nucleation

LIT-1447: EMA: A process model of appraisal dynamics

3 min read

Properties4
StatusActive — read 2026-10-23 (NOTE-1100): in full, from the published version on a University of Kentucky course reading list, checked against the authors' manuscript on Gratch's USC page. Worth reading as the worked computational case of appraisal as a continuous process over the agent's causal interpretation, with an explicit arbitration rule. Appraisal is a single, fast, parallel mapping from that interpretation to appraisal variables. Its dynamics come from the slower perception, inference and action that change the interpretation, and from coping, which alters beliefs, desires and intentions and so drives reappraisal. Appraisal frames, one per proposition, compete for the coping response. The winner is the most recently accessed frame with the highest mood-adjusted intensity, and intensity is a decision-theoretic product of likelihood and utility. That is arbitration on a common currency among concerns, not a negotiation among perspectives.
Tagsemotion-and-affect, cognition, agency
SupportsCLAIM-440 — An organization's appraisal is the negotiation among its perspectives, a common currency is that negotiation's admitting an ordering of the options it faces, and goal-conduciveness valence is a regime in which an ordering of the organization's own states by their bearing on its maintenance prevails in the negotiation; so valence presupposes contested rankings, not frustrated demands, its sign is whether the organization runs ahead of or behind a reference course along the potential ordering, its intensity is how far the negotiation's decisions follow that ordering, and ambivalence, where no ordering prevails, differs from neutrality, where one prevails and the organization keeps to its course
Cited as “paper” byNOTE-1100 — EMA: A process model of appraisal dynamics

Graph View

Stacy C. Marsella & Jonathan Gratch (2009), Cognitive Systems Research 10(1):70–90 — DOI-10.1016/j.cogsys.2008.03.005

Key takeaways

  • One appraisal process, many inference processes (pp. 71–74). The authors reject multi-level appraisal (Scherer’s sequential checks, Smith and Kirby’s two processes) as confusing appraisal with inference. “We distinguish between the construction of this representation, which may be slow and sequential, and its appraisal, which is fast, parallel and automatic” (p. 71).
  • A cycle of appraisal, coping and reappraisal (p. 76), following Lazarus (LIT-550): “we assume a cyclical relationship between appraisal, coping and re-appraisal.” Inference changes the interpretation, and “they indirectly trigger automatic reappraisal” (p. 76).
  • The causal interpretation (pp. 77–79). A blackboard holding the agent’s working memory, built on decision-theoretic planning: states, actions, causal links, beliefs, intentions, probabilities and utilities. “Utilities represent the preference agents’ have for states” (p. 79). Appraisals “do not change the causal interpretation but provide a continuously updated “affective summary” of its contents” (p. 80).
  • Appraisal variables (p. 80). Relevance, perspective, desirability, likelihood, expectedness, causal attribution, controllability and changeability, computed per proposition. “Perspective” is whose utilities are used, the agent’s own or another agent’s as the agent represents them; the paper uses only the agent’s own.
  • Arbitration (pp. 81–82). EMA “maintains multiple appraisal frames (one for each proposition in the causal interpretation) each of which are labeled with a specific emotion type and intensity, and each competing to determine the agent’s coping response” (p. 81). The winner is “the most recently accessed appraisal frame with the highest mood-adjusted intensity” (p. 81). Mood, an aggregate of frames by emotion label, “bleeds over” into appraisal (p. 81). Intensity comes from “a decision-theoretic calculus” of likelihood times utility (p. 84).
  • Coping (pp. 82–83). Strategies that work “in the reverse direction of the appraisal that motivates them” (p. 82) act on attention, belief (shift responsibility, wishful thinking), desire (distancing, positive reinterpretation) and intention (planning, procrastination, resignation, avoidance). “EMA proposes strategies in parallel but adopts them sequentially. A set of preferences resolves ties” (p. 83).
  • Limits (pp. 87–88). Some predictions miss, such as anger at the bird’s injury where the video suggests concern for it. Time courses are “free parameters that allow us considerable latitude in fitting our model to data” (p. 88).

Standing in the record

Filed on 2026-10-23 for the record’s proposal that appraisal is the negotiation among an organization’s perspectives (CLAIM-440, TERM-076). EMA is the clearest working model of appraisal as a continuous process with explicit arbitration among concurrent evaluations. Its arbitration is winner-take-all on one scalar, mood-adjusted intensity, so it presupposes the common currency that the proposal treats as an achievement. Its competitors are evaluations of propositions, not rankings held by perspectives. Coping that lowers a desire’s utility is the one place where the ordering itself is revised. The bearing is set out in NOTE-1100.

Not flagged for the Anthology of the SOTA: EMA is an architecture for virtual humans and carries no instruction for machine-learning practice. Whether the anthology wants it as a computational model of emotion for agents is the anthology’s question.

Backlinks