Contribution
An updated statement of EMA, with previously unpublished details, and an argument against multi-level appraisal. The argument is that a single appraisal level suffices, because the fast and slow dynamics of emotion come from the processes that build the representation appraisal reads. It is illustrated by encoding a naturalistic episode, an actor’s surprise, fear, defensive stance, relief and concern when a pigeon flew into a lab, as a sequence of EMA states.
Key insight
Appraisal is cheap and constant. What is slow is building the thing appraised. Keep one uniform representation of the person–environment relation, appraise every proposition in it continuously, and let coping act back on it. Emotional dynamics then follow from perception, inference and action, with no appraisal-specific timing.
Assumptions
- Appraisal causes emotion (Frijda’s law of situated meaning, p. 75), with mood allowed as an incidental influence.
- A symbolic, decision-theoretic representation of the agent–environment relation, after the Physical Symbol System Hypothesis (p. 76): beliefs binary with probabilities over futures, and utilities over states.
- One agent’s utilities as the measure of desirability, with others’ utilities representable from the agent’s point of view (p. 80).
- SOAR’s operator cycle: operators proposed in parallel and executed sequentially, about 50 ms each (p. 79).
Key results
Model and illustration, not experiment:
- Per-proposition appraisal (p. 80). Every proposition in the causal interpretation, past, present or future, gets an appraisal frame, continuously updated, including relevance, desirability, likelihood, expectedness, causal attribution, controllability and changeability.
- Emotion labels as a convenience (p. 81, Table 2). Hope, joy, fear, sadness, surprise, anger and guilt are read off appraisal patterns, but “it is the specific configuration of appraisal variables” that drives coping.
- Mood (p. 81). Frames aggregated by label, squashed to [0,1], added back to frames of the same label. Mood “tends to change slowly over time as appraisal frames are added or removed in response to changes in the causal interpretation.”
- Arbitration (pp. 81–82). The coping response follows “the most recently accessed appraisal frame with the highest mood-adjusted intensity”. So “if and event is appraised as equally hope and fear provoking, the agent will focus on its fears if its mood state contains more fear than hope” (p. 82).
- Coping as a gate (p. 82). “Coping acts as a sequential gate-keeper that sanctions the action congruent with the current appraisal pattern.” Strategies are chosen by controllability and changeability (p. 83).
- The bird (pp. 84–87). Surprise leads to attending, then fear from an uncontrollable threat, then avoidance. Distance raises controllability and turns fear to anger, then planning and hope. The bird caught in the other actor’s hair removes the threat, and help follows. The model’s “anger” at the bird’s injury is a miss.
- Common appraisal as coordination (p. 89). “Appraisal theory suggests a general set of criteria and control strategies that can be uniformly applied to characterize, inform, and coordinate the behavior of heterogeneous cognitive functions.”
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | A single-level appraisal process suffices to explain fast and slow emotional dynamics | moderate | architectural argument plus one encoded episode |
| C2 | Apparent ordering of appraisal checks is a by-product of the time course of inference, not of appraisal | moderate | argued against Scherer; illustrated |
| C3 | Coping acting on beliefs, desires and intentions, followed by reappraisal, generates emotional transitions | moderate | the bird encoding; earlier vignette studies (Gratch and Marsella 2005) |
| C4 | Multi-process appraisal models are hard to falsify | strong (as a methodological point) | argued; the authors grant EMA’s own timing parameters are free |
Concepts
- causal interpretation: the agent’s working-memory representation of the agent–environment relation, holding states, actions, causal links, beliefs, intentions, probabilities and utilities.
- appraisal frame: the continuously updated appraisal values for one proposition, with an emotion label and intensity.
- mood-adjusted intensity: a frame’s intensity plus the mood component of the same label, the quantity on which frames compete.
- coping: strategies that maintain desirable or overturn undesirable in-focus appraisals by acting on the world or on attention, belief, desire or intention.
Connections
Follows Smith and Lazarus (1990) and Lazarus (LIT-550, unread), and is set against Scherer’s sequential checks (Scherer 2001; the record reads his model in LIT-1451) and Smith and Kirby’s two-process model. It takes the interrupt view of emotion from Simon (1967) and the law of situated meaning from Frijda (LIT-704, unread). It is a sibling of Lewis’s dynamic-systems model (LIT-1455), which also refuses a fixed appraisal sequence but makes the competition one of attractors, not of discrete frames.
Bearing on the record
On the negotiation reading (CLAIM-440, part 1). EMA has an explicit, implemented arbitration among concurrent evaluations: frames “each competing to determine the agent’s coping response” (p. 81), and a rule for the winner. To that extent it anticipates “appraisal as the process by which competing perspectives negotiate”. It anticipates the Moors form of the idea (LIT-1434), though, not the record’s. The difference is the same, and sharper.
- The competition is on one scalar. A frame’s intensity is “a function of the likelihood of the threat times the utility of the goal that is threatened” (p. 84), mood-adjusted. Selection is an argmax, with recency breaking ties. That is a common currency by construction, the compensatory regime of CLAIM-435 and CLAIM-436 with nothing else possible. Sen-type failure of the negotiation to produce an ordering cannot arise in EMA.
- The competitors are concerns, not perspectives. What competes is evaluations of propositions, all on the one agent’s utility. EMA’s appraisal variable called “perspective” (p. 80) is not the record’s perspective. It names whose utility function a proposition is judged by, and the paper uses only the agent’s own. The record should not borrow the word from EMA without saying so. A multi-perspective EMA, with frames judged under different utility functions and no common scale between them, is exactly what the proposal would need and what EMA does not provide.
- No authority distinct from intensity. In EMA the precedence a frame claims (its intensity) is the authority it is granted (selection for coping). CLAIM-283’s distinction has no room in it, except through recency, which grants authority on another ground.
On the marks of potential (TERM-076).
- Priority is the argmax rule.
- Persistence comes only through mood. A frame type that has won often raises the mood component of its label, and that biases later competitions toward the same label. That is a concrete mechanism for persistence, and it works as it should, by a slow aggregate biasing a fast competition.
- Resumption is not modelled, and the recency rule works against it. After an interruption, the interrupting frame, being most recently accessed, wins ties. An EMA agent returns to an interrupted concern only if it re-accesses it or mood favours its label. Frijda’s resumption would need something EMA lacks: a standing ranking whose authority outlasts the frames it wins.
On valence. EMA’s valence is per proposition: desirability positive or negative, on the agent’s utilities. It is Carver’s “each goal loop its own valence” (LIT-1441 through LIT-1449) again, the view the proposal sets against one valence for the potential ordering. EMA’s mood is the nearest thing to a system-level valence, and it is a sum by label, not an ordering.
What the record can take. Two things. First, desire-related coping: “Distance/Mental disengagement: Lower utility attributed to a desired but threatened state” (p. 82). Here appraisal revises the ranking itself, not only which ranking wins. A negotiation that can change its parties’ rankings is a case the proposal has not considered, and it bears on how an ordering comes to prevail (by winning, or by others conceding). Second, the engineering point of p. 89. Appraisal as a common characterization across heterogeneous modules, “By casting the state of each module in these same general terms, it becomes possible to craft general control strategies that apply across modules”, is Veit’s sharing job for a common currency (THEORY-101), built in by design. It is a worked example of the second half of the proposal’s part 2.
Candidly. EMA does not threaten the proposal’s novelty beyond what Moors already does. It shows what an arbitration that presupposes the currency looks like when implemented, and so it serves as a baseline model against which the record’s ordinal, multi-perspective negotiation could be specified.
Limitations
One encoded episode, with timing parameters set by hand (p. 88), so the illustration shows sufficiency, not fit. The anger prediction at t9 is wrong by the authors’ own reading (p. 87). Mood is a placeholder for “sub-symbolic (brain or bodily) processes … that we do not yet know how to model” (p. 81). The world is assumed fully observable in the present (p. 79, note 3).
Open questions
Whether frames judged under other agents’ utilities could compete with the agent’s own without first being mapped to a common intensity. How EMA would represent a concern that keeps returning after being displaced, other than by mood. Whether desire-related coping converges, or can cycle when two frames each lower the other’s supporting utility.