Cognition in the structure of emotion
This chapter focuses on the content of human emotions from a structural point of view. The various elements of emotion identified in this discussion will figure in the next chapter's account on the dynamics of emotion. I divide my discussion on the role of cognition in emotion into these two approaches because even if almost all theorists pay lip service to the dynamic character of emotions, most of them -and philosophers in particular with the exception of Jenefer Robinson (2005) -have remained firmly on the structural side in their accounts of emotional content. However, both perspectives must be kept in sight if we want to understand what emotions truly are.
The question about the role of cognition in emotion can be put at two different levels of analysis, functional and algorithmic, identified in the introductory chapter. Existing theories of emotion seldom distinguish between these two levels, which makes it difficult to evaluate and compare their claims about the involvement of cognition in emotion. Therefore, I have decided to separate these two levels of analysis in order to put the cognitive claim into a more comprehensive test. Another issue at stake is the robustness of the cognitive claim. Emotions are cognitive in a less robust sense if they involve cognition at only one level of analysis than if they involve cognition at both functional and algorithmic levels of analysis. I shall suggest that something like this is the case: that cognition is always involved in emotion at the functional level of analysis, whereas emotions divide into two kinds, cognitive and noncognitive, at the algorithmic level of mechanisms and types of representation. However, this distinction makes sense only from the structural perspective that does not take into consideration the interaction and integration of different kinds of processes and representations that occur in actual human emotions. This dynamic approach to the role of cognition on emotion will be taken up in the next chapter in which I shall argue that even those post-infantile human emotions that are triggered by noncognitive representations and processes involve cognition at the later stages of appraisal and emotion regulation that should be seen as parts of emotion. Therefore, all post-infantile human emotions qualify as robustly cognitive also at the algorithmic level of analysis. This chapter will focus on building a typology of those representations and processes that figure in the dynamic account.
Cognition in emotion at the functional level of analysis
Functionalist theories of mind identify types of mental states by their causal roles in the information processing system of which those states are part. All token states that serve in the same causal role belong to the same type in the functional sense, independently of their internal structure. Thus mental states of the same type can be realized by different kinds of physical states. This principle of multiple reliability allows us to ascribe cognitive and motivational states to both humans and animals as well as to robots and fictitious aliens whose internal constitution differs significantly from ours insofar as their states of the same kind share similar causal role in relation to their sensory input, other mental states, and behavioural responses. No wonder then that the leading metaphor of a functionalist theory of the mind is software - a computational program that can operate in organisms with different kinds of hardware. Accordingly, functionalism is a prevalent approach to information processing in cognitive psychology, cognitive science, and artificial intelligence in addition to philosophy of mind where it is an influential paradigm (see e.g. Lycan, 1994).
If we take a functionalist perspective to cognition, what is the causal role of cognition in the mental economy in general and in emotions in particular? Fred Dretske (1981) ascribes cognitive mental states to all systems that are capable of converting perceptual information into structures with a semantic content. He regards perception as a process by means of which information is delivered within a rich matrix of information to the cognitive centers for their selective use. The process of successful conversion of analogue sensory information into digital semantic form constitutes the essence of cognitive activity and the results of this activity are structures with a semantic content. More precisely, the semantic content of a token internal state is defined by the kind of information to which internal states of that type are selectively sensitive. A type of internal state, in turn, constitutes a semantic structure whose content derives from the sort of situations that the structure was either biologically pre-wired or culturally developed to represent. Semantic structures qualify as cognitive insofar as it is their content that is causally efficacious in determining the system's response. This entails that the system displays flexibility by assigning "a sameness of output to differences of input" (Dretske, 1981, p. 183, original italics).
Human concepts are paradigm examples of semantic structures. However, Dretske maintains that we must assign semantic contents to higher animals' internal states as well if we want to explain their discriminatory responses. Both humans and animals have semantic structures for representing various kinds of emotionally relevant situations, for instance. We attack or flee from objects that we categorize as threatening and seek to avoid disgusting substances. Dretske notices the reluctance of some philosophers to assign cognitive structures to non-linguistic animals. Yet he argues that their specialized yet flexible response patterns compel us to assign them internal states and functional concepts with a specific cognitive content. After all, an internal state qualifies as a cognition when it functions in a cognitive role.
We can now see why emotions necessarily involve cognitions at the functional level of analysis. Emotions are not reflexes or fixed action patterns that activate automatically in certain stimulus situations. Emotional responses may have some reflex-like features insofar as they unfold in the manner of complex, coordinated, and automated affect programs (see e.g. Ekman, 1992, 1999; Levenson, 2003). Yet these responses are flexible both in terms of their eliciting situations and behavioural output. This indicates that emotions involve mental representations that mediate between perceptions of emotion-eliciting stimuli and emotional responses. These mental representations are cognitive in the functional sense as they causally function in the role of cognition.
On the one hand, emotions display sameness of output in relation to varieties of input. For instance, both rabbits and humans respond with fear - in proper circumstances - to perceptions of a snake or a wolf or a fast approaching truck. Since the physical properties of these objects are dissimilar, we must assume that there is an internal mechanism that represents those objects as somehow similar; as threatening or dangerous in the case of fear. The sameness of response presupposes sameness of mental representation at the functional level of analysis. On the other hand, there are changes in output in relation to the same input. To quote Agnes Moors (2007, p. 1241):
First, repeated encounters with the same stimulus can lead to a change in response - e.g. faster, more positive (cf. mere exposure); less intense (cf. habituation); more intense (cf. sensitization). Second, the same stimulus can lead to a change in response due to (repeated) co-occurrence with other stimuli (cf. classical and evaluative conditioning). Third, the same stimulus can lead to different responses depending on the varying internal states such as motivation.
All these changes in output require flexibility in internal representations of input; in their accessibility, or in their association or integration with other representations. For instance, fear responses can adjust to repeated encounters with threatening stimuli by becoming either more intense (e.g. PTSD) or less intense (e.g. unlearning phobias in therapy); threatening stimuli may associate with innocuous ones, transmitting fear responses also to the latter; and other activated goals can ward off fear in a dangerous situation. This kind of flexibility in the input and output of emotions - that anatomically is founded on connections between the baso-lateral nuclei of the amygdale and the orbitofrontal and anterior cingulate cortical areas (e.g. Pessoa, 2008; Pessoa & Adolphs, 2010; Rolls, 2013) - can be explained only if emotions involve mental representations (e.g. Moors, 2007; Smith & Ellsworth, 1985).
Both cognitive and noncognitive theorists agree on the role of mental representations in the elicitation of emotions. Thus, for instance, Jesse Prinz (2004) maintains that all mental states that elicit instances of a particular emotion type together constitute the calibration file of that emotion type. However, he rejects cognitivism by arguing that "there is no internal state that always plays the role of triggering a higher cognitive emotion. Different items in our calibration files play that role on different occasions" (Prinz, 2004, p. 101). For instance, the elicitation file of fear may contain perceptions of a snake or a pointing gun, as well as judgments like "That's poison!" or "I'm in danger!" Yet from a functional perspective, the heterogeneity in the elicitation files of emotions is superficial because it hides an underlying homogeneity at the causal level of analysis: all items of the same elicitation file represent the same core relational theme as they give rise to token emotions of the same type. Since these inner causes have an essential role in emotions, emotion always involves cognition.
The functional view of cognition is capable of accommodating emotions elicited at different levels of information processing within the same theoretical model. Accordingly, the functional account has been very popular among multi-level appraisal theories which maintain that emotions involve representations of different types in the appraisal of their eliciting stimuli. In accordance with popular dual-process models of cognition developed in social and cognitive psychology and cognitive science, most contemporary appraisal theories distinguish between two levels of processing; "one conscious, deliberate, and under volitional control, the other automatic, unconscious, and uncontrollable", as Richard Lazarus (1991, p. 153) proposed. The former types of representations or processes have been characterized in various theories as "propositional" (Teasdale & Barnard, 1993), "deliberative" (Ben-Ze'ev, 2000), "rule-based" (Clore & Ortony, 2000; Smith & Neumann, 2005), "conceptual" (Leventhal & Scherer, 1987); or "reasoning" (Smith & Kirby, 2001), whereas representations or processes of the latter type are termed "schematic" (Leventhal & Scherer, 1987; Ben-Ze'ev, 2000), "associative" (Clore & Ortony, 2000; Smith & Kirby, 2001; Smith & Neumann, 2005), or "implicational" (Teasdale & Barnard, 1993). Emotions involve other components besides appraisals as well, but the causal and constitutive role of appraisals at either or both levels of information processing renders emotions cognitive according to these theories. Louis Charland (1997) captures the strategy of multi-level accounts by subsuming different types of processes and representations into his explicitly functionalist "Representational Theory of Emotion", which distinguishes between "affective cognition" and "affective perception". In this early theory, Charland purports to synthesize both cognitive and noncognitive approaches in the philosophy of emotions by ascending to a higher level of description, that of representations, which are neutral in both epistemological and ontological terms.
The functional notion of representation has allowed multi-level appraisal theorists to maintain that mental representations at different levels of information processing have the same or similar evaluative content insofar as they elicit emotional responses of the same type. For instance, some fears are elicited quickly and effortlessly from perceptual stimuli, whereas others emerge only through slow and effortful understanding of conceptual information. It is obvious that representations of danger that have the same causal role in emotion elicitation emerge as the result of different types of processes in somewhat different areas of the human brain in these two kinds of fear. However, multi-level appraisal theories maintain that these differences do not significantly affect the nature of emotions. To quote Clore and Ortony (2000, p. 42), "the cognitive claim is that emotions are reactions to (or representations of) the personal meaning and significance of situations, not that emotions originate in the cerebral cortex". The recognition of personal meaning is automatic and nonconscious if it takes place in the subcortical pathway of the brain. However, from an informational and functional point of view it is still a recognition, and as such a kind of cognition, as Robert Solomon (2004, p. 79) points out.
In spite of their popularity, multi-level appraisal theories with their functional view of cognition have been subject to persistent criticism especially in philosophy. One of the most vehement critics of this approach has been Paul Griffiths (1997; 2004), who argues that functional similarities are shallow in comparison to homological similarities that are manifested at the level of causal mechanisms and physiological implementation. Appraisal theories are vulnerable to such criticism as they avoid commitments in these matters or provide only tentative hypotheses (Moors, 2009). Thus, Griffiths claims that instances of the same vernacular emotion type such as anger, fear, and sadness fall into two subtypes - affect program emotions and cognitively complex emotions - which have little in common. Affect programmes are phylogenetically ancient, informationally encapsulated, and reflex-like emotional responses that are homologous with the emotions of other vertebrates, and therefore, constitute a plausible natural kind. In contrast, complex emotions are culturally specific responses that operate on semantically structured representations of objects, events, or states of affairs, without possessing a distinct, emotion-specific physiology. Griffiths (1997, p. 16) concludes that "the two kinds of emotion have different phylogenies, different adaptive functions, different neuroscience, and different roles in human psychology, [and therefore t]he concept that groups them together has no discernible theoretical utility."
With these theoretical tools Griffiths attacks multi-level appraisal theories that purport to bridge the gap between affect programme emotions and complex emotions by maintaining that all instances of the same vernacular emotion type share the same or similar content. Griffiths argues that this view contradicts the hierarchical architecture of the emotional brain. Representations with the same content cannot operate at different levels of implementation. Further still, he suggests that there are several differences between low- and high-level appraisals. First of all, lower-level emotional appraisals involve action-oriented representations that unite the functions of belief and desire. This kind of "collapse of the attitudes" (Millikan) is typical of mental representations in simple organisms. Moreover, Griffiths argues that lower-level representations differ from higher-level ones in terms of their impoverished inferential role:
First, low-level appraisal processes do not have access to most of what is represented elsewhere in the brain... Secondly, the processes of affective computing, as opposed to their final output, are not available for inspection by other cognitive subsystems.... Finally, the inferential principles used in affective computing are not truth-preserving, but heuristically survival-enhancing. (Griffiths 2004, p. 247)
Together these arguments suggest that theories based on functional accounts of cognition skate over important dissimilarities between different kinds of emotions and their eliciting appraisals, which therefore cannot have the same content even in terms of their causal-representational role. Even if emotions involve mental representations that function as inner causes of emotion, those representations and the ensuing emotional responses are too dissimilar to each other to ground an informative representational theory of the emotions in the manner suggested by Charland (1997). This conclusion leads to the next level of analysis where the question about the role of cognition in emotion is put in terms of mechanisms and formats of representation.