Introduction

The near consensus among theorists working on emotion, in both philosophy and psychology, is that emotions have objects. When we are angry, we are angry at someone. When we are frightened, we are frightened of something. We can call the theory that emotions have particular objects the Token Object Theory (TOKEN), though the defenders of this view are so numerous and diverse that they agree on little else. Adherents of TOKEN rarely treat it as a theoretical position in need of defense, with empirically falsifiable implications.

In fact, some relevant empirical work has been done. I will argue that this research supports an alternative model, the Type Influence Theory (TYPE), according to which emotions lack particular objects. Fear, for example, influences judgments about anything that seems dangerous, without any special status for whatever the subject is “really” afraid of. TYPE proposes that emotions are, with respect to intentionality, no different from moods. The conclusion of the arguments in this paper will remain neutral about whether emotions are directed at general features of the world, such as danger or offensiveness, or whether they lack direction altogether.Footnote 1

I will focus exclusively on the class of emotions that psychologists such as Ekman (1992) refer to as basic emotions, such as happiness, anger and fear. Even if my arguments are successful regarding these emotions, there would still remain a further question about whether other emotions, such as shame and guilt, are amenable to the same analysis. Ekman (1992) argues that all emotions can be understood as variations on the basic kinds, and I am sympathetic to that view, but I will not pursue it here.

My proposal is so far outside the mainstream that readers familiar with the emotions literature might wonder how it could possibly be right. After all, much of the philosophical discussion of emotions, including the treatment of key features such as normativity and motivation, is based on the assumption that emotions have objects. I cannot hope to reproduce here a full alternative to the existing emotions literature. However, I will briefly reply to a number of challenges that may initially appear to be fatal. I hope to show that TYPE presents a robust alternative to prevailing views.

In Part 2 I will introduce a criterion for determining whether emotions have objects. In Part 3 I will discuss in detail a psychological study that forms the basis for my argument against TOKEN, and in Part 4 I will explain why this study supports TYPE over TOKEN. In the remainder of the paper I will consider a variety of objections: in Part 5 I will address our ordinary sense that emotions are directed, and in Part 6 I will discuss objections based on the semantics of emotion terms and the functions of emotions.

The criterion

Before discussing empirical results it is useful to introduce a criterion that will guide their interpretation. Here is my proposal: if emotions have particular objects, then they will influence experimental participants’ behavior in ways that are particular to those objects. On the contrary, if emotions influence participants in ways that are the same for all objects, or all objects of the same type, then those states do not have particular objects.Footnote 2 This criterion is derived from a conventional argument in favor of taking any mental states to have objects (Fodor 1987). On a daily basis we attribute beliefs and desires to our peers. But merely knowing that Jane believes tells us nothing about what Jane will do. We need to know what Jane believes. Knowing that Fred desires tells us nothing about what Fred will do. We need to know what Fred desires.Footnote 3 The same goes for emotions. If they have objects, predicting the influence of an emotion on an individual’s behavior should require correctly attributing an object to that emotion.

It is not important for these purposes whether the purported intentional object is a particular, a property of a particular, an event, a state of affairs, or a proposition. If Will is afraid of a bear, and the object of his fear is a particular bear, then Will’s fear should influence him in ways that are specific to that bear, rather than other particulars. If the object of his fear is the fierceness of the bear, then it should influence him in ways that pertain specifically to that property of that bear. If, instead, it is more accurate to say that he is afraid that the bear will harm him, where the object of the fear is the proposition ‘that the bear will harm him’, then the state of fear should influence Will’s behavior in ways that depend on the specific content of that proposition. That includes, at a minimum, influencing Will’s response to that bear. It is unwieldy to repeatedly address these different variations of TOKEN, so in places I will write as though I am targeting one or another. However, my arguments should apply equally to theories that attribute any sort of intentional object.

Of course in our ordinary lives we do often attribute objects to our emotions and to the emotions of others. Typically we take an emotion to be directed at an object when a perception, thought, or imaginative act with that intentional object was responsible for eliciting the emotion. But it is one thing to say that a person made you angry, and another to say that your anger is in any meaningful sense directed at that person. Again, a comparison with belief should clarify this point. Suppose Sam believes that there is a pen on her desk. This belief is about that pen not merely because it was caused by perceiving the pen, but because it will play a special role in pen-related thought or behavior she may engage in. If the belief influenced her in ways that were perfectly neutral between pen and pencil-related thought or behavior, we would have no basis for taking her to believe that there is a pen, rather than a writing implement or some other more general type of object. The conventional view seems to be that emotions, like beliefs and desires, play a special role in determining how individuals in such states will relate to their intentional object. I will argue that emotions do not have this sort of special role.

This criterion might seem flawed for requiring that emotions manifest their object-directedness in behavior. If an emotion influences other mental states in a way that is token-specific but not apparent from observing the individual’s behavior, that is clearly enough reason to consider the emotion to have an object. Any token-specific effects provide evidence that a mental state is object-directed. However, if emotions do have token-specific effects on other mental states, then at least some of the time this should be observable through subsequent influence on behavior. If we cannot find any evidence that emotions have any token-specific influence on behavior, then we have no reason to assert that they have any token-specific influence on other mental states either.

Some critics might claim that behavior aside, criteria based on phenomenology or normativity are sufficient in themselves for positing intentional states. Behavioral criteria for intentionality play an essential role in my argument, so I’ll make two points in response to this challenge before proceeding. First, behaviorial criteria for intentionality make a lot of sense if you assume naturalism, and you assume that our faculty for entering into intentional states arose via natural selection.Footnote 4 If intentionality is an adaptation then we should expect it to influence behavior, so a behavioral criterion is appropriate.

Second, criteria for intentionality aside, prototypical intentional states such as beliefs and desires do tend to influence thought and behavior in ways that are specific to their objects. Otherwise, folk-psychological explanations would not be effective for predicting behavior. If it turns out that emotions are not like beliefs and desires in that respect, that would itself be a surprising discovery. It would certainly put pressure on views that identify emotions as judgments or perceptions, since judgments and perceptions do normally have a systematic, content-driven influence on behavior.Footnote 5

Discounting and correcting

The psychological work on misattributed emotions is fairly well-known among psychologists and among philosophers interested in psychology. Using the criterion presented above, I will investigate how this literature bears on the question of whether emotions have objects. I will begin by describing the Mood as Information hypothesis, which was developed in response to research done in the early 1980s, before moving on to a study by DeSteno et al. (2000) that gives us reason to revise that hypothesis. This study, interpreted with my criterion, supports TYPE over TOKEN.

Schwarz and Clore (1983) performed a pair of studies on the effects of mood on judgment. Based on these experiments and others like them, they developed the Mood as Information hypothesis (INFORMATION).Footnote 6 According to INFORMATION, we take our moods to be informative about the world. Positive moods alert us to positive features, and negative moods to negative features of our environment. For example, Schwarz and Clore (1983) found that bad weather produced negative moods, which in turn influenced judgments about general well-being. However, we are somewhat selective about how we use that information. We generally assume that our current mood is informative about whatever we are currently considering, but if there is evidence to the contrary we discount the mood in our deliberation. In that same study, when participants were alerted to the effects of weather on mood, their judgments of well-being did not reflect any mood-related bias.

It is worth emphasizing at this point that INFORMATION is explicitly a theory of moods. The idea is that mood-related affective phenomenology does not provide the subject with any information about the identity of the mood’s object, it only provides information about the character of that object. In a later commentary Schwarz and Clore (2003, p. 300) say that emotions, despite their similar phenomenology to moods, are not as prone to these effects because they do carry with them information about their objects. Emotions only produce misattributions via lingering moods that no longer carry information about their original objects. I will ultimately argue that this view of the emotion-mood distinction is not supported by the evidence. However, since it is important not to beg any questions, I will use the term ‘affective state’ to refer to either emotions or moods when there is doubt about the nature of a state.

DeSteno et al. (2000) defend an alternative to INFORMATION. According to their proposal (CORRECTION), when participants make judgments about a given object, affective states automatically influence that judgment. Participants then consider whether the object they are judging is the intentional object of their affect. If not, they adjust their judgment about the object to correct for the influence of the affect, which in such cases is seen as a harmful bias.

The results of the processes posited by INFORMATION and CORRECTION are similar. Affective states will only affect a participant’s considered judgment when the participant believes that the affective state is about the same token object as the object of that judgment. A considered judgement, for these purposes, is the judgement that results after any discounting or correcting has been done. The difference between the theories concerns how participants reach those considered judgments. According to CORRECTION, affective states have an automatic influence on judgment, and that influence occurs prior to participants comparing the objects of their affect and judgment. INFORMATION, on the contrary, holds that affective states have no influence on judgment unless participants decide that they should, and this influence only occurs after they decide that the intentional objects match.

The fourth study from DeSteno et al. (2000) is designed to distinguish between these views. The experiment is based on the realization that emotion regulation requires work, and INFORMATION and CORRECTION differ over what sort of work participants need to do in order to regulate their emotions. There is a personality scale, Need for Cognition (NFC), which measures an individual’s willingness or unwillingness to do cognitive work. By observing the difference between participants with high (cognitively industrious) and low (cognitively lazy) levels of NFC, they can test the two models.

First, the experimenters induced anger, sadness or happiness by having participants describe events in their lives when they had strong emotions of that type. Second, participants were given a questionnaire to test the effectiveness of the emotion-elicitation. Finally, they were asked how likely certain types of events were. For example, they might be asked, ‘Of the 50,000 students currently enrolled at Ohio State, how many will experience the death of a loved one (e.g., close relative, close friend) within the next year?’ or ‘Of the 20,000 people who will drive on U.S. highways today, how many will be sharply cutoff by another driver?’ (p. 415). Depending on how participants correct or discount, we would expect sadness to increase estimates of the likelihood of losses, anger of offenses, and happiness of benefits. These are considered ‘affect-congruent’ effects: effects in which emotions (e.g. sadness) inflate estimates of the type of event that tends to elicit that kind of emotion (e.g. losses). All of the event-types from the third stage of the experiment involved losses, offenses, or benefits, so it was possible to compare the prevalence of these effects in high and low NFC participants.

Both INFORMATION and CORRECTION, as described above, predict that affective states will only influence considered judgments when the participants think that their affect is about the subject matter of their judgments. Via discounting or correcting, participants should avoid affect-congruent effects when their affective state is clearly not directed at the object of their judgment. The stages of the experiment were sequenced in such a way as to make it perfectly clear to participants that they were made angry, sad or happy by thinking about the autobiographical events that they described, and that these emotions were unrelated to the estimates in the third stage of the study, so a naive version of either theory would predict that no emotion-congruent effects would be observed.

This is where cognitive effort becomes relevant. DeSteno and colleagues (p. 408) predicted, on the basis of CORRECTION, that participants with low NFC would demonstrate affect-congruent effects. CORRECTION states that affective states automatically bias judgements, which are then corrected afterwards when necessary. Lazy, low NFC participants will not correct consistently and thoroughly, so their judgments should still reflect affect-congruent biases. That is exactly what they found.

This alone could also be compatible with INFORMATION. The experiment was designed to provide participants with clear information about the origin of their affective states. However, participants who are cognitively lazy might ignore that information, and have a policy of assuming that affective states are directed at whatever they are currently considering. They would rarely bother to discount, and as a result, demonstrate affect-congruent effects as observed.

The more striking results concern participants with high NFC. In general, the cognitively industrious actually over-corrected. They gave lower, rather than higher, estimates of the likelihood of affect-congruent events (p. 410). It is perfectly clear why, on the CORRECTION view, this would happen. First of all, according to CORRECTION, affect influences judgments of affect-congruent events automatically, and does so prior to any regulation by the participants. When participants decide that this influence is a harmful bias and decide to correct for it, there is no precise way for them to tell how strong a correction is required. Apparently high NFC participants are prone to over-correct in that type of situation. This contrasts with low NFC participants, who tend to under-correct or fail to correct at all.

While defenders of INFORMATION can easily explain affect-congruent biases in low NFC participants, they do not seem to have any way to account for affect-incongruent biases in high NFC participants. According to INFORMATION, the way that you avoid bias is by selectively discounting affect when making judgments. Perfect bias avoidance in this task would be reflected in a lack of any correlation between participants’ current affective state and their performance in the task. Good but imperfect bias avoidance would be reflected in a small emotion-congruent bias. Only CORRECTION can explain affect-incongruent biases.

Emotions and moods

When I introduced INFORMATION I mentioned that its defenders distinguish between emotions, which inform subjects about their particular objects, and moods, which do not. The theory predicts that misattribution should primarily occur, therefore, with moods rather than emotions. This is reflected in the choice by Schwarz and Clore (1983) to design experiments which produce states which seem intuitively to be moods rather than emotions. It would be possible to defend a version of CORRECTION while maintaining a similar view of the emotion-mood distinction. On such a view, only moods would automatically have a type-specific influence on judgment, while emotions would either have an automatic token-specific influence (with no need for subsequent correction), or they would not have any automatic effect at all.

However, Desteno et al. (2000) explicitly designed their experiment to induce the emotions of happiness, sadness, and anger, and these emotions are responsible for the subsequent misattribution effects. According to the version of CORRECTION that is supported by that research, emotions themselves have a merely type-specific influence on judgment, so the theory supports TYPE over TOKEN. The correction, which is distinct from the emotion itself, is entirely responsible for sometimes producing a net token-specific effect on judgment. There are two factors which influence the correction. The first factor is epistemic. If we do not know how our emotion originated, we will not know when it is appropriate to correct its influence.Footnote 7 Most misattribution literature capitalizes on this factor by obscuring the origin of the emotion. The second factor is motivational. We have to care enough to do the cognitive work required to correct.

What does CORRECTION have to say about the emotion-mood distinction? First of all, the notion that emotions but not moods have particular objects results from conflating the correction process with the emotion, and taking epistemic/motivation factors to be intrinsic to the state. Sometimes epistemic/motivational circumstances are such that the correction will produce what appears to be an emotional response targeted at a particular object, and in those cases we take the affective state to be an emotion. Other times, the epistemic/motivational features are not so cooperative, so we call the affective state a mood. However, those circumstance directly affect the correction, and only indirectly affect emotional responses. There may be some legitimate reason for preserving the emotion-mood distinction, perhaps related to the time-course of the state. But interpreting that distinction in terms of intentionality is not consistent with CORRECTION.

Primary effects and side effects

Some critics might accept my interpretation that participants in the study developed emotions, not merely moods. They might also accept that emotions have type-specific effects, as this study demonstrates. However, these critics might claim that this is only a side effect. Emotions, the argument goes, primarily have token-specific effects, but also have some type-specific side effects. In this study only the side effects are manifest, but that may be due to the peculiarity of the design.

This objection may be framed in a number of ways. In another version, the emotion itself is targeted at the object, and therefore has token-specific effects, but it is accompanied by a mood, which is untargeted. My mistake, on this view, is focusing on the indiscriminate effects of the mood, and losing sight of the emotion itself. I consider this to be equivalent to the primary effect/side effect objection, since it posits distinct mental processes responsible for type-specific and token-specific effects.

This may be the most persuasive defense of TOKEN, and it is difficult to decisively refute. My critics and I agree that there is explanatory value in positing mental processes responsible for type-specific effects on judgment and behavior. In my view these processes are emotions, while my critics either view them as moods or as aspects of emotions. The question is whether we have any reason to posit an emotional process that produces token-specific effects, in addition to the afore-mentioned type-specific effects. I have argued that the type-specific effects, plus correction processes which are due to distinct states, do all of the explanatory work.

I am ultimately making a parsimony argument.Footnote 8 I say that there are two things: a type-specific emotional process and a correction process. Defenders of TOKEN say there are three: the two that I posit plus a token-specific process. If we can explain all of the phenomena with a subset of the entities, then we have no reason to posit any others. The most productive way to criticize this kind of argument is to propose phenomena that I cannot explain. So far I have focused on one psychological study. The goal of the rest of the paper is to argue that my view does in fact have the power to explain normal emotional behavior.

Explaining appearances

In the preceding sections I have presented empirical reasons for denying TOKEN. However, in our daily lives we frequently attribute objects to our emotions. If I cannot explain why we make this error, critics could fairly suggest that it is not an error after all. I will argue that we ordinarily attribute emotional objects for a wide range of reasons. In developing my explanation I will appeal to four contributing factors: the conflation of occurrent and dispositional states, the salience of emotion-elicitors, the need to comply with social norms, and the utility of selective suppression more generally.

The first source of error is that we often conflate occurrent and dispositional states of all kinds, including emotions. Dispositional fear is not, strictly speaking, fear. It is just a disposition to enter into a state of fear. The systems that elicit emotions in different people respond to different things. My fear-eliciting system is highly sensitive to the perception or vivid imagination of spiders. To speak more colloquially, I am afraid of spiders. It is natural to describe dispositional fear in object-directed terms. However, this does not mean that fear itself is an object-directed state. My occurrent state of fear, whether it is elicited by spiders or heights, will have the same effects on my subsequent responses.Footnote 9

It may seem that I am walking into a familiar trap. Anthony Kenny (1963, p. 72) argued that the object of an emotion is not always its cause. After all, sometimes we seem to to be afraid of future events, which cannot cause anything in the present. Deonna and Scherer (2010, p. 45) provide a different type of example: “My fear of the elephant may have been caused by rumors concerning its having escaped the zoo. Here the object of my fear is still the elephant, but the cause is the rumors.” I argued that one source of TOKEN is that we notice the eliciting conditions of emotions, and take the emotions to be about, in an intentional sense, the elicitors. But this strategy seems unable to deal with cases where the apparent object of the emotion is different from its cause.

My solution is actually the same one that Kenny (p. 72) attributes to Descartes. We do not take emotions to be about their causes. Instead, we take emotions to be about the objects of the mental states that elicit them. We notice that spider-representations elicit fear, so we take ourselves to be afraid of spiders. This takes care of emotions that are apparently about the future, since in those cases the emotions are elicited by representations concerning the future. As for the elephant case, the fear seems to be directed at the elephant because it was caused by imagining a runaway elephant rampaging through town. Kenny (p. 73), in rejecting this solution, said, “If the relation between an emotion and its object were one of effect to cause, then it would be only by induction and tentative hypothesis that one knew on any particular occasion what one was afraid of or excited about. But this is sometimes obviously untrue.” Setting induction aside, the empirical evidence that I have presented suggests that we do, in fact, attribute objects to emotions via tentative hypothesis. While an occurrent emotion does not require any cognitive input to influence judgment, DeSteno et al. (2000) demonstrate that the scope of that influence is only limited to its purported object via an effortful and fallible cognitive process. This is not consistent with Kenny’s notion that having an emotion is sufficient for knowing its object.

Second, the object that elicits an emotion is in some cases the only present object of a type that is salient to that emotion. Imagine that David is typing away at his desk when he notices a giant hairy spider. Seeing the spider will elicit fear, which will make him more sensitive to any apparent danger. Since the spider is the only dangerous-seeming thing in his environment, the effect of his emotion will only be manifest in his response to the spider. It will seem as though his fear influences him in a way that is specific to the spider, which is also the object of the perception which elicited that fear. This sort of case reinforces Kenny’s intuition that we immediately know the objects of our emotions. In fact, the evidence I have presented suggests that David’s fear would have increased his sensitivity to any other apparent dangers, if he were exposed to any.

Since I deny that emotions have targeted effects, it may seem as though I cannot explain targeted motivational behavior. We need to explain why David would flee the spider if his fear were not directed at that spider. However, we should not overlook the motivational role played by desires. Setting fear aside, most of us desire to avoid bodily harm. If I am driving a car at a moderate speed, I will avoid wrenching the wheel over in a way that would send me crashing into a tree or into oncoming traffic. I avoid doing this even when I am blissed out, listening to music as I drive. If I happen to be in a state of fear, like David in the spider example, my desire to avoid any apparent dangers will magnified, but the only kind of targeted motivational state we need to appeal to is desire. Fear magnifies my desire not to be harmed, but that desire, along with my beliefs about which actions will lead to avoiding harms, explains why I perform one action rather than another.

We can clarify this point by considering an analogy between emotions and drives such as hunger and thirst. We sometimes say things like “I’m hungry for a big bowl of ice cream,” which on the face of it implies that states of hunger can be directed at types of food. However, we do not take these claims at face value. Hunger is just a mechanism that regulates our general desire to eat. Instead, when we say that we are hungry for ice cream, we really mean that we desire ice cream, and that desire is strengthened by (among other factors) a global state of hunger. Anyone who has ever attempted to maintain a diet is well aware that desire to eat appealing foods can persist in the absence of actual hunger. Likewise, fear globally strengthens our desire to avoid danger, but that desire persists in the absence of fear.

Third, attributing objects to our emotions helps us comply with social norms. Susan may have a contentious argument with Janet, and then encounter Sue, who makes a mildly obnoxious comment. Anger causes all offensive things to seem more offensive, so Susan will find it difficult to be as tolerant of Sue as she normally would be. It may seem perfectly appropriate at that moment to respond tersely or with a sarcastic remark, but if she does so, the hostility of her behavior will be out of proportion to Sue’s offense. Showing a bit of anger now and then is seen as normal, reflecting strength and passion. However, people who let their anger at one person color their responses to others just end up looking like jerks. In order to avoid looking like a jerk without suppressing our anger entirely, we need to attribute objects to our states of anger, and suppress our responses accordingly.

It would be natural to describe this norm by saying that if you are angry at one person, you should not take out that anger on someone else. This description presumes that states of anger are, at least in some cases, directed at people. However, it is possible to describe this norm by appealing to the etiology of the anger rather than its object: if a state of anger is elicited in response to a thought about p, then it is inappropriate to act towards q in a way that is influenced by that anger, unless q is suitably related to p. This description even preserves the force of the norm. If someone makes you angry,Footnote 10 then given human nature, they should expect you to respond forcefully, and perhaps out of proportion to the original offense. The acceptability of that response will depend on the context. However, if you are angry, and you are relating to someone who was not responsible for making you angry, then it would be unfair for them to deal with your belligerent behavior under almost any context.

This point becomes more clear when we consider the norms governing moods. When we encounter someone who is in a bad mood, the norms are much the same as encountering someone who is angry, in cases where we played no relevant role in the etiology of that anger. Out of prudence and sensitivity we step lightly around such people. At the same time, we consider it inappropriate when they do lash out at us without our doing anything to deserve it. This shows that these norms do not rely in any critical way on emotions having particular objects.

Fourth, attributions can help us identify when to suppress an emotional response that will interfere with the task at hand. Emotions can be beneficial because they get us in the right mindset to address certain kinds of problems or opportunities. However, a mindset that is appropriate for addressing one situation may not be helpful when dealing with another. If Joanne loses a job she will become sad, which will make her reflective and risk-averse (Lerner and Keltner 2001; Bodenhausen et al. 1994). This response is appropriate when dealing with the aspect of her life in which he suffered the loss, but general moroseness might lead her to pass up opportunities for growth or pleasure. Joanne, by attributing her sadness to the loss of her job, sets the groundwork for suppressing the symptoms of her sadness when dealing with other issues. Attributions are a prerequisite for targeted emotional responses, so they play an important role in our emotion regulation.

Opacity and function

I will finish by considering a pair of objections that would likely occur to philosophers working on emotion. First, one might argue that emotions have intentional objects by appealing to the semantics of emotion ascriptions. As Forbes (1997) puts it, ‘Lex Luthor is afraid of Superman’ is true, but ‘Lex Luthor is afraid of Clark Kent’ is false. Intentional state terms like ‘believes’ and ‘desires’ create opaque contexts. Emotion terms also create opaque contexts, so it is natural to suppose that emotions are themselves intentional states.

Graeme Forbes begins his explanation of opacity in transitive verbs by suggesting that, ‘Luthor fears Superman, so-labelled, but not Clark, so-labelled’ (Forbes 1997, p. 10). This analysis seems to be correct, as long as Luthor’s fear of Superman is understood to be a dispositional emotion. When we talk about emotions in a dispositional sense, we are interested in their elicitation conditions, so the elicitation conditions of an emotion determine the truth values of sentences that describe them. Since it is possible for ‘Superman’-representations to elicit fear in an individual for whom ‘Clark Kent’-representations do not, ‘is afraid of’ can create an opaque context. Occurrent states of fear, however they are elicited, do not have objects.

Second, Deonna and Scherer (2010 p. 48) argue that emotions have objects by appealing to utility:

Why should this important mechanism be so strongly affected by transient noise? Would it not seem more reasonable to assume that emotions are elicited and differentiated by the judgment or appraisal of what philosophers have called the “intentional object” and that conscious feeling reflects the content of appraisal as well as its mental and bodily consequences, in particular, adaptive action tendencies?

I have three responses to this objection. First, DeSteno et al. (2000) show that emotions really are strongly affected by transient noise. A satisfactory theory will not only explain the utility of emotions, but also the respects in which that utility is limited.

Second, states with only type-specific effects can still do a great deal of good. DeSteno et al. (2000) argue persuasively that we use correction processes to limit useless or harmful effects. As long as we are willing to devote the necessary cognitive resources, we can have the next best thing to truly token-specific effects. But even unfiltered type-specific effects can be useful. Emotions are often fleeting, and they are often elicited by perceptions of people or things in our immediate proximity, or thoughts about people or things that are not present, but which require immediate attention. If we assume that emotions are most likely to be useful when they influence our response to p, where p is the intentional object of the state which elicited that emotion, we should expect emotions to be useful much of the time. The effects that are documented in the misattribution literature are of great theoretical interest, but they are produced under highly contrived circumstances, so it is easy to exaggerate the scope of their practical importance.

Third, thinking of emotions within an embodied framework helps clarify the function of emotions as understood by TYPE. The peripheral bodily changes that occur during emotional episodes have two roles: they alter the effectiveness and efficiency of different types of actions, and they send signals to our peers. As a general rule, it is better to perform the types of actions for which your body is currently optimized, and it is more appropriate to act in ways that are consistent with the signals you are sending. If we were designing an organism from scratch, and we were going to include bodily changes which have these two functions, we would therefore include a system to steer decision-making in a manner that promotes actions that are optimal and appropriate given current bodily conditions.

Assume, for the sake of argument, that this organism that we are designing will sometimes undergo bodily changes that resemble ours when we are angry. These changes improve the effectiveness and efficiency of any aggressive action, not just aggressive action directed at some particular target. They send a signal that promises willingness to take aggressive action in general, rather than aggressive action toward a specific target for a specific reason. In the interests of effective and appropriate action, we would therefore decide that our organism’s decision-making processes, when it is in this body state, should be systematically altered in a manner that promotes aggressiveness in general. Our emotions may produce type-specific effects for just this reason. It is useful for emotional motivations to be as general as emotional physiology.

Defenders of TOKEN explain many features of emotions by appealing to their particular intentional objects. My goal in this section was to suggest that this is not necessary. I argued that emotions without particular objects can be quite useful, particularly when they are appropriately regulated. In my treatment of the linguistic features of emotion terms, instead of appealing to the particular object of an emotion, as defenders of TOKEN likely would do, I appeal to the particular object of the intentional state that elicited the emotion. I also used this strategy in my explanation of why emotions appear to have particular objects, and I suspect it would be useful for addressing other theoretical challenges that are beyond the scope of this paper.

Conclusion

In this paper I had two main tasks. The first was to show, through close analysis of psychological research, that emotions do not have particular intentional objects. Instead, the impression that they do have objects derives in large part from the way we selectively suppress emotional effects. The second task was to look beyond the laboratory, and explain how this model can account for the common impression that emotions are directed at objects. Philosophers working on emotion are likely to think that there are numerous considerations in favor of TOKEN, many of them individually compelling. I have attempted to show that TYPE can diffuse a broad range of objections, and in fact seems powerful enough to account for all emotional phenomena.

Defenders of TOKEN may remain unpersuaded. A good number may be sympathetic to the primary-effect/side-effect objection, and deny I have proven that emotions do not have any token-specific influences on thought and behavior. This is certainly right, since it is challenging to prove that an effect does not exist. What I hope to have done is present a compelling alternative to the received view, an explanation of how emotions influence our thought and behavior which does not appeal to particular objects of emotions. Ideally readers would be convinced that TYPE is the best available model. My more modest goal is to cast some measure of doubt on TOKEN, so that proponents see it as a position which requires defense.