Comment: Language and Dimensionality in Appraisal Theory

Causality

The relation of causality to conceptual connection has given rise to much confusion. Moors’ (2013) careful discussion of the question whether appraisals cause emotions or constitute emotions clears up the muddle very nicely. She examines a number of interpretations of the claim that appraisals are causally involved in emotions, and defends it against objections. In the end, she finds a sensible way, based on a distinction between the process and the output of appraisal, of evading the apparent incompatibility of the causal claim with the competing thesis that appraisals are definitionally tied to emotions. She concludes that there is no conflict between saying that “the appraisal process causes the appraisal output, which in turn, causes the other components” and claiming that “the appraisal output and the other components may be part of the content of feelings” (Moors, 2013, p. 138). And she sensibly points out that causal relations between the various so-called components of emotion remain an interesting area of research.

As if to answer her prayer, Scherer’s (2013) compact diagram in Figure 1, showing a forest of causal influences and feedback loops between different levels of appraisal, seems to have the right order of complexity. It can be grasped in theory, but that level of intricacy is susceptible to being implemented only by the subpersonal, nonexplicit levels of brain activity, and to being understood only by computer analysis: those are features of the “intuitive” processing system as opposed to the “analytic,” as I shall discuss in what follows.

I suspect, however, that the preoccupation with questions of causation might be unhelpful. Here is one reason. Moors (2013) refers to the very appealing three-tiered model proposed by David Marr (1982) for understanding a complex system. But Marr’s model, especially when applied to biological systems rather than to artefacts, does not throw very much light on the causal situation. The kludgy character of many organs suggests that, causally speaking, some prior configuration determined the organ’s design. In some other cases, the algorithmic level, concerned with the system of representation used by the organ, might be causally crucial. Yet the most important level for understanding is what Marr calls the “computational” level, which requires us to understand what the thing is meant to do. Ellsworth (2013) hints that the distinction between being a cause and an ingredient is sometimes a matter of perspective. Indeed. It may be useful to recall that for Aristotle, the “material cause” and the “efficient cause” were just two of four modes required for a complete explanation. One also needed to know what something was made of, its “material cause,” to have a functional understanding or “final cause,” as well as insight into the essence of the phenomenon or “formal cause.” While it is important to know what we are talking about at any particular moment, many philosophers have come to think that discussions of “essence” too often reduce to mere semantics. In most cases, it is quite appropriate to speak, as Ellsworth does, of “correlates of appraisal” (p. 126), without needing to specify how these correlations come about before specifying the precise explanatory goals to be served.

Almost a century ago, Bertrand Russell (1918) argued that causation is not a scientific notion. It belongs to what we now call folk psychology, ruled by the practicalities of intervening in complex processes (Hacking, 1983). What caused the match to light? Its being struck on a rough surface, we say. But what makes that useful is not that friction is the only causally necessary condition of the match’s lighting; neither is it sufficient. Rather, unlike the presence of oxygen and the absence of excessive humidity, striking the match is the condition we are most apt to control. An appraisal can claim to be involved in all four of Aristotle’s explanatory aims: first, it is “essential” in that an ascription of emotion implies at least some form of appraisal; second, the emotion type is likely to be functionally useful in standard circumstances; third, appraisals do seem to be “ingredients” in emotions; and fourth, getting a grip on the appraisal is often our best chance of changing the course of the emotions—making them what we call “efficient cause.”

Once we relax about causation, we can broaden our understanding of “understanding.” Ellsworth (2013) writes: “We can go much farther than we have in refining appraisals, but I do not think that we will ever develop a complete set that accounts for all emotional experience” (p. 127). But what is meant by “accounts for”? One way to account for an emotion is to describe how it unfolds. That narrative aspect of emotions is an important aspect of emotions largely left out of appraisal theories. A Shakespeare soliloquy, or a few pages of Proust, convey in minute detail the vicissitudes of an emotional episode in which appraisals may change, following an arc that may have its roots in what I have called “paradigm scenarios” learned in early childhood (de Sousa, 1987). As Ellsworth suggests, these are apt to change over time (p. 130). They are enriched and modified by countless memories, associations, analogies, and refinements in one’s background value system. So it is not surprising that when people try to describe novel emotions they typically appeal to situations and scenarios: “‘It was like finding out that there was no school today’; ‘it was as though I had cheated’; ‘it was like being betrayed’” (Ellsworth 2013, p. 128). In accounting for an emotion in this sense science’s role is auxiliary. Its contribution to “accounting for,” in the sense of explaining, the full details of a particular episode, consists in identifying the parameters for which specific values are filled in according to the particulars of each singular episode. ¹ That is where the exploration of the dimensions of appraisal can be helpful, and it is to some issues raised by dimensionality that I now turn.

The Proliferation of Dimensions

Many of the original dimensions included in Scherer’s (2013) and Roseman’s (2013) theory admit to many refinements. I shall focus exclusively here on the key dimension of valence. Aided by refinements in the understanding of the brain mechanisms of pleasure and pain (Grahec, 2007), philosophers have cast doubt on the commonsense idea that these merely represent what is above and below the line of indifference in a single dimension of experience. Some have specifically taken aim at the assumption that emotions are either “positive” or “negative” (Kristjansson, 2003). Nevertheless, that terminology persists in many quarters. Indeed, it is implicitly present as just one dimension in Scherer’s Table 1, which displays “intrinsic pleasantness” as the second dimension in the category of “relevance.” Scherer also refers to “the fundamental implications of approach and avoidance tendencies used by Lewin as the basis for introducing the notion of valence” (p. 154). But that “basis,” which psychologists repeat like a religious incantation, is persuasive only to the faithful. It is simply false that positive valence can be defined in terms of a tendency to approach, and negative valence in terms of a tendency to avoid. Anger is undoubtedly negatively valenced; yet an angry person will aggressively approach the target of their anger. Entire traditions of religion and morality have been built around the idea that what is desired should by all means be avoided. Of course, an objector will say, approach and avoidance must not be construed literally! Precisely: that’s what they say about faith. Psychologists who insist on it must explain why their conception of approach and avoidance isn’t just identified by definition with positive and negative valence respectively.

In the light of these problems, it is gratifying to find Scherer (2013) breaking up the supposed unity of valence into “at least six types of qualitatively different valences” (p. 154). But one of those dimensions itself illustrates the point I have just made. According to Scherer (2013, Table 2), familiarity is listed as a dimension of valence. But is it familiarity or surprise that is correlated with approach? Surprise captures attention: that would seem to be a form of approach. On the other hand, it is familiarity that is coded as positively valenced in Scherer’s scheme. One could similarly quibble about whether “self-worthiness” is something to be approached or something to be avoided. You could see it both ways, like the duck–rabbit illusion.

One more detail, Scherer (2013, p. 152) remarks that “the un/pleasantness of some stimuli is intrinsic to the stimulus, that is, that it does not depend on the motivational state of the appraiser.” (Just a few lines down he mitigates this, using the phrase “largely independent of the motivational state,” p. 152). This seems to be contradicted by the familiar example of the athlete who does not notice, until after the end of the race, that she has been seriously wounded. It would be daunting to devise an experiment in which you caused a major laceration in the midst of a subject’s extreme physical exertion, obtain an appraisal of its effects at the time, and then compare that appraisal to one made later in tranquillity. The answer presumably lies with the consideration that in the case of the athlete, the motivational state suppresses, rather than conditions, perception of the state’s intrinsic unpleasantness. But the nature of whatever underlying mechanisms are involved and the difference between them seem to be further issues that require investigation.

More generally, it seems somewhat arbitrary to limit the number of valence dimensions to six. In addition to self-worthiness and moral worthiness, why is there not a distinction to be made between both of those and prudential worthiness? And within long-term prudential worthiness, should one not distinguish the short-term from the long-term? The problem here is that the very word “valence” is more or less equivalent (in meaning as well as in etymology) to “value.” And value, as many philosophers have stressed, admits to an indefinite number of dimensions that can clash in practice despite having equally impregnable claims to being genuinely worthy (Berlin, 1958; Williams, 1986). So to open up the dimension of valence—and certainly it does seem to require to be opened up—is to expose oneself to the prospect of a proliferating pullulation of potential dimensions.

Emotions and Language

Ellsworth (2013) highlights the problem of the connection between the psychological reality of emotions and the linguistic repertoire of emotion words. Structurally, the vast multidimensional space of appraisals in which specific emotions find their place is a chaotic dynamic one (Lewis, 2005; Magai & Haviland-Jones, 2002). Such spaces typically have “attractors.” The named emotions can be regarded as identifying those attractors: “the words of a language may act as magnets in the multidimensional universe of appraisals” (Ellsworth 2013, p. 128). The impressive results obtained in the GRID study conducted by Scherer and his collaborators (Fontaine, Scherer, & Soriano, in press) suggests that some cross-cultural commonalities exist to ground a widely shared, if not universal, localization of attractors in the vast space of possible emotions. But that structure, in turn, requires an explanation: Why those points and not others? We have a limited vocabulary to designate some types of weather, mental health, and colours (Ellsworth, 2013), but these obviously pick out only a very few salient types. If we turn inwards and try to experience our own emotional state in its rich specificity, we don’t need to use words, though we very likely will be tempted to use words, if only because the habit of verbal categorization leads to quick and efficient ways of deciding what, if anything, one should do. (But as soon as we are asking what to do, we are no longer simply attending to the character of our own inner experience.) But it is in social life that most emotions have their being, not merely because they arise in social situations, but because conversation is among the chief activities conducted socially. Conversation is driven by four goals that we can conveniently sum up as PEGGing: Predicting, explaining, generalizing, and gossip. Gossip, according to persuasive evidence marshalled by Robin Dunbar (1993), played an important role in the development of our brains during that period of our ancestors’ evolution in which groups became too large for everyone to groom one another efficiently. Dunbar estimates that gossip takes up about 60% of all conversation; others have found it to be closer to “80% to 90%” (Elmer, 1994, p. 131). For the purposes of PEGGing, we need labels that function in the sort of way described by Eleanor Rosch (Rosch & Lloyd, 1979), referring to “prototypes” easily recognizable in terms of intuitively perceived family resemblance rather than clear lists of necessary and sufficient conditions. The so-called basic emotions will tend to be those which can be identified from the outside, in terms of the scenarios that embody them, rather than in terms of the feelings of the emoters. If a man with a knife lurches towards another man, our inclination to attribute anger to the first and fear to the second in no way depends on having interviewed either man concerning the profile of his appraisals. If Y’s spouse X has discovered that Y has slept with Z, X may, in actual fact, experience any of a very large and disparate range of emotions. Which it is could hardly be discovered without asking. Most observers won’t bother; they will attribute jealousy to X and assume they are thereby entitled to predict what X is likely to do, to explain how X actually responded, and to be understood when they gossip about it—all in virtue of generalizations widely taken to be axiomatic. Labelled emotions, in short, are those that are salient because they are particularly useful for the purposes of PEGGing. That implies that the categories in question are not, in fact, primarily psychological, but social; and it seems quite compatible with the result of the GRID study that attributes particular importance to the four dimensions of valence, power/control, arousal, and unpredictability (or novelty; Scherer, 2013, p. 154).

Levels of Processing

A broader issue about the role of language in appraisal and emotion concerns the contribution of various levels of processing. Moors (2013) writes that

appraisal theorists generally agree that various mechanisms can underlie appraisal and that they can operate on a wide range of representations: conceptual and/or propositional versus perceptual and/or embodied; symbolic versus subsymbolic; locationist versus distributed…. appraisal often proceeds automatically (i.e., uncontrolled in the promoting or counteracting sense, unconscious, efficient, and/or fast…) but can also sometimes proceed nonautomatically. (p. 120)

This, together with her allusion to “dual mode” and “triple mode” theories (Moors et al., 2013, pp. 120–122), suggests an affinity with the “dual processing hypothesis” (Evans & Frankish, 2009). ² Descartes had already pointed out in 1649 that “those who are most violently moved by their passions are least likely to know them” (Descartes, 1649/1984, §74). In the perspective of dual processing theory, emotions pretty clearly demand to be classed as belonging to the “intuitive” system, S1. At the same time, they are eminently susceptible to being triggered by verbal information, which typically belongs to S2, the “analytic” system. This happens not only in learning facts that touch our concerns, but also, of course, when reading fiction. In addition, it is surely true that “rule-based mechanisms can also be automatic … and that the associative mechanism can also operate on conceptual codes” (Moors et al., 2013, p. 122). Indeed, this is surely a certainty. But the question is, how exactly is it that emotions manage in such an elusive way to straddle the different “modes”? Language also works in nonstandard ways, that is, in virtue of the associative properties of words rather than the logical properties of sentences, as in priming effects that work below the level of awareness to change one’s moods and likely responses to other verbal or nonverbal input; emotions are affected by both systems, even in their linguistic dependencies.

Thus emotions and appraisals navigate between the intuitive and the analytic systems. This may be particularly important given the role of the “individual, cultural and developmental differences” to which Moors et al. (2013, p. 120) allude. Much of the process of development would seem to consist precisely in the conversion of System 2 processes into System 1 processes which is what “overlearning” amounts to. One would like to know more about the mechanisms involved. For these reasons, Scherer’s (2013) articulation of levels of processing into four (low sensorimotor, schematic, associative, and conceptual; . 151) is very much to be welcomed as an advance. I find it a little unclear, however, how the second is separate from the third, given that both are learned, automatic, unconscious, and distinct from both the innate first level and the S2-like fourth level.

Second-Order Appraisals

I shall end with a question that intrigues me, but that does not fit neatly under any of the previous headings: the question of second-order appraisals. Moors et al. (2013) point out that “many events are congruent for one concern and incongruent for another. One both wants something and doesn’t want it” (p. 123). Scherer’s (2013) example of drinking a fine wine forbidden for health reasons is a case in point. He notes that this type of case is “rarely studied” (p. 156). It should be. So should the importantly different case where the conflict is between a first-order and a second-order emotion or desire. One can evaluate the very same thing as a bad thing at the first level of appraisal and a good thing at the second or metalevel. To borrow an example from Tim Schroeder (personal communication, June 2012), one can suffer the pain of punishment while deeming it to be justified. That could be described as being bad in the light of one concern (present pleasure) and good in the light of another (justice). But it is quite different from the conflict of those same two concerns when the question is, say, whether to spend my money on expensive entertainment or send it to Amnesty International. In the former case, but not the latter, it is the pain itself (of the punishment) that is positively appraised. This is broached in an article by Lambie and Marcel (2002) to which Moors et al. refer, but I don’t see in the articles here reviewed that any attention is paid to either of the two questions that can arise about second-order appraisals: the question of the objective justification for first-order appraisals—such questions as “Is it reasonable to be frightened by an inoffensive spider?”, for example, and the question of the mechanism and the nature of self-assessment, as in one’s emotional response to one’s own first-order appraisals (“How shamefully unreasonable of me to be frightened”). I’d love to know more about how both these conflicts work.