The Problem of Scope

According to Barrett, the trouble with basic emotion theory and appraisal theory lies in the assumption that “emotions have definable boundaries in the brain or body” (Barrett, 2006, p. 29). In Lindquist, Wager, Kober, Bliss-Moreau, & Barrett (in press, p. 5), it is explicitly argued that “[t]he natural kinds approach assumes that emotion categories like anger, sadness, fear, et cetera, map on to biological categories that are given by the brain and body.”

The implication is that scientific models of anger, sadness, and fear that do not put emphasis on brain or body would not encounter the problems faced by basic emotion theory and appraisal theory. This implication naturally leads to social constructionist theories, proposed as an antidote to the “tendency to treat emotions as biologically primitive, instinctive response patterns” (Averill, 1980b, p. 57). Barrett’s own conceptual act theory, the details of which I cannot consider in this article, proposes that “emotion categories are … socially constructed” (as cited in Lindquist et al., in press, p. 5).

I will now argue that the problems encountered by basic emotion theorists and appraisal theorists have nothing to do with the fact that they map traditional emotion categories to biological categories, and all to do with the fact that they map traditional emotion categories to theoretically homogeneous categories for purposes of scientific extrapolation. This can be shown by demonstrating that theories which search for the distinctive social signature of traditional emotion categories have as much trouble finding it as theories which search for the distinctive biological signature of traditional emotion categories.

Consider Averill’s social constructionist theory. It is worth considering not only because it is highly influential, but because it clearly aims for universal scope. In Averill’s (1980a) view, “[a]n emotion is a transitory social role (a socially constituted syndrome) that includes an individual’s appraisal of the situation, and is interpreted as a passion rather than as an action” (1980, p. 312). The interpretation of passivity, Averill suggested, results from a limited understanding of the social role played by emoting, which is “resolving conflicts which exist within the social system” (1980b, p. 37).

For example, Averill proposed that anger is a social role people take on in order to be justified in the exercise of aggression. This account seems to capture some forms of anger, for example the “wild pig” syndrome characteristic of the Gururumba tribe of New Guinea (Newman, 1964). The syndrome consists of a ritualized sequence of aggressive behaviors the locals interpret as resulting from being bitten by the spirit of someone who died.

As pointed out by Averill, this attribution manifests a limited insight into the social function played by the syndrome, namely giving the “victims” a means to communicate how overwhelmed they are by their social responsibilities. The syndrome generally occurs to newly married males who have just begun sharing communal responsibilities, and it leads the community of elders to reduce their share of social responsibilities.

What Averill’s theory certainly does not do is to apply to all instances of what we call “anger” in English. Consider the sort of anger in which infants engage when they are suddenly restrained. It has been detected in infants as early as their 5th month of life in a variety of cultures (Camras, Campos, Oster, Miyake, & Bradshaw, 1992), it is associated with characteristic facial expressions, it is automatic, it has quick onset, brief duration, unbidden occurrence, it exists in homologous form across species, and so on.

In other words, it fits Ekman’s (1999) definition of a basic emotion. In contrast, it does not fit Averill’s (1980a, b) own definition of anger, because there is no theoretically interesting sense in which the infant is playing a social role, let alone the specific social role Averill ascribes to anger. ³ The case of infant anger due to restraint is one of many to which Averill’s theory does not apply.

The inability to provide definitions that fit all instances of the target emotion is not specific to Averill’s formulation of social constructionism, nor to Ekman’s formulation of basic emotion theory. I have argued in Scarantino (2005) that all scientific theories which try to tell us what all emotions are like, or what all angers are like, end up facing the following:

Problem of scope: For every scientific theory T that tells us that an emotion/anger/fear/etc. is X, we can find counterexamples consisting of things called “emotion”/“anger”/“fear”/etcetera in English that are not X, and/or things not called “emotion”/“anger”/“fear”/etcetera in English that are X.

The problem of scope is what frustrates models that endorse the natural kind assumption (NKA). To reiterate, these are not models that map traditional emotion categories to biological categories (Barrett’s [2006] suggestion), but, more broadly, models that map traditional emotion categories to theoretically homo- genous categories, namely natural kinds in the sense I discussed earlier. I conclude that social constructionist models are as guilty of the natural kind assumption as basic emotion and appraisal models.

Ex Cathedra Strategy

One possible response to the charge that one’s favorite scientific definition of emotion/anger/fear/etcetera does not apply to all instances of the relevant traditional emotion categories is to stipulate that the contradictory data are not “really” about emotions/angers/fears/etcetera, but about some other affective phenomenon. I call this the ex cathedra strategy, because it consists of stipulating from a position of authority that what does not fit the proposed definition not only should not count as an emotion for the purposes of the theory (this would be a legitimate move), but is not an emotion simpliciter.

A conspicuous example of the ex cathedra strategy can be found in Ekman’s edict that “I do not allow for ‘non-basic’ emotions” (Ekman, 1999, p. 57). If we accept this edict, we can reject all empirical data in contradiction with basic emotion theory by stipulating that they do not concern emotions. Consider Barrett’s (2006) data calling into question the assumption that there are facial expressions distinctive of all instances of what we call “anger” in English, such as furrowing of the brow and tightening of the lips. If distinctive facial expressions are an essential part of the definition of a basic emotion, as Ekman has often suggested, and if basic emotions are the only emotions that there are, anything that does not have the requisite facial expressions cannot be a genuine instance of the emotion of anger. Therefore, it cannot be a counterexample to a scientific theory of anger. We may therefore dismiss the data on alleged instances of anger that do not involve furrowing of the brow and tightening of the lips as being about affective phenomena other than anger.

This is the strategy used by Ekman to account for traditional emotion categories of which no members fit the profile of a basic emotion, for instance grief and romantic love (in the case of anger, some members do fit such profile). According to Ekman, grief and romantic love are not emotions, but emotional plots. Emotional plots are “more specific” and “more enduring” than emotions (Ekman, 1999). The emotional plot of grief, for instance, lasts longer than the emotion of sadness, and it is “much more specific than sadness. We know that in grief a death has occurred, in sadness we only know that the person has suffered an important loss, but not what kind of loss” (Ekman, 1992, p. 194).

This characterization of emotional plots strikes me as unable to capture any relevant distinction between grief and sadness, but this is not the key issue. The key issue is that the ex cathedra strategy is a bad way to make a good point, namely that there are important theoretical differences between members of the same traditional emotion category such as emotion or anger. If we assume that a good theory of emotion/anger/fear/etcetera must apply to all emotions/all angers/all fears/etcetera, we are forced to respond to seeming members of such categories that do not fit one’s definition by expelling them ex cathedra from membership.

The reason why this is a bad way to acknowledge the presence of theoretical differences is that it insulates the theory from the possibility of falsification. Empirical data contradicting one’s theory of emotion/anger/fear/etcetera should not be rejected because they do not fit the theory, lest the theory becomes true by stipulation. As we shall see, a better way to acknowledge theoretical differences is to give up on the assumption that a good theory of emotion/anger/fear/etcetera should apply to all emotions/all angers/all fears/etcetera.

Hidden Unity Strategy

A second possible response to the charge that one’s favorite scientific definition of emotion/anger/fear/shame/etcetera does not apply to all instances of the relevant traditional emotion categories is to acknowledge the presence of differences, but to argue that, over and above such differences, emotion/anger/fear/shame/etcetera are still theoretically homogeneous categories for purpose of scientific extrapolation.

I call this the hidden unity strategy, because it consists of finding some nonobvious source of unity among instances of the same traditional emotion category, while acknowledging that there are important theoretical differences among them. Here, I consider one example of this strategy at work, namely Izard’s (2007) defense of basic emotion theory from Barrett’s (2006) attack.

Izard proposed that there are two main kinds of emotions: basic emotions and emotion schemas. Basic emotions are “affective processes generated by evolutionarily old brain systems upon the sensing of an ecologically valid stimulus” (Izard, 2009, p. 7). Emotion schemas, on the other hand, consist of “emotion interacting dynamically with perceptual and cognitive processes to influence mind and behavior” (Izard, 2009, p. 8).

A key difference with Ekman’s account is that emotion schemas, unlike emotional plots, are acknowledged to be emotions. In other words, Izard allows for nonbasic emotions. His view is that “[t]he cumulative evidence suggests that the following basic emotions meet criteria for classification as natural kinds: interest, joy/happiness, sadness, anger, disgust, and fear” (Izard, 2007, p. 261). At the same time, he states that “emotion schemas are not natural kinds because they have properties that differ across individuals and cultures” (Izard, 2007, p. 261).

This last passage makes clear that the notion of natural kind Izard presupposes differs from the one I relied on in this article. In Izard’s view, natural kinds are categories whose instances do not have properties that differ across individuals and cultures (call this the social invariance sense of “natural kinds”). In my view, natural kinds are theoretically homogenous categories for purposes of scientific extrapolation (call this the inductive sense of “natural kinds”).

These two notions of natural kinds are orthogonal to one another (Scarantino, in press). For instance, a category may be nonnatural in the social invariance sense because its members have properties that differ across cultures, but natural in the inductive sense because its members participate in the same body of empirically discovered reliable generalizations (e.g., generalizations about cultural variation). Since this article focuses on whether traditional emotion categories are theoretically homogeneous, I will disregard the social invariance sense of natural kinds in what follows.

Relative to the inductive sense of natural kinds, both basic emotions and emotion schemas are natural kinds by Izard’s own lights. This is because they both map on to theoretically homogeneous categories for purposes of scientific extrapolation. Izard himself proposes a number of empirically-based generalizations that are meant to apply to, respectively, basic emotions and emotion schemas.

The problem is that some of the instances of Izard’s basic emotions—interest, joy/happiness, sadness, anger, disgust, and fear—satisfy the generalizations characteristic of emotion schemas. Izard argues that there is evidence for the “universality of expressions of a limited set of basic emotions,” for “their recognition via processing in evolutionarily old brain stem and amygdala systems,” and for the fact that basic emotions do “not depend on or include complex appraisals or higher order cognition” (Izard, 2007, p. 262).

This would not do as a response to Barrett (2006). There are instances of interest, joy/happiness, sadness, anger, disgust, and fear—the alleged basic emotions—that lack distinctive expressions, are not recognized via amygdala systems, and depend on higher order cognition (Barrett, 2006; Lindquist et al., in press). These instances fully qualify as emotion schemas.

Izard admits that when we speak of, say, “anger” in English we can refer either to a basic instance—for example, “anger to explain the frustration of blocked goal responses” (2007, p. 267)—or to a schema instance, for example, “a pattern of … anger schemas to explain reactions to the terrorist disaster of 9/11/2001” (2007, p. 267). This would seem to force Izard into accepting the conclusion that traditional emotion categories are not theoretically homogeneous categories because they contain emotions of different types.

It is at this juncture that Izard engages in the hidden unity strategy. His proposal to preserve unity is to say that basic emotions and emotion schemas, different though they may be, have evolved feelings at their motivational core. On this view, the trademark thesis that all emotions are biological adaptations can be upheld, because “all emotion feelings, whether they are a component of a basic or nonbasic emotion, are products of evolution” (Izard, 2007, p. 265). “Although there are important differences at the cognitive and action levels between an anger schema and a basic anger episode,” Izard concludes, “the quality of the anger feelings is the same” (2007, p. 265).

But why should we believe that? The quality of the feeling associated with a given anger episode is determined by what it is like to experience such an episode. This in turn depends on which specific emotion components are involved in it. Since basic anger and schema anger are instantiated by different components (e.g., basic anger involves facial changes, schema anger doesn’t; basic anger involves automatic appraisal, schema anger doesn’t; etc.), there is no reason why the quality of the anger feeling should be the same, except in the trivial sense that it will be a feeling of anger in both cases.

But let us assume for the sake of argument that emotion feelings have the exact same experiential quality in all cases in which a discrete emotion exists in both basic and schema forms. Still, there are emotions—for example, shadenfreude (joy at someone else’s misfortune)—which only seem to exist in the form of emotion schemas. How would evolved feelings manage to get to their motivational core? One could argue that all emotion schemas lacking a basic version come from a combination of different basic emotions, the way cocktails result from a combination of spirits (Prinz, 2004b). If that were true, a combination of evolved feelings may still be at the motivational core of shadenfreude. This combinatorial project, however, faces a number of serious difficulties, and it is far from clear that they can be overcome (Scarantino & Griffiths, 2011).

I will mention a final problem, namely that there is at least preliminary evidence that some emotions do not involve feelings at all because they are unconscious (Berridge & Winkielman, 2003). These results are admittedly controversial, but they put further pressure on the thesis that all emotions have an evolved feeling at their core.

I conclude that even Izard’s (2007) sophisticated version of basic emotions theory faces the problem of scope: There are things we call “emotion” or “anger” in English which are not at their core biological adaptations, not even in the weak sense that they have evolved feelings at their motivational core.

One case study is clearly not enough to establish that the hidden unity strategy does not work. In Scarantino (2005), I have considered many more attempts to find hidden unity in traditional emotion categories, and argued that they all fail in one of two ways. They either do not detect a genuine way in which all emotions/angers/fears are alike (Izard’s case), or they detect a genuine way in which they are alike but one that fails to be relevant to the predictive and explanatory purposes of affective science.

My interpretation of this collective failure is that there is no unity to be found within traditional emotion categories. Although we cannot be positively sure that traditional emotion categories are not theoretically homogeneous for the purpose of scientific extrapolation, we have good reason to interpret the long history of failure in finding a single scientific theory that fits all cases as a sign that no single scientific theory can fit all cases. In other words, we have good reason to infer from the failures of traditional emotion theory that emotion, anger, fear and so on are not natural kinds (Scarantino, 2005).

Primitivism versus Pluralism

If the classificatory scheme of traditional emotion theory is no longer suitable to the needs of affective science, what should replace it? Two main proposals for conceptual renewal should be considered:

Primitivism: Infer from the failures of traditional emotion theory that there are no natural kinds of emotion/anger/fear/etcetera, and search for natural kinds at the level of primitive components of discrete emotions.

Pluralism: Do not infer from the failures of traditional emotion theory that there are no natural kinds of emotion/anger/fear/etcetera, and search for natural kinds at the level of discrete, but nontraditional, emotion categories.

Primitivism is the position advocated by psychological constructionists, and pluralism is the position I wish to defend. According to primitivism, there are no natural kinds of emotion/anger/fear/etcetera and the search for natural kinds should focus exclusively on emotion components that are psychological primitives “elemental—but still psychological—building blocks” (Russell, 2003, p. 146). Psychological constructionists are convinced that “[t]he goal of psychology is to identify … the most basic psychological descriptions that cannot be further reduced to anything else mental (because at that point they would describe biological mechanisms)” (Lindquist et al., in press, p. 4).

A prime candidate for psychological primitivity is core affect, a blend of hedonic and arousal values (Russell, 2003). Core affect need not be directed at anything, but it is a building block of psychological events such as discrete emotions, which are instead directed at something (at least generally). For example, a fear episode caused by a bear in the woods will include a feeling of high arousal and high displeasure directed at the bear. Barrett states that “the empirical case supporting the hypothesis that core affect is a natural kind is suggestive” (2006, p. 48), even though she admits that further empirical investigation is needed to close the case.

I support the search for natural kinds at the level of emotion components. What I reject is the inference that, since emotion, anger, fear and so on are not natural kinds, there must not be any natural kinds of emotion/anger/fear/etcetera. We can appreciate that this is a bad inference by considering the case of memory. What we call “memory” in English is not a theoretically homogeneous category for purposes of scientific extrapolation. Yet it clearly does not follow that there are no natural kinds of memory.

Long-term memory and short-term memory, just to pick two examples, are promising candidates for natural-kind status. They each participate in a large body of empirically discovered reliable generalizations on account of the many scientifically interesting properties they have in common. For example, lots of reliable generalizations have been discovered about short-term memory, concerning its dissociation from long-term memory and other forms of memory, its neural and chemical underpinnings, its maximum capacity, its information storage modality, and so on.

By the same token, the fact that the traditional categories of emotion/anger/etcetera are not natural kinds is perfectly compatible with the existence of natural kinds of emotion/anger/etcetera. As I argue below, these will have to be categories that bear some similarity with traditional emotion categories, but that differ from them in that they constitute homogenous groupings of emotion instances that participate in a large body of empirically discovered reliable generalizations.

Another reason why psychological constructionists infer that there are no natural kinds of emotion/anger/etcetera may be the implicit conviction that natural kinds exist only at an elemental level of analysis. If so, we won’t be able to individuate natural kinds of emotions, not because traditional emotion categories are not natural kinds, but because all emotions have more primitive building blocks (e.g., core affect).

I see no reason why there should not be natural kinds at both lower and higher levels of analysis. An analogy may help. Water is a natural kind for chemistry. Yet water is a chemical compound constituted by hydrogen and oxygen molecules, so it has other chemical substances as components. Furthermore, the atoms of the basic chemical elements are themselves composed of more basic parts, namely protons, neutrons, and electrons. So neither chemical compounds like water nor chemical elements like oxygen and hydrogen are without components, but they both fully qualify as natural kinds for chemistry.

The take-home message here is that not having constitutive components is not a necessary condition for being a natural kind. Consequently, although I consider the project of psychological constructionists well worth pursuing, I reject the idea that only psychological primitives can be natural kinds, and conclude that affective science should engage in an open search for natural kinds, at both elemental and nonelemental levels of analysis.

Folk Emotion Project versus Scientific Emotion Project

It is time to provide some details on how the pluralistic search for natural kinds at the level of nontraditional discrete emotion categories should be carried out. The first order of business is to distinguish between two projects that have been systematically confused in the history of emotion science.

One is the Folk Emotion Project, which aims to offer a descriptive definition of the conditions of membership of traditional emotion categories such as emotion, anger, and so on. The other is the Scientific Emotion Project, which aims to offer a prescriptive definition of the conditions of membership of natural kinds of emotion, natural kinds of anger, and so on. ⁴ Whereas the Folk Emotion Project has the accurate reconstruction of the boundaries of traditional emotion categories as its primary objective, the Scientific Emotion Project has the transformation of such categories into useful scientific tools as its primary objective.

Emotion theorists have not distinguished between these two projects because they have implicitly assumed that one and the same scientific definition can simultaneously individuate a natural kind of emotion/anger/fear/etcetera and account for all empirical data about what we call in English “emotion,” “anger,” “fear,” etcetera. To be pulled out, this achievement would require that traditional emotion categories designate natural kinds. But I have argued that we have good reason to conclude that they do not.

If so, two important consequences follow. First, no single scientific definition can be both descriptively and prescriptively adequate. Second, affective scientists must choose whether to engage in the Scientific Emotion Project—formulating good prescriptive definitions—or whether to engage in the Folk Emotion Project, formulating good descriptive definitions. If a definition aims to be descriptively adequate, we should ask: Does it account for all empirical data about what we call in English “emotion,” “anger,” etcetera? If the definition aims to be prescriptively adequate, we should ask: Does it individuate a natural kind of emotion, or a natural kind of anger? What we should stop asking of any scientific definition is the achievement of both objectives at the same time.

I won’t say much about the Folk Emotion Project, except to note that Fehr and Russell’s (1984) prototype proposal goes in the right direction. Since traditional emotion categories are unlikely to belong to the restricted class of classically definable categories, family resemblance accounts in the style of Fehr and Russell’s strike me as promising candidates to account for all empirical data about what we call in English “emotion”/“anger”/“fear”/etcetera.

But, as pointed out by Widen and Russell (2010, p. 378), “a [good] descriptive definition does not provide a good prescriptive definition.” The reason should now be clear: If traditional emotion categories are not natural kinds, what tells us what makes something a member of a traditional emotion category does not tell us what makes something a member of a natural kind.

What I want to focus on in the rest of my article is the Scientific Emotion Project, namely the project of discovering natural kinds of emotions.

Essentialism versus Antiessentialism

So far I have worked with a thin notion of natural kinds, understood simply as categories that are homogeneous for purposes of scientific extrapolation. This general idea has been developed in two main ways in philosophy. According to essentialist accounts, the members of a natural kind share an essence—a set of individually necessary and jointly sufficient properties. For example, all samples of water have the same chemical constitution (H₂O), and this is what makes them samples of water. In this view, natural kinds have sharp edges and feature in exceptionless laws of nature that hold uniformly across time and space.

This essentialist notion applies to natural kinds in chemistry and physics, but it is an unsuitable theory of natural kinds for the biological and social sciences (Boyd, 1999; Machery, 2005; Samuels, 2009; Wilson, Barker, & Brigandt, 2007). In such disciplines, variability among kind members is the norm, borderline cases often emerge, and generalizations tend to be exception-ridden and only locally valid.

Biological species such as tiger are a case in point. There is a great deal of variation among species members. Furthermore, species are products of evolution and as such generalizations about them cannot be unrestrictedly valid across space and time. But despite this, classification of organisms into species is of proven utility in biology, and reflects a real division in nature, rather than being solely an expression of human interests.

To accommodate variation, borderline cases, and the lack of exceptionless and universal generalizations in the special sciences, antiessentialist accounts of natural kinds have been proposed. The most influential is Boyd’s homeostatic property cluster (HPC) theory (Boyd, 1991), according to which “[t]he natural definition of … homeostatic property clusters kinds is determined by the members of a cluster of often co-occurring properties and by the (‘homeostatic’) mechanisms that bring about their co-occurrence” (Boyd, 1991, p. 141).

In the HPC view, tiger would be defined by a cluster of genetic, morphological, physiological, and behavioral properties that species members tend to reliably share on account of causal mechanisms such as interbreeding, being exposed to common selection pressures, and sharing ancestors. This definition is compatible with the absence of individually necessary and jointly sufficient properties for tigerhood, and with the presence of individuals that have just enough of the cluster properties to place them on the borderline for membership.

A final requirement for being a natural kind, on both the essentialist and the antiessentialist approach, is being a maximal class of items that share an essence or a cluster of causally correlated properties. For instance, consider the proposal that tiger living in the Western hemisphere is a natural kind. The reason why we would reject the proposal is not that tigers living in the Western hemisphere do not share a cluster of properties on account of causal mechanisms, but that there is a larger class—the class tiger—whose members tend to share the exact same cluster of causally correlated properties. The problem with tiger living in the Western hemisphere is that it is a subset of a larger class of items that share the same cluster of causally sustained properties.

In what follows, I presuppose Boyd’s HPC account of natural kinds. I have argued elsewhere that affective scientists should explicitly embrace an antiessentialist theory of natural kinds (Scarantino, in press; Scarantino & Griffiths, 2011). This would allow them to acknowledge the presence of variability among members of affective kinds, of borderline cases, and of exceptions to generalizations, while preserving the idea that some kinds are maximally suitable for induction and explanation in affective science.