I am very much in sympathy with the model of emotion presented by Lindquist et al. in the target article, which successfully manages to integrate psychological data on emotion with current neuroscientific findings, and provides the possibility of grounding phenomenological accounts of intentionality and mood in the biology of affect. My points in this commentary are therefore not intended as criticisms of the model. I offer instead a clarificatory question, and a proposal for developing the model further.

I start by flagging a concern about the possible conflation of core affect with the representations of afferent homeostatic information. Core affect is an explicitly psychological phenomenon, yet the target text is sometimes ambiguous as to whether it denotes: (1) a psychological feeling state, (2) the representation of interoceptive information, or (3) the actual bodily state. For example, take this statement from the discussion of situated conceptualisation in section 3:

Here, “one's own body” must either mean the physiological body or the brain. The first option appears to conflate core affect (the feeling of physiological changes) with affect (the physiological changes themselves). The second option implies the quoted statement should read, “a prediction about what caused core affective changes which arise as a result of changes in one's body being represented in the brain.” But if so, then it is unclear whether the prediction underpinning the situated conceptualisation is supposed to concern bodily changes, the representation of those changes, or the feeling of those changes. Some clarification here would be helpful.

Although I agree that core affect is grounded in interoception, conflating the two risks obscuring the mechanisms for representing biological value; is it (1) core affect – the feeling of the interoceptive information – that is linked to sensory information, or (2) the neural representation of interoceptive information? It might be suggested that the feeling of the interoceptive changes just is the representation of those changes, but it seems unlikely that this is what the authors are claiming, since this would mean that their use of “feeling” did not necessarily imply that this information was experienced. While this is not an unprecedented use of the term (see Damasio Reference Damasio1999), the authors state specifically in the target article that core affect is a psychological phenomenon which always presents itself in experience, either in the foreground or the background. This raises an important question: Is the biological value that is being imposed on the environment as a result of situated conceptualisation an experienced value or can it be restricted to some sub-personal domain?

The second question that I want to raise concerns the dimensions of core affect. Barrett has argued elsewhere (Barrett & Bliss-Moreau Reference Barrett and Bliss-Moreau2009) that Wundt's original dimensions of affect should be translated as valence, arousal, and intensity. But Barrett and Bliss-Moreau reject intensity as a dimension, arguing that it is realised by the ends of the dimensions of valence and arousal. While Wundt does speak of intensity and quality in regard to affective and sensational systems, he does not do so in such a way as to suggest that these are the dimensions of affect (see Wundt Reference Wundt, Judd and Engelmann1897, sect. 3). Wundt's dimensions of feelings are: (1) pleasurable and unpleasurable; (2) arousing and subduing; and (3) strain and relaxation (Wundt Reference Wundt, Judd and Engelmann1897, sect. 7.7). Although (1) and (2) neatly map on to valence and arousal respectively, as Barrett and Bliss-Moreau suggest, I do not think that (3) maps on to intensity. I suggest rather that, where arousal is a distinct affective dimension likely to be grounded in the afferent wing of the continuous – but varying – activity of the sympathetic nervous system, and valence may be grounded in the motor aspect of the homeostatic/interoceptive loop (accounting for its relation to approach/withdrawal states), Wundt's third dimension of affect can be grounded in the force with which prediction error is being reduced.

According to the generalised predictive coding framework for neuroscience (Bar Reference Bar2009a; Friston Reference Friston2009; Friston & Kiebel Reference Friston and Kiebel2009), unexpected sensations are intrinsically motivating in virtue of the nervous system's function being to reduce error in predictions. According to such a story, any sensation that was not predicted will inherently motivate the system either to move and act, or to update its hypothesis concerning the cause of the predicted sensations. Action and/or hypothesis change are mandated until there is no error and the sensation corresponds to that which was predicted. Feldman and Friston (Reference Feldman and Friston2010) argue that attention may be the process of optimising the post-synaptic responsiveness of the units reporting prediction errors. What this means in practice is that if the system regards some sensory information as high-grade (hence reliable), it can increase the gain on the error-units reporting prediction error, thus giving added weight to that sensory data. In such a case, the units encoding that error become highly responsive (synaptic gain). This means they exert a greater force by which to resolve prediction errors. The searchlight of attention thus gets progressively finer until all prediction errors have been resolved, or a greater error arises elsewhere.

I propose that the directions of strain and relaxation correspond to the force and resolution of synaptic gain as evidenced in attention and the corresponding affective experience of being drawn towards that which captures our attention, or releases us. This fits Lindquist et al.'s compelling model of many-layered predictions generating situated conceptualisations. It also neatly incorporates attentional phenomena into the more basic operations of prediction and corresponding experience within the domain of core affect.