Contribution
The first study to relate the CPM’s componential features, rated per event, to whole-brain activity. It decomposes emotional experience into data-driven “functional core processes” (FCPs), each a weighted combination of features with a brain map. The CPM is taken as “a presumption to reveal potential core processes”, not a theory under test (p. 7994).
Key insight
The building blocks of emotion in the brain are not the CPM’s five components one by one, nor valence and arousal. They are mixed processes that cut across components. One of them looks like wanting and another like liking.
Assumptions
- That ratings on re-viewing, given by pausing the video immediately after each event, reflect the experience during the first viewing in the scanner.
- That one canonical HRF per event is adequate: “we used a single HRF function for every emotional event in movie clips to model the BOLD signal” (p. 8007), conceded as neglecting carry-over and habituation.
- That physiology should not be regressed out of the BOLD signal, being a component of interest.
Key results
- Behaviour. Clustering of emotion profiles splits positive from negative, then pleasant feelings, distress, annoyance and surprise. Six factors explain 47.8% of the variance: pleasantness, arousal, expression, novelty, action tendency, social norms.
- PLSC. Six significant latent variables, explaining 19.0, 11.0, 6.0, 5.1, 4.6 and 3.4% of the covariance. “Their weighted sum represents a specific combination linked to a particular brain activation pattern, whose expression characterizes a given functional core process (FCP) underlying the generation of an emotion response” (p. 8002).
- FCP-1, valuation. Low unpleasantness, low incongruence with standards, wanting to continue and not to stop, feeling good. Positive in vmPFC, lateral OFC, amygdala and VTA, and negative in anterior and dorsal insula, mid-cingulate, thalamus and striatum. Read as “wanting”, since it carries action features and low feeling intensity.
- FCP-3, hedonic impact. Expression features (eyes, lips, eyebrows) with warm, good, intense and lasting feelings. Positive in sensorimotor operculum (face area), SMA, dorsal ACC, posterior insula, ventral pallidum, VTA and pons. Read as “liking”.
- Physiology. HR and RR load negatively on LV5 and LV6. LV2 is negative and LV4 positive for HR alone. “We also note that we found no component uniquely associated with peripheral physiology changes during emotion events” (p. 8007).
- Valence–arousal control. PLSC on the feeling items alone gives two latent variables, valence and arousal, with the familiar brain correlates and less explained covariance.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Emotional experience decomposes into at least six brain–behaviour core processes | moderate | PLSC, permutation p < 0.01, bootstrap stability over subject subsets; 20 participants |
| C2 | Valence has two distinct substrates, a wanting-like valuation network and a liking-like hedonic one | weak–moderate | two latent variables with valence features and different maps; interpreted post hoc |
| C3 | Arousal is not a single functional process | weak | absence of an arousal FCP in the full model |
| C4 | The core processes mix features of several CPM components | moderate | loading profiles (Figure 6) |
| C5 | ”Being moved” has its own process, in the default mode network | weak | LV5, interpreted post hoc |
Concepts
- functional core process (FCP): a latent variable of PLSC, a weighted feature profile paired with a brain saliency map, shared across emotion categories.
- PLSC: partial least squares correlation, an SVD of the brain–behaviour correlation matrix giving paired latent variables of maximal covariance.
Connections
Realizes in part the five-network programme of Sander, Grandjean and Scherer (LIT-1471, NOTE-1103). The processes it finds do not correspond to the five networks. Reuses the film set and CoreGRID items of Mohammadi and Vuilleumier (LIT-1443). Its two valence networks are read through Berridge’s wanting and liking, which the record holds first-hand in Berridge and Kringelbach 2015 (LIT-1459).
Bearing on the record
A per-episode measure of synchronization? No. Synchronization is not computed. Each event’s brain signal is a single GLM contrast over 12 seconds, and each event’s features are a single rating. The covariance PLSC extracts is across events and subjects. What the method does give per episode is a score. The latent variables are defined per observation (the behavioural scores are the rating matrix times the saliences), so every event has a signed value on each process, both from its ratings and from its brain map. The paper reports loadings and maps, not these event scores. For QUESTION-268 this is a per-episode, signed, brain-and-report measure of which pattern an episode expresses. It is still not a measure of potential. A high FCP-1 score means the episode had the valuation pattern. It does not mean that pattern won a conflict against another one.
Signed, or which tendency prevails? FCP-1 is the one process here that is signed along an action-tendency axis. Its positive pole is the item “Want the situation to continue?”, its negative pole the items “Feel the urge to stop what was happening?” and “Want to undo what was happening?” (Table 1). On TERM-076 it is the closest thing in these five papers to a measure that tracks potential and valence together: an evaluation network whose sign follows the direction of the prevailing tendency. Two cautions. The tendencies are reported, not enacted, and in passive viewing there is nothing to continue or stop. And the sign is of a bipolar factor, which by construction cannot represent “continue” and “stop” both high, the ambivalent case. PLSC on a design with choice points would be needed.
On QUESTION-266 and the wanting/liking objection. This is the result that matters for the record. In human brain data recorded during naturalistic episodes, valence features load on two separate processes:
- FCP-1 carries the action tendencies (continue, stop, undo) with vmPFC, OFC, amygdala and VTA, and the authors read it as “wanting”.
- FCP-3 carries feeling intensity and facial expression with the sensorimotor face area, posterior insula and ventral pallidum, and they read it as “liking”. “Our findings of 2 distinct networks (FCP-1 and FCP-3) correlating with valence features converge with growing evidence that this dimension may not have a unique psychological or neural underpinning” (p. 8006).
For QUESTION-266’s layers this is close to a human dissociation of layer 1 (potential, read off the tendency that governs: FCP-1) and layer 2 (a hedonic marker that does not depend on choice: FCP-3, whose rated expression features are a rough human analogue of taste reactivity’s orofacial signs). It agrees with Berridge’s animal work (LIT-1459), which THEORY-893 and CLAIM-438 use against the proposal that valence is potential by a self-maintenance ordering (CLAIM-440). It is weaker evidence than it looks, for three reasons:
- The two processes are orthogonal by construction (PLSC latent variables are orthogonal), so their separation is partly the method’s. What is not imposed is the difference between their brain maps.
- Both are read from one retrospective report. FCP-3’s “liking” is rated expression and rated feeling, not a measured face. Facial EMG was attempted and lost.
- They correlate with the same discrete emotions in the same direction (FCP-3 “shows strong positive correlation with joy and love, but negative correlation with sadness and anxiety, like FCP-1”, p. 8006). So across these films the two coincide in sign, even though they load on different features. Wanting-without-liking needs episodes where they come apart, and the paper does not report any.
So the paper does not show the dissociation THEORY-893 needs. It shows that the two layers have separable substrates in humans and could be measured separately, with event scores on FCP-1 and FCP-3 as candidates. That is what QUESTION-266 asks for: each layer measured without the others.
Arousal. The absence of an arousal process bears on the intensity clause of TERM-076 only weakly. If arousal is not one process, an unsigned intensity read off arousal would not be one quantity either.
Limitations
Twenty participants. Ratings given on re-viewing, after the scan. One HRF per 12-second event. Vicarious, passive emotion. Post hoc interpretation of every latent variable, with the risk of reverse inference the authors acknowledge. Facial EMG and electrodermal data lost. The data are available only on request.
Open questions
Whether the per-event FCP-1 and FCP-3 scores ever come apart in sign within this dataset (wanting without liking, or the reverse), which would be the human dissociation the record lacks. Whether the analysis on a design with choice points (the VR route of LIT-1440) would let FCP-1’s sign be read as the tendency that prevailed.