RESEARCH DOSSIER
Constructed Emotion and Predictive Affective Processes
Abstract. Constructed emotion theories explain differentiated emotional episodes through interactions among body regulation, predictive inference, situated conceptualization, and learned categories. This critical narrative monograph reconstructs the historical transformation from bodily-feeling and discrete-emotion theories to psychological construction and active interoceptive inference. It analyzes experimental manipulations of situated emotion, affective taxonomy, autonomic meta-analysis, neuroimaging syntheses, cross-cultural expression studies, linguistic mechanisms, temporal dynamics, individual differences, computational models, and the 2025 debate over basic and constructed emotions. The evidence supports substantial context dependence, distribution of neural contributions, and heterogeneity within named emotion categories, while the specific computational mechanisms and necessity claims of predictive construction remain underdetermined. The synthesis establishes testable model comparisons and a research agenda for causal identification, replication, and cautious human–artificial comparisons. This work is a critical narrative evidence synthesis, not an original experiment, registered systematic review, or meta-analysis.
Keywords: constructed emotion; theory of constructed emotion; predictive processing; affective inference; interoception; allostasis; active inference; emotion concepts; categorization; psychological construction; situated conceptualization; emotion prediction; affective neuroscience; autonomic specificity; emotion granularity; cross-cultural emotion; basic emotion theory; computational affective science; human–artificial emotion inference
Introduction: The Explanatory Problem of Constructed Emotion
Constructed emotion concerns how an episode identified as fear, anger, sadness, amusement, or another emotion emerges from interacting bodily, perceptual, conceptual, and situated processes. Its central problem is explanatory: what must be organized for a particular experience and course of action to count as an instance of an emotion, and how can investigators distinguish that organization from the categories used to describe it? The theory of constructed emotion proposes that a brain regulates the body and anticipates sensory consequences while categorizing an ongoing situation through concepts derived from experience. The relevant category is implemented as a context-sensitive event, not merely recovered from a fixed physiological template. Barrett (2017) Barrett et al. (2025)
This monograph investigates construction as an account of emotion generation and predictive affective processing, rather than treating all affective phenomena as instances of the same mechanism. Affect refers here to valenced and arousal-related aspects of experienced bodily and environmental significance; emotion refers to an episode categorized in relation to a situation, action possibilities, and social meanings. Predictions are model-dependent expectations about sensory states and their causes; categorization is the application or construction of conceptual knowledge that makes a current instance intelligible. These analytic definitions make it possible to compare theories that differ in which components are primitive, necessary, or causally sufficient. Russell (1980) Barrett (2017)
The evidence encompasses conceptual analyses, experimentally manipulated situated interpretations, cross-cultural categorization tasks, neuroimaging, autonomic meta-analysis, experiential sampling, and computational accounts of interoception. Each method answers a different question. An fMRI correlation can locate a distributed contribution to reported affect without proving that a network implements an exact Bayesian algorithm; a forced-choice facial-labeling study can reveal how participants use categories without directly reading another person's hidden emotional state. The goal is to identify convergent and discriminating results and specify where theory still outruns available measurement. Lindquist et al. (2012) Barrett et al. (2019) Siegel et al. (2018)
The approach is a critical narrative synthesis of traceable original studies and relevant systematic or theoretical reviews. Historical statements are distinguished from currently supported conclusions, and interpretations are connected to research designs rather than inferred from the status of an influential theory. No new human-subject data, prospective review protocol, preregistered systematic search, meta-analysis, or original computational fit is reported. The final chapters address testable predictions, explanatory alternatives, conceptual scope, and possible comparisons with artificial systems as an emerging methodological question. van Heijst et al. (2025) Barrett et al. (2025) Hoemann et al. (2021)
From Bodily Feeling to Psychological Construction
William James's 1884 essay placed bodily changes at the center of the explanation of emotion. Its enduring scientific contribution was the demand that emotion theories account for bodily experience rather than regarding feeling as an independent faculty. Later two-factor reasoning associated with Schachter and Singer emphasized how interpreted arousal could contribute to different emotional experiences, while subsequent appraisal approaches investigated the relation between significance and emotional differentiation. These traditions generated different causal hypotheses and cannot be collapsed into one uninterrupted history of predictive processing. James (1884) Schachter & Singer (1962)
During the twentieth century, models of discrete emotions proposed evolved configurations with characteristic elicitors, rapid organization, and expressive or physiological properties. Ekman's argument for basic emotions offered one influential program specifying features such as adaptive significance, quick onset, and coordinated response organization. In parallel, Russell's circumplex analyzed how self-reported affective descriptors relate along pleasantness and activation dimensions. The former investigates kinds of emotion; the latter maps a structure of ratings and concepts. Neither observational space automatically settles the processes by which a given emotional event is generated. Ekman (1992) Russell (1980)
Psychological construction approaches developed the proposition that psychological events recruit more general processes whose combinations produce varied emotional instances. Situated conceptualization models added that a concept is enacted with sensory, action-related, and contextual features rather than residing as a static dictionary entry. The 2011 neuroimaging experiment by Wilson-Mendenhall and colleagues manipulated physical danger versus social evaluation while participants experienced fear or anger, offering an empirical way to ask how the same category varies with situation. Such evidence speaks to heterogeneity, although interpretation still depends on what participants simulated and what the imaging contrast measured. Wilson-Mendenhall et al. (2011)
Barrett's 2017 formulation integrated construction with interoceptive prediction and active inference, shifting the focus toward how a brain's internal model coordinates future body needs and incoming sensory information. The historical trajectory is therefore a convergence of questions about bodily feeling, categorization, situated cognition, and computational regulation. The 2025 debate with evolutionary basic-emotion accounts shows that even basic terms such as emotion and feeling remain theoretically contested. A sound historiography retains the distinct claims of each research program and tests their consequences rather than retrospectively awarding the entire field to one lineage. Barrett (2017) van Heijst et al. (2025) Barrett et al. (2025)
Defining Affect, Emotion, Concept, and Category
Affect, in this discussion, denotes the ongoing organization of valence, activation, and experienced significance without assuming that a named emotion category is already present. Valence identifies a pleasant–unpleasant dimension and arousal a dimension of activation. Russell's circumplex is influential because these dimensions structure many affect judgments; it does not show that the psychological content of grief, anger, relief, or pride can always be recovered from two coordinates. Investigators must identify whether their measure concerns subjective ratings, physiological activation, semantic similarity, or a particular time-varying episode. Russell (1980) Cowen & Keltner (2017)
An emotion concept organizes knowledge about possible situations, bodily changes, expected consequences, meaningful action, and social interpretation. Under constructionist accounts, categorizing an episode as anger can help constitute its experienced meaning and shape subsequent action; it does not require that the person consciously chooses a verbal label before the episode begins. A category can support many physically and phenomenologically different instances. Situational meaning and category membership remain distinguishable in analysis even when they are tightly coupled during experience. Wilson-Mendenhall et al. (2011) Lindquist (2017) Barrett (2017)
Emotion perception is the inference of another person's situation or state from incomplete evidence such as facial movements, vocal features, posture, words, and contextual history. A displayed scowl may inform an anger judgment, yet the same movement can participate in effort, concentration, or non-angry communication. Emotion experience is also a separate target from emotion recognition: success in selecting a conventional label for a posed expression does not establish that the label describes the expresser's lived event. Research on contextual perception and facial expression has made this construct distinction indispensable. Barrett et al. (2011) Barrett et al. (2019)
Operational definitions should identify inputs, temporal grain, measurement procedure, and criterion variable. For example, an experiment can manipulate linguistic access to category words and then measure changes in emotion perception; another can alter a bodily cue and estimate effects on reported arousal; a third can ask whether held-out physiological measurements distinguish emotion categories across situations. Each design probes a restricted component of the general account. The scientific value of constructed emotion depends on whether such components generate predictions that can be separated from their competitors under transparent measurement assumptions. Lindquist (2017) Siegel et al. (2018) Hoemann et al. (2021)
The Generative Architecture of Predictive Affective Processing
Predictive-processing models describe nervous systems as using internal generative models to anticipate sensory input and update in response to mismatches between expected and observed signals. In a hierarchical formulation, higher-level expectations constrain lower-level estimates, and prediction errors influence learning according to estimated reliability. Friston's free-energy account is an influential mathematical framework for perception, action, and learning; applying its formal machinery to an emotional episode requires specifying latent states, sensory channels, model parameters, and measurable outcomes. The general usefulness of a predictive vocabulary is not by itself evidence for one exact implementation of emotion. Friston (2010) Seth (2013)
In Barrett's constructed-emotion account, predictive regulation of the body and categorization of the sensed situation are linked. Prior experiences supply concepts or probabilistic regularities used to anticipate what a bodily condition and environment signify and which actions could meet anticipated needs. Incoming signals constrain this process, while ongoing regulatory actions change the signals that will arrive. Thus an emotional episode emerges within perception–action loops rather than following a single mandatory path from stimulus to feeling to response. This proposal gives allostasis and concept learning explanatory roles, while leaving room for experimentally distinct implementations. Barrett (2017) Barrett & Simmons (2015)
Prediction error does not have one straightforward emotional interpretation. A surprising bodily event, a violation of a social expectation, and a discrepancy between predicted and observed reward have different experimental structures and neural dependencies. Precision, in computational accounts, refers to how strongly a discrepancy should influence updating given estimated uncertainty; it must not be conflated with a person's subjective confidence or the moral importance of a situation. Formal experiments should manipulate expected contingencies and reliability while recording changes in perception, action, physiological control, and model fit over time. Friston (2010) Seth & Friston (2016)
The account also distinguishes a functional explanation from an anatomical claim. A computational model may specify recurrent estimation without locating a unique site at which emotion is produced. Distributed systems can contribute to prediction, bodily regulation, memory, and category deployment, and several models may fit a restricted behavioral dataset. Consequently, strong evidence requires comparisons with simpler models, out-of-sample prediction, and perturbations that separate hypothesized feedback from alternative causes. Computational vocabulary becomes explanatory through constrained predictions and model comparison, not through re-describing every observation as the reduction of surprise. Kober et al. (2008) Kleckner et al. (2017) Pessoa (2008)
Interoception, Allostasis, and Anticipatory Body Regulation
Interoception concerns sensing, integrating, and representing internal bodily conditions. Signals associated with the cardiovascular, respiratory, gastrointestinal, endocrine, and other regulatory systems contribute to bodily experience, yet they are sampled through different pathways and tasks. Allostasis refers to anticipatory adjustment of physiological conditions in response to expected demands. Constructed-emotion accounts place the prediction of body-relevant costs and needs near the center of emotional generation, while interoception as a broader field includes many processes that occur without named emotions. Barrett & Simmons (2015) Critchley & Garfinkel (2017)
Barrett and Simmons proposed that interoceptive experience reflects descending expectations constrained by ascending visceral signals. Seth's account of interoceptive inference and the joint analysis by Seth and Friston express related computational possibilities concerning bodily prediction and regulation. These are mechanistic proposals that integrate experimental observations but should be distinguished from direct demonstrations of the precise neural messages or parameter values posited in a specific active-inference model. A body signal can contribute to emotion without being a unique signature of an emotional category. Seth (2013) Seth & Friston (2016) Barrett & Simmons (2015)
Kleckner and colleagues reported evidence consistent with a large-scale cortical system involved in interoception and allostasis. Their neuroanatomical and connectivity findings support an integrated network view and are relevant to candidate mechanisms for anticipatory regulation. The research does not demonstrate that a specific subjective category is encoded exclusively in a particular node, or that every observed network fluctuation represents one fixed mathematical prediction-error signal. Studies of functional architecture and studies of experience must be connected by task-level predictions and convergent interventions. Kleckner et al. (2017) Lindquist et al. (2012)
Interoceptive assessment introduces its own measurement problem. Performance in heartbeat-related tasks, self-reported sensitivity to bodily sensations, and metacognitive calibration can reflect different processes, and their relations depend on methodology. Physical exercise, pharmacological conditions, respiratory manipulation, and social anticipation may influence bodily input in different ways. An informative emotion study therefore records which channel was manipulated, whether the person actually sensed the change, which aspects of experience shifted, and whether the inferred link survives adjustments for demand characteristics and context. These distinctions prevent a predictive interpretation from becoming immune to contrary observations. Critchley & Garfinkel (2017) Hoemann et al. (2021)
Conceptual Knowledge and Situated Emotion Construction
Situated conceptualization treats concepts as flexible assemblies of remembered sensory, motor, affective, and contextual regularities. The meaning of fear during physical danger can differ from fear during anticipated social rejection, even when both occasions are named with the same word. The constructionist proposal is that concept knowledge enables a present situation to be categorized in a form that coordinates perception, action, and subjective experience. It predicts within-category heterogeneity that follows changes in relevant situations rather than random physiological noise. Wilson-Mendenhall et al. (2011) Wilson-Mendenhall et al. (2013)
Wilson-Mendenhall and colleagues' 2011 study varied imagined physical-threat and social-evaluation contexts while investigating neural contributions to anger and fear. Its design illustrates a discriminating logic: if a category always depends on one invariant regional response, changing context should leave a characteristic neural pattern relatively stable; if emotion construction recruits contextual assemblies, within-category responses should vary systematically. The study contributes evidence for situated variation, although mental imagery introduces individual differences in vividness, beliefs, and task compliance. Replication across naturalistic contexts and participant groups is needed to estimate generality. Wilson-Mendenhall et al. (2011)
The 2013 investigation of situated emotional experience extended the question of context dependence, while work on neural correlates of core affect examined dimensions shared across several named emotions. These results favor research programs that represent both common affective properties and variable situational configurations. Their interpretation remains sensitive to choices of stimulus, rating scales, neural contrast, and how category labels guide participants' imagination. A report that two emotional occasions share valence or arousal does not imply that they share identical mechanisms, and contextual differences do not establish that every proposed component is necessary. Wilson-Mendenhall et al. (2013) Wilson-Mendenhall et al. (2013)
Conceptual organization is also socially learned. Speakers acquire words, narratives, scripts, and norms that may alter what a particular feeling is expected to mean and which actions count as appropriate. Cultural variability therefore belongs to the explanatory model as a possible source of changed concept deployment. At the same time, recurrent bodily constraints and learned contingencies can stabilize affective patterns without producing perfectly fixed categories. The scientific challenge is to estimate how much variation is predicted by language, situational interpretation, shared biology, and individual history in the same study. Gendron et al. (2014) Lindquist (2017) Barrett et al. (2025)
Language, Labels, and Emotional Meaning
Language provides stable and revisable resources for distinguishing emotion categories, telling others about experiences, and making predictions about socially meaningful situations. Constructionist proposals assign concepts an active role in emotional experience, but the strongest version of a linguistic necessity claim exceeds the evidence. A person can show affective behavior without retrieving an explicit word, and emotion-related understanding develops through embodied interaction as well as speech. The empirical question is whether access to particular concepts changes measurable aspects of an otherwise comparable affective episode, by how much, and in which populations. Lindquist (2017) Barrett (2017)
Lindquist's review brought together behavioral, neural, and lesion-related findings to argue that emotion words can help shape perception and experience. Such evidence is important because causal manipulations of verbal or semantic access provide more specific tests than observing a correlation between vocabulary and questionnaire scores. Nevertheless, a semantic task may change attention to stimuli, response selection, or reporting standards without changing the underlying feeling in the same way. Designs need multiple outcome measures to distinguish categorization effects from altered response criteria or instructional demand. Lindquist (2017) Barrett et al. (2011)
The semantic structure of emotional experience is itself complex. Cowen and Keltner analyzed reports elicited by a broad set of videos and found category-like clusters connected by graded transitions. Their result challenges the adequacy of an exclusively two-dimensional summary for the selected reports, while also demonstrating that reported categories are not isolated natural partitions. It cannot directly establish what bodily mechanisms generated the states elicited by the videos, since the method mainly characterizes the structure of judgments. Affective taxonomy and generative emotion theory must therefore be connected through independent causal measures. Cowen & Keltner (2017) Russell (1980)
Emotion concepts also differ in everyday granularity. Some people distinguish irritation, disappointment, apprehension, and grief in a given context, while others use broad descriptions of feeling good or bad. Reviews of emotional expertise show that differentiation, clarity, complexity, and awareness have related but distinct operationalizations. Repeated sampling methods can themselves affect observed granularity, making measurement reactivity part of the object under study. Reliable interpretation requires repeated reports, transparent scale construction, and independent outcomes beyond self-description. Hoemann et al. (2021) Hoemann et al. (2021)
The Experimental Logic of Situated Affective Prediction
Experiments aimed at construction should vary candidate causes while preserving an interpretable comparison. A paradigm can hold a bodily perturbation approximately constant and alter its situational explanation; another can change conceptual access with the same set of external cues; a third can manipulate learned expectations about internal sensation. The decisive outcome is not that participants report different words under different instructions. It is whether changes in experience, action, physiology, and perceptual judgment follow a specified causal pattern that a rival theory would predict less well. Schachter & Singer (1962) Wilson-Mendenhall et al. (2011)
Schachter and Singer's 1962 experiment remains historically important for proposing interactions between physiological activation and situational cognition. Its methodological lesson is to study interpretive conditions rather than treating bodily arousal as a sufficient explanatory variable. The study predates the current computational constructionist theory and should not be described as a direct validation of it. The relevant research logic persists in modern experiments that measure expectations, contextual interpretations, and outcomes separately, with controls for induced arousal and participants' awareness of experimental hypotheses. Schachter & Singer (1962) Barrett (2017)
An interoceptive prediction experiment needs an explicit generative model: what signal is expected, what uncertainty governs the expectation, what error is produced by an unanticipated observation, and which dependent variables should move in response? Measures might include trial-wise expectancy ratings, respiratory or cardiac recordings, autonomic responses, and reports of feeling or categorization. A temporal model can compare how rapidly predictions update against the delayed changes in peripheral physiology. Predictive processing earns explanatory credit only when these predicted dependencies survive controlled tests and formal model comparisons. Seth & Friston (2016) Friston (2010)
Laboratory outcomes require ecological interpretation. An imagery experiment can isolate conceptual content but may differ from unprompted emotional episodes; a posed face task offers standardized stimuli while encouraging an observer to treat categories as more stable than they are in daily social life. Intensive longitudinal designs can measure within-person variability across situations, but self-report frequency and vocabulary affect data quality. Triangulation across controlled experiments, ambulatory sampling, behavioral observation, and physiological monitoring helps identify causal features that persist beyond a particular task format. Wilson-Mendenhall et al. (2013) Barrett et al. (2019) Hoemann et al. (2021)
Autonomic Variation and the Emotion Fingerprint Hypothesis
The emotion fingerprint hypothesis predicts recurrent configurations of peripheral physiological activity that discriminate named categories across relevant conditions. In stronger forms it implies characteristic autonomic profiles for anger, fear, sadness, or other discrete emotions. A constructionist account anticipates greater heterogeneity because bodily mobilization reflects the demands of action, context, individual history, and regulatory conditions. These predictions are testable but depend on defining what degree of category specificity, temporal stability, and contextual invariance is expected. A weak statistical tendency is different evidence from an indispensable biological signature. Ekman (1992) Siegel et al. (2018)
Siegel and colleagues' 2018 meta-analytic investigation of autonomic features compared evidence relevant to fingerprints and emotion populations. The review is important because it addressed cross-study heterogeneity and the conditional nature of autonomic associations rather than relying on a memorable individual demonstration. A meta-analysis aggregates measures that vary in elicitation, sampling, and instrumentation; even when a mean category effect appears, it does not identify a universal within-person pattern. The investigation supports evaluating categories as distributions of context-sensitive instances while leaving room for recurrent tendencies under restricted conditions. Siegel et al. (2018)
Physiological measurements require deliberate design. Heart rate, skin conductance, respiration, heart-rate variability, facial electromyography, and hormonal assays have different latencies, noise sources, and relations to action. A pattern discriminating fear from anger during seated film viewing might lose accuracy during exercise, social confrontation, or anticipatory decision making. Classifiers should be trained and tested across people, elicitation methods, and contexts, with calibration data separated from evaluation data. Otherwise apparently accurate emotion recognition may exploit confounds such as physical movement or stimulus properties. Critchley & Garfinkel (2017) Barrett et al. (2019)
The absence of a singular fingerprint cannot establish that emotion is unrestricted or arbitrary. Embodied agents face real physiological constraints: preparing to flee differs metabolically from remaining still, and chronic states may shift baseline regulation. What matters is whether a candidate physiological configuration is specific to the named psychological category rather than to an action demand, level of arousal, or experimental manipulation. Competing hypotheses should identify invariance targets and report distributions, uncertainty, and failures as well as mean differences. Barrett & Simmons (2015) Siegel et al. (2018)
Neural Systems and the Limits of Localization
Neural construction theories propose that emotional episodes emerge from interacting networks participating in sensation, regulation, conceptual memory, attention, salience, and action. An anatomical localization hypothesis instead expects particular emotional categories to map reliably onto specialized regions or circuits. These are families of hypotheses with variations in their specificity, not a single binary test. Kober and colleagues' meta-analysis identified distributed functional groupings across neuroimaging studies, and Lindquist and colleagues' later meta-analytic review assessed evidence about category-specific and shared neural contributions. Both invite caution about equating one labeled experience with one circumscribed location. Kober et al. (2008) Lindquist et al. (2012)
The amygdala, insula, orbitofrontal cortex, cingulate regions, sensory areas, and memory-related systems can participate in emotionally significant processing, but their contributions depend on the task and network context. Pessoa's review argued for dynamic coalitions across systems commonly classified as cognitive and affective. Functional coactivation or network connectivity does not by itself tell us whether a process is a sensory prediction, a category representation, a regulatory command, or a consequence of felt significance. Brain maps become explanatory when paired with behavior, temporal measurement, and well-matched perturbations. Pessoa (2008) Kober et al. (2008)
Interoceptive and allostatic circuitry supplies one plausible bridge between bodily regulation and experience. Kleckner and colleagues characterized a large-scale human system connected to these functions. This contributes anatomical plausibility to predictive affective accounts without demonstrating an emotion-specific command center. Neural evidence is especially vulnerable to reverse inference: observing a region during an emotion task does not entail that the region's activation denotes that emotion or that computational function. Investigators should specify which regions, directed connections, signals, and time scales are expected under competing models. Kleckner et al. (2017) Barrett & Simmons (2015)
Multivariate analyses can classify patterns more successfully than single-region analyses, yet generalization matters more than within-dataset accuracy. If a pattern discriminates emotion labels only for one film set, it may reflect shared sensory input or narrative content. Cross-context validation and experimental intervention can test whether the pattern carries an emotion-relevant variable. Constructionist neurobiology is strongest when it predicts structured variation and re-use of mechanisms, then shows that these predictions outperform equally plausible category-specialist or generic arousal accounts. Lindquist et al. (2012) Pessoa (2008) Wilson-Mendenhall et al. (2013)
Faces, Voices, and Contextual Emotion Perception
Emotion perception is a case of inference under uncertainty. An observer combines information from facial movement, vocalization, body position, words, relationship history, and current circumstances. A smile, frown, or widened eye can participate in communicative and regulatory behavior without unambiguously disclosing the maker's internal state. Research on context demonstrates that visual surroundings, voices, and language change interpretations of facial movement. These observations are relevant to construction because an emotional category is applied to a situated pattern rather than directly read from one isolated physical sign. Barrett et al. (2011) Barrett et al. (2019)
Barrett, Adolphs, Marsella, Martinez, and Pollak reviewed evidence about inferring emotions from facial movements and highlighted substantial limitations of simple face-to-state mapping. Their conclusions have practical relevance to automated affect recognition and psychological assessment. The review does not imply that faces carry no emotional information; it asks how diagnostic that information is and whether a particular label is justified without context. Predictive research can specify priors, facial and contextual likelihoods, and behavioral consequences of different assumptions, then validate them against independent descriptions and naturalistic observations. Barrett et al. (2019)
Contemporary basic-emotion research has also expanded beyond the six labels often used in forced-choice studies. Keltner and colleagues described advances in the study of emotion-related expression involving multiple channels and more varied emotional categories. This literature offers an essential alternative to claims that every expressive pattern is entirely unconstrained. A viable comparison must separate stereotyped posed displays, spontaneous social behavior, multimodal expression, and perceivers' judgments. Each may exhibit a different degree of consistency or cultural generalization. Keltner et al. (2019) Barrett et al. (2019)
For artificial systems, the methodological implication is immediate: a facial video or voice recording yields features from which a model estimates states under uncertainty, not direct access to private experience. Evaluations should document sampling populations, contextual coverage, domain shift, protected attributes, and calibration. Systems should represent the possibility of mixed or unclassifiable signals rather than forcing a confident discrete label for every input. These standards follow from the science of human emotion inference; they do not establish how, or whether, a digital system itself experiences affect. Barrett et al. (2019) Gendron et al. (2014)
Culture, Learning, and Cross-Population Generalization
Emotional categorization is learned within social and cultural practices, and theories that posit universally recognized expressions must be evaluated with methods that minimize imported conceptual cues. Cross-population studies differ in translation, sampling, exposure to media, interviewer expectations, and whether participants must choose among researcher-provided labels. These design details matter because a forced-choice list can change how stimuli are grouped. Constructionist models predict some cultural variation in concept use while allowing shared bodily and social constraints to produce recurrent similarities. Gendron et al. (2014) Barrett et al. (2011)
Gendron and colleagues compared emotion-related sorting of posed facial expressions among U.S. participants and Himba participants in northwestern Namibia. In the absence of supplied emotion concepts, the sorting patterns differed from the conventional presumed universal pattern; providing conceptual cues increased similarity while cultural differences remained. This original result is informative because the manipulation targets a potential methodological cause of apparent universality. It cannot establish that no affective regularities cross cultures, nor can a single group comparison summarize global emotional diversity. Gendron et al. (2014)
Related work on perceiving emotion from vocalizations found evidence of cultural relativity and shows why expressive channels should be tested separately. Vocalizations, speech prosody, and facial configurations have different learned conventions and physical production constraints. Robust comparative studies should use participant-generated categories and open-ended interpretation alongside controlled recognition tests, sampling within cultures as carefully as between them. The research target is the distribution of inference strategies and meanings under specified circumstances, not a competition between claims of absolute universality and unrestricted variation. Gendron et al. (2014) Keltner et al. (2019)
Longitudinal developmental evidence could clarify how categories become available and how linguistic acquisition changes emotion differentiation. Cross-cultural experiments can also test whether learned conceptual labels alter the same physiological or action outcomes across populations, which would provide a stronger causal bridge than observed verbal similarity. Translation equivalence, measurement invariance, and culturally appropriate elicitation must be demonstrated rather than presumed. Such studies directly address whether the generative account predicts what people feel and do, or only how they describe other people's expressions. Lindquist (2017) Hoemann et al. (2021)
Affective Dynamics Across Time and Context
Emotion unfolds as a process, with changes in anticipation, bodily activation, appraisal, action readiness, subjective labeling, and recovery. A single questionnaire score collapses this temporal organization. Constructionist and predictive approaches treat current predictions as dependent on accumulated experience and on the consequences of previous actions, implying that an episode may be reorganized as new cues arrive. To test the proposal, researchers need repeated measures of the same person across situations and a model specifying whether context and expected consequences alter transitions between states. Barrett (2017) Seth (2013)
Predictive processing is especially relevant when the sequence of an affective episode is uncertain. A person awaiting a medical test may anticipate harm, experience elevated bodily activation, revise expectations when receiving reassuring news, and still retain unpleasant feelings. Different components may recover at different rates. In such cases, using the final verbal label as the sole dependent variable loses the history of model updating. A dynamic design distinguishes anticipation from surprise, intensity from duration, and physiological adaptation from retrospective interpretation. Barrett & Simmons (2015) Critchley & Garfinkel (2017)
Ecological momentary assessment offers repeated observations in everyday settings, potentially improving validity for context-sensitive emotions. Yet prompting participants to label feelings changes attention and may encourage differentiation, a methodological effect documented in work on emotional granularity. Sensor streams add temporal detail but can introduce artifacts from posture, movement, location, and device reliability. The strongest analyses examine within-person associations, account for autocorrelation and missingness, and report whether model parameters remain stable when occasions or participants change. Hoemann et al. (2021) Hoemann et al. (2021)
Constructionist accounts can gain specificity by predicting which changing contexts should produce particular trajectories. A model in which learned categories help guide preparation might predict rapid reorganization when the meaning of an event changes, even if some peripheral signals persist. A simpler arousal-only model could predict a slower decrease related to bodily recovery. Observing such dissociations would be informative, although alternative appraisal or learning models can predict them too. Temporal resolution is therefore a basis for discriminating explanations, not an automatic confirmation of construction. Wilson-Mendenhall et al. (2013) Russell (1980)
Individual Differences, Emotion Granularity, and Expertise
People differ in the number and structure of distinctions they regularly make among emotional experiences. Emotional granularity describes the degree to which a person's reports discriminate emotional categories across occasions, while clarity, awareness, and complexity capture related but different competencies. Constructionist theory suggests that accessible concepts can help organize situated experience; it does not establish that any single granularity index is a universal measure of predictive skill or psychological health. Reviews urge precise construct definitions and attention to correlations among overlapping assessments. Hoemann et al. (2021) Lindquist (2017)
Experience-sampling methods can estimate whether a person uses different emotion words for distinguishable situations or repeatedly applies the same broad labels. But these estimates depend on the selected words, range of sampled contexts, number of observations, and statistical aggregation. Hoemann, Barrett, and Quigley reported that intensive ambulatory assessment can alter observed granularity, making the observation schedule part of the measurement model. A result obtained after prolonged repeated labeling should not automatically be generalized to a person's unobserved baseline. Hoemann et al. (2021)
Emotional expertise can be investigated through independent tasks: discrimination among situations, precision of experience reports, accurate predictions of likely reactions, adaptive choice, and interpersonal communication. These outcomes need not covary perfectly. Someone may possess a detailed emotional vocabulary but struggle to anticipate a physiological stress response, or accurately distinguish experiences while avoiding overt disclosure. Studies should ask which skills transfer across tasks and cultures and whether training or developmental changes causally alter specific outcomes rather than assuming that all competencies share one latent variable. Hoemann et al. (2021) Gendron et al. (2014)
Constructed-emotion models are well suited to hypotheses about learning because they connect prior experience to categorization. A program could measure how a novel emotional concept is acquired, how it guides prediction, and whether its effect generalizes to untrained contexts. Randomization, active controls, delayed assessments, and transparent performance criteria would establish whether observed changes reflect improved affective differentiation or simply greater willingness to use newly taught words. Such experiments remain an important route from an attractive theory of concept use to constrained causal explanation. Lindquist (2017) Barrett (2017)
Appraisal, Prediction, and the Meaning of Events
Appraisal accounts explain emotion through evaluations of circumstances relative to goals, responsibility, coping possibilities, uncertainty, and anticipated consequences. Predictive construction emphasizes how expected sensory and bodily states are organized through learned concepts. These explanatory traditions often investigate similar events, but they make different commitments about what constitutes a mechanism. An appraisal variable may be a conscious judgment, a rapidly inferred evaluation, or an analytically reconstructed relation; a predictive model must specify how such information enters a generative process. Barrett (2017) van Heijst et al. (2025)
A useful comparison considers an ambiguous interpersonal message. One reader expects criticism and experiences rising tension; another expects a request for help and responds with concern. Appraisal research can specify differences in perceived intention and goal relevance. A predictive construction account adds claims about anticipated body-state consequences and the conceptualization that gives the situation an emotional form. To separate them, investigators must manipulate expectations independently of goal relevance when possible and observe whether resulting affective changes match the distinctive predictions of each model. Wilson-Mendenhall et al. (2011) Barrett & Simmons (2015)
These models need not always disagree at the level of verbal description. Both can predict context-dependent emotions and rapid responses to changed meaning. A discriminating test should therefore target the sequence of mediation, the necessity of conceptual access, or the relation between inferred body needs and evaluations. A theory that allows any eventual response to count as a successful prediction lacks empirical constraint. Preregistered operational definitions and comparison with simpler appraisal-only or learned-association models provide stronger evidence of explanatory advantage. Friston (2010) Lindquist (2017)
Emotion categorization may also change subsequent appraisal, producing reciprocal influence over time. Labeling a state as anger can direct attention toward perceived obstruction; labeling it as anxiety can emphasize uncertainty and anticipated loss. Evidence from concept manipulation and contextual variation makes this feedback plausible, but causal identification requires designs that separate instruction effects from naturally emerging experience. The scientific task is to map observable reciprocal dependencies without assuming that one process always occurs first or that every affective change is mediated by an explicit semantic decision. Barrett et al. (2011) Wilson-Mendenhall et al. (2013)
Basic Emotion Programs and Evolutionary Explanations
Basic-emotion theories attribute important explanatory weight to evolved affective systems and recurrent response organizations. Ekman's 1992 account argued that characteristic configurations can serve adaptive life tasks and share properties such as rapid onset and automatic evaluation. Other research on emotional expression recognizes multiple expressive signals and dimensions. Such programs should be assessed against their actual claims about constraints and functional organization rather than reduced to the view that every feeling has one exclusive facial expression or isolated brain center. Ekman (1992) Keltner et al. (2019)
Constructed emotion raises a different question: how does a particular episode become an instance of a recognizable emotion within a brain's predictive regulation of its body and environment? In Barrett's account, evolved regulatory capacities contribute to the conditions that make emotion construction possible, while learned conceptual categories help organize differentiated episodes. The position thus retains evolutionary commitments concerning nervous-system regulation while rejecting a required one-to-one mapping of named emotion categories to dedicated inherited packages. A scientific comparison must specify which structural and functional invariants each theory expects. Barrett (2017) Barrett et al. (2025)
Van Heijst, Kret, and Ploeger's evolutionary analysis, published online in 2023 and in a 2025 issue, used Tinbergen's explanatory questions to propose ways of clarifying the dispute. Their account suggested that basic-emotion and constructed-emotion traditions may focus on different aspects of affective life and could be supplementary. Barrett and coauthors responded in 2025 that identifying one with objective bodily emotion and the other with subjective feeling mischaracterized the constructionist theory's biological claims. The disagreement concerns definitions and explanatory assumptions as well as results; the literature should present both positions accurately. van Heijst et al. (2025) Barrett et al. (2025)
An informative evolutionary research design would examine phylogenetic distributions of defensive, affiliative, and reward-related capacities alongside human concept-dependent categories. Comparative evidence for ancient survival functions does not, by itself, establish that the modern linguistic category fear is implemented in the same way across species. Conversely, evidence that human emotion terms vary culturally does not negate biological continuity in body regulation. Integration requires careful identification of the trait being compared and evidence appropriate to developmental, mechanistic, functional, and evolutionary questions. Keltner et al. (2019) van Heijst et al. (2025)
Prediction, Learning, and Neural Plasticity
Predictive affective processes depend on how experience changes future expectations. Repeated encounters can modify which cues imply safety, loss, social evaluation, or opportunity, and those expectations can in turn influence bodily preparation and conceptual interpretation. Predictive theories describe this using learning and model updating, but their empirical value depends on specifying which variables are learned and how change is estimated. An account that explains each new episode solely by appealing to unspecified prior experience becomes difficult to challenge experimentally. Friston (2010) Barrett (2017)
One approach separates acquisition from transfer. During training, participants learn that a previously ambiguous cue predicts a change in breathing load, social feedback, or environmental outcome. Later trials can alter the cue's reliability or situational meaning while measuring expectation ratings, physiology, action, and verbal categories. If a generative model predicts the timing or size of changes better than reinforcement learning with a simpler state representation, its added complexity is supported. Such evidence requires out-of-sample evaluation rather than retrospective compatibility with a flexible narrative. Seth & Friston (2016) Barrett & Simmons (2015)
The neural implementation of conceptual learning can involve distributed memory and control processes that also operate outside emotion tasks. Kober's and Lindquist's meta-analytic evidence cautions against assigning each learned emotion category to a single invariant anatomical circuit. Recurrent circuits can support multiple functions, while different combinations can realize similar reports in distinct contexts. This plasticity supports a research agenda for reconfiguration, though observed neuroplasticity alone cannot identify the computational variables proposed by constructed-emotion theory. Kober et al. (2008) Lindquist et al. (2012)
Longitudinal work is needed to estimate durable learning effects. Immediate differences after category instruction could reflect priming, demand, or strategic response; delayed generalization across contexts would provide stronger evidence that conceptual prediction changed. Long-term measures should also address whether changes in self-reported emotional granularity covary with physiological regulation, behavioral flexibility, or social understanding. Distinct outcome trajectories could reveal multiple learning mechanisms rather than a single global emotional competence. Hoemann et al. (2021) Hoemann et al. (2021) Lindquist (2017)
Critically Evaluating Claims of Explanatory Sufficiency
An explanatory model should connect general principles to observations through explicit intermediate assumptions. Constructed emotion proposes a rich relationship among allostasis, prediction, concepts, and situated categorization. Yet some empirical findings often cited in its favor are compatible with multiple theories. The absence of a one-region emotion center can support distributed mechanisms without proving active inference; contextual effects on labels can support inferential perception without proving that categorization constitutes the entire emotional episode. Evidence must be matched to the strength of the particular claim. Lindquist et al. (2012) Barrett et al. (2019)
Predictive coding is sometimes expressed using highly general equations of Bayesian inference and free energy. Such formulations can be useful organizing frameworks, but they do not automatically specify a unique mechanistic account of emotion. Two models may fit the same average ratings by assigning different latent variables to expectation, bodily signal, and concept selection. Model identification requires distinctive interventions, parameter recovery, predictive checks, and transparent alternative models. A failure to specify these steps risks transforming a proposed mechanism into a description applicable after any outcome. Friston (2010) Seth & Friston (2016)
Interpretation should also distinguish theory-associated evidence from theory-exclusive evidence. Wilson-Mendenhall's situated emotion experiments support sensitivity to context, and Siegel's autonomic synthesis challenges simple fingerprint assumptions. Neither alone establishes that top-down predictions causally produce every named emotional experience. Similarly, category-rich emotion reports support high-dimensional phenomenology, but their structure is not a direct test of interoceptive precision weighting. Strong synthesis preserves separate conclusions for conceptual, physiological, and neural measures. Wilson-Mendenhall et al. (2011) Siegel et al. (2018) Cowen & Keltner (2017)
A cumulative research program can adjudicate explanations by preregistering what counts as success and failure, measuring variables independently when feasible, using cross-context replication, and comparing models on held-out data. Investigators should report effect sizes and uncertainty, representative sampling limits, and the extent to which their design distinguishes competing causal pathways. This critical standard applies equally to basic-emotion, appraisal, and constructionist theories. Productive disagreement depends on precisely testable claims rather than treating broad descriptions of emotional complexity as sufficient demonstrations. van Heijst et al. (2025) Barrett et al. (2025) Hoemann et al. (2021)
Measurement, Construct Validity, and Multiple Indicators
Constructed emotion is a multilevel research object and requires a measurement design that identifies which level a variable represents. Ratings of valence, named emotion, perceived arousal, bodily activation, expressive behavior, and inferred social meaning are not interchangeable indicators. The same report can arise through different combinations of processes, while one physiological change can appear in different episodes. A study should therefore specify whether it seeks a latent category, a dynamic functional relation, or a predictive mapping from observed signals to an outcome. Russell (1980) Siegel et al. (2018)
Researchers can improve construct validity by collecting separate measures of expectation, categorical report, subjective affect, physiological dynamics, and context. Simultaneous measurement alone is insufficient: responses occur at different time scales, and any task can alter the process under study. For example, frequent prompted labeling may change differentiation, while video-based elicitation samples a limited range of life events. Measurement models should examine indicator reliability, missingness, reactivity, and the extent to which participants understand the items in comparable ways. Hoemann et al. (2021) Cowen & Keltner (2017)
An affective classifier should report how it performs when participants, contexts, and sensor distributions change. Cross-validation that divides adjacent observations from the same person across training and test folds may overstate real-world generalization. Better evaluation includes participant-held-out and context-held-out performance, calibration, uncertainty, and predeclared outcomes. A reliable predictor of a label is evidence of statistical prediction of that label; it does not establish a privileged biological essence or unique causal mechanism for the emotion. Barrett et al. (2019) Lindquist et al. (2012)
Studies that compare construction with discrete-emotion alternatives require mutually fair operational definitions. If one account is modeled as a highly specific one-to-one neural map and the other as an unconstrained flexible network, differences in fit can reflect asymmetry of model flexibility. Competing accounts should be implemented at comparable complexity and linked to the same observed data. Negative findings need sensitivity analysis and uncertainty intervals, while favorable results need independent replication and tests of alternative mediation pathways. van Heijst et al. (2025) Barrett et al. (2025) Kober et al. (2008)
Computational Modeling and Causal Identification
A computational account should define its latent variables, observations, model dynamics, and learning or inference rule. A candidate generative model of emotion might include expected bodily state, sensory evidence, contextual latent causes, concept availability, and prospective action costs. These variables must be connected to observable tasks with measurable predictions. Friston's free-energy framework provides general mathematics for agent–environment regulation, while interoceptive-inference formulations show how that vocabulary can be applied to internal sensation; neither supplies a completed validated psychological model for every human emotion. Friston (2010) Seth & Friston (2016)
The causal test is more demanding than good retrospective fit. To establish that a learned concept contributes causally, researchers could randomize access to category cues and measure changes in experience or action that cannot be explained by response bias alone. To test precision weighting, a task could systematically vary reliability of internal and external cues, then assess trial-level adjustments predicted by a formally fitted model. For allostatic prediction, investigators could manipulate expected energetic demands and distinguish anticipatory from reactive regulation. Each design faces ethical constraints and possible non-emotional explanations. Barrett (2017) Barrett & Simmons (2015)
Model comparison should include parsimonious alternatives: category-conditioned regression, generic arousal models, appraisal-based prediction, associative learning, and models in which language affects only reporting. Parameter recovery and simulation-based calibration help establish whether the proposed task can identify its own model. Fit to the same observations used for tuning can make flexible theories appear unusually persuasive; out-of-sample predictive adequacy offers a more demanding criterion. Even a winning predictive model leaves questions about neural implementation unless causal interventions connect the inferred variables to actual biological processes. Friston (2010) Lindquist (2017)
A fully integrated experiment would analyze repeated events across individuals, sensory conditions, and social contexts, with neural and peripheral measures aligned to an explicit temporal model. Researchers could predefine which interaction between bodily uncertainty and category learning is uniquely predicted by the constructed-emotion account, then compare it against evolutionary and appraisal models. Such an experiment would provide direct evidence about the distinctive synthesis rather than separately confirming unsurprising claims that the body matters, context matters, and predictions matter. Barrett et al. (2025) van Heijst et al. (2025) Kleckner et al. (2017)
Evidence Quality, Replication, and Generalizability
The evidence base includes controlled experiments, meta-analyses, neuroanatomical studies, cultural comparisons, and theoretical papers, with different inferential strengths. Reviews of autonomic specificity and functional imaging summarize broad literatures and clarify heterogeneity; they cannot remove design limitations of every original study. Situated conceptualization experiments provide closer manipulation of putative causal factors but often depend on restricted tasks. The strongest conclusion across these levels is that emotion-related signals are distributed, context-sensitive, and multiply determined; the unique formal computation underlying construction remains a hypothesis requiring specific tests. Siegel et al. (2018) Lindquist et al. (2012) Wilson-Mendenhall et al. (2011)
Replication should preserve the proposed causal contrast while varying ancillary features: eliciting stimuli, cultural context, recording device, label language, participant age, and social interaction. A finding dependent on a forced-choice response list may reveal properties of that list rather than stable emotion structure. Similarly, physiological signatures identified within a narrowly standardized task should be tested during spontaneous interaction and movement. Researchers must distinguish a successful same-protocol replication from transportability to new conditions and document both. Gendron et al. (2014) Barrett et al. (2019) Keltner et al. (2019)
Publication selection and researcher flexibility create risks even when an individual study is technically correct. Selective contrasts, variable region-of-interest definitions, repeated feature selection, unreported questionnaire transformations, and post hoc labels can inflate interpretability. Transparent preprocessing, preregistered comparisons where feasible, open analytical code, and full disclosure of unsuccessful tests strengthen inference. For a narrative monograph such as this one, the absence of a systematic-search protocol means it can synthesize important findings and disagreements but cannot claim exhaustive coverage or calculate an unbiased pooled estimate. Kober et al. (2008) Hoemann et al. (2021)
Recent theoretical controversy is relevant evidence about unresolved definitions, not a measure of which team has more supporters. The 2025 publications by van Heijst and colleagues and Barrett and colleagues make divergent assumptions explicit. They provide a useful basis for discriminating studies that assess what is meant by emotion, biological state, feeling, and construction. Empirical advance will depend on transparent tests with shared outcome definitions and independently reproducible analyses, rather than on resolving the disagreement by terminology alone. van Heijst et al. (2025) Barrett et al. (2025)
Applied Implications and Human–Artificial Emotion Inference
Constructionist findings inform the interpretation of emotional reports in education, healthcare, social interaction, and digital systems. A person's emotion label may be informative yet context-sensitive, and a perceived facial expression is an uncertain inference rather than direct inspection of another person's experience. Applications should therefore gather relevant situational information, accommodate mixed states, and distinguish prediction accuracy from psychological understanding. They should also avoid assuming that one physiological measurement can determine an individual's emotional category across circumstances. Barrett et al. (2019) Barrett et al. (2011)
In clinical and health research, altered interoceptive attention or unusual emotion differentiation can become candidate mechanisms for investigation. Existing evidence supports distinctions among bodily sensing, subjective appraisal, and linguistic awareness, yet it does not authorize diagnosis or treatment solely from a constructionist label. Clinical translation requires disorder-specific studies, validated assessments, controlled interventions, and trained professional review. The present monograph evaluates theory and measurement rather than issuing individual diagnostic recommendations, medication advice, or clinical efficacy claims. Critchley & Garfinkel (2017) Hoemann et al. (2021)
Artificial affect-recognition systems provide a particularly clear test of the inferential problem. A system that predicts emotion labels from audio, images, or text produces estimates conditioned on training categories and contextual coverage. Its apparent accuracy can depend strongly on recording conditions, demographic composition, and choice of reference labels. Evaluation should disclose uncertainty, test cross-domain generalization, and avoid presenting predicted categories as transparent measurements of private emotional states. These implications follow from human behavioral and perceptual research. Barrett et al. (2019) Gendron et al. (2014)
The Artificial Era adds a distinct scientific question: could an artificial system implement functional organization involving anticipatory regulation, learned concepts, and self-modeling without reproducing human visceral physiology? This is a proposed comparative research question, not a demonstrated property of current AI. A research program would define artificial regulatory variables, observable predictive updates, persistent history, and independently testable outcomes, then avoid inferring subjective emotion from verbal fluency or a probabilistic label alone. Comparisons require explicit system-level criteria rather than transferring human emotion vocabulary without validation. Barrett (2017) Friston (2010)
Unresolved Problems and a Discriminating Research Agenda
The first unresolved problem is how to identify the generative role of concepts without conflating experience with reporting. Label manipulation experiments can change choices and attention; to establish a stronger claim, they must also track behavior, experience, and physiology under conditions that control for expectancy and task demands. Developmental or cross-language studies could test whether concepts contribute differently as they are acquired. Explicit causal estimates would clarify whether conceptualization is necessary for particular classes of emotion or one of several available routes to affective differentiation. Lindquist (2017) Gendron et al. (2014)
The second problem concerns formal specificity. A predictive account should indicate which interoceptive error signals, uncertainty estimates, and policy-related variables distinguish affective episodes from other predictive processes. Without such details, all human perception and action might satisfy the same framework while explaining little about the particular organization of shame, panic, wonder, or grief. Computational and neurophysiological studies should pre-register discriminating trajectories, evaluate alternative models, and test whether estimated quantities correspond to measurable neural or peripheral phenomena. Friston (2010) Seth & Friston (2016) Kleckner et al. (2017)
The third problem concerns the proper levels of categorization. Affective dimensions, named emotional concepts, adaptive functional systems, and distributed neural configurations describe different structures. A claim about universal defensive capacity can coexist with cultural variation in fear concepts, yet stronger claims about exclusive physiological fingerprints can conflict with heterogeneous autonomic evidence. Future studies should formulate predicted invariants at a stated level and test transportability across contexts. Explicitly tracking the level of explanation would make the comparison between constructionist and basic-emotion programs empirically productive. Ekman (1992) Siegel et al. (2018) van Heijst et al. (2025)
The fourth problem is scientific generalization. Research has sampled particular languages, industrialized populations, imagery tasks, laboratory elicitors, and self-report instruments more extensively than the full range of emotional life. Naturalistic longitudinal sampling, culturally collaborative study design, representative developmental evidence, perturbation experiments, and transparent model comparison would reduce this imbalance. The mature form of constructed-emotion research will be measured by clear predictions and surviving failures, including observations where basic-emotion or appraisal models explain the data more efficiently. Barrett et al. (2025) Hoemann et al. (2021) Keltner et al. (2019)
Conclusion
Constructed Emotion and Predictive Affective Processes is a research program concerned with the generation of emotional episodes through bodily regulation, inference, situated conceptual knowledge, and dynamic interaction with an environment. Its central insight is that an emotion label does not function as a sufficient specification of neural, autonomic, or expressive processes. The same category can encompass heterogeneous episodes, while common physiological and cognitive resources participate in multiple named experiences. This conclusion rests on diverse evidence concerning context, autonomic variation, neural distribution, and culturally conditioned categorization. Barrett (2017) Lindquist et al. (2012) Siegel et al. (2018)
Prediction-based accounts supply a plausible framework for linking interoception, allostasis, and action with emotion construction. Contemporary evidence supports the importance of bodily expectation and distributed neural control, and experimental work demonstrates situational modulation of emotion-related experience. The exact mathematical mechanisms, necessary role of concepts, and distinctive causal relationships remain incompletely established. The existence of predictions in general nervous-system function is not equivalent to direct validation of a complete predictive theory of differentiated emotions. Barrett & Simmons (2015) Kleckner et al. (2017) Wilson-Mendenhall et al. (2011)
Competing basic-emotion, dimensional, appraisal, and constructionist approaches retain distinct explanatory claims. Meta-analyses and cross-cultural studies challenge overly rigid assumptions about universal physical signatures, while recurrent expressive and adaptive patterns justify continued research on constraints and regularities. The 2025 debate clarifies that agreement on the biological importance of emotion does not dissolve differences concerning what an emotional category explains. Progress depends on shared measurement standards, stronger causal interventions, and models that specify what observable result would count against them. van Heijst et al. (2025) Barrett et al. (2025) Keltner et al. (2019)
The resulting scientific synthesis establishes a durable research structure: concepts and situated meaning, bodily inference and allostasis, distributed neural dynamics, cultural learning, experimental measurement, and comparative model testing. These structures connect psychology of Homo to wider questions about the organization of adaptive artificial systems while preserving empirical boundaries between human emotional experience and system-level analogies. An account of constructed emotion becomes fundamental when its powerful integrative language is made precise enough to guide independently reproducible experiments. Friston (2010) Hoemann et al. (2021) Barrett (2017)
References
- Barrett, L. F. (2017). The theory of constructed emotion: An active inference account of interoception and categorization. Social Cognitive and Affective Neuroscience, 12(1), 1–23. https://doi.org/10.1093/scan/nsw154
- Barrett, L. F., Atzil, S., Bliss-Moreau, E., Chanes, L., Gendron, M., Hoemann, K., Katsumi, Y., Kleckner, I. R., Lindquist, K. A., Quigley, K. S., Satpute, A. B., Sennesh, E., Shaffer, C., Theriault, J. E., Tugade, M., & Westlin, C. (2025). The theory of constructed emotion: More than a feeling. Perspectives on Psychological Science, 20(3), 392–420. https://doi.org/10.1177/17456916251319045
- Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements. Psychological Science in the Public Interest, 20(1), 1–68. https://doi.org/10.1177/1529100619832930
- Barrett, L. F., Mesquita, B., & Gendron, M. (2011). Context in emotion perception. Current Directions in Psychological Science, 20(5), 286–290. https://doi.org/10.1177/0963721411422522
- Barrett, L. F., & Simmons, W. K. (2015). Interoceptive predictions in the brain. Nature Reviews Neuroscience, 16(7), 419–429. https://doi.org/10.1038/nrn3950
- Cowen, A. S., & Keltner, D. (2017). Self-report captures 27 distinct categories of emotion bridged by continuous gradients. Proceedings of the National Academy of Sciences, 114(38), E7900–E7909. https://doi.org/10.1073/pnas.1702247114
- Critchley, H. D., & Garfinkel, S. N. (2017). Interoception and emotion. Current Opinion in Psychology, 17, 7–14. https://doi.org/10.1016/j.copsyc.2017.04.020
- Ekman, P. (1992). An argument for basic emotions. Cognition & Emotion, 6(3–4), 169–200. https://doi.org/10.1080/02699939208411068
- Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787
- Gendron, M., Roberson, D., van der Vyver, J. M., & Barrett, L. F. (2014). Perceptions of emotion from facial expressions are not culturally universal: Evidence from a remote culture. Emotion, 14(2), 251–262. https://doi.org/10.1037/a0036052
- Gendron, M., Roberson, D., van der Vyver, J. M., & Barrett, L. F. (2014). Cultural relativity in perceiving emotion from vocalizations. Psychological Science, 25(4), 911–920. https://doi.org/10.1177/0956797613517239
- Hoemann, K., Barrett, L. F., & Quigley, K. S. (2021). Emotional granularity increases with intensive ambulatory assessment: Methodological and individual factors influence how much. Frontiers in Psychology, 12, Article 704125. https://doi.org/10.3389/fpsyg.2021.704125
- Hoemann, K., Nielson, C., Yuen, A., Gurera, J. W., Quigley, K. S., & Barrett, L. F. (2021). Expertise in emotion: A scoping review and unifying framework for individual differences in the mental representation of emotional experience. Psychological Bulletin, 147(11), 1159–1183. https://doi.org/10.1037/bul0000327
- James, W. (1884). What is an emotion? Mind, os-IX(34), 188–205. https://doi.org/10.1093/mind/os-IX.34.188
- Keltner, D., Sauter, D., Tracy, J., & Cowen, A. (2019). Emotional expression: Advances in basic emotion theory. Journal of Nonverbal Behavior, 43(2), 133–160. https://doi.org/10.1007/s10919-019-00293-3
- Kleckner, I. R., Zhang, J., Touroutoglou, A., Chanes, L., Xia, C., Simmons, W. K., Quigley, K. S., Dickerson, B. C., & Barrett, L. F. (2017). Evidence for a large-scale brain system supporting allostasis and interoception in humans. Nature Human Behaviour, 1, Article 0069. https://doi.org/10.1038/s41562-017-0069
- Kober, H., Barrett, L. F., Joseph, J., Bliss-Moreau, E., Lindquist, K., & Wager, T. D. (2008). Functional grouping and cortical–subcortical interactions in emotion: A meta-analysis of neuroimaging studies. NeuroImage, 42(2), 998–1031. https://doi.org/10.1016/j.neuroimage.2008.03.059
- Lindquist, K. A., Wager, T. D., Kober, H., Bliss-Moreau, E., & Barrett, L. F. (2012). The brain basis of emotion: A meta-analytic review. Behavioral and Brain Sciences, 35(3), 121–143. https://doi.org/10.1017/S0140525X11000446
- Lindquist, K. A. (2017). The role of language in emotion: Existing evidence and future directions. Current Opinion in Psychology, 17, 135–139. https://doi.org/10.1016/j.copsyc.2017.07.006
- Pessoa, L. (2008). On the relationship between emotion and cognition. Nature Reviews Neuroscience, 9(2), 148–158. https://doi.org/10.1038/nrn2317
- Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39(6), 1161–1178. https://doi.org/10.1037/h0077714
- Schachter, S., & Singer, J. E. (1962). Cognitive, social, and physiological determinants of emotional state. Psychological Review, 69(5), 379–399. https://doi.org/10.1037/h0046234
- Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends in Cognitive Sciences, 17(11), 565–573. https://doi.org/10.1016/j.tics.2013.09.007
- Seth, A. K., & Friston, K. J. (2016). Active interoceptive inference and the emotional brain. Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1708), Article 20160007. https://doi.org/10.1098/rstb.2016.0007
- Siegel, E. H., Sands, M. K., Van den Noortgate, W., Condon, P., Chang, Y., Dy, J., Quigley, K. S., & Barrett, L. F. (2018). Emotion fingerprints or emotion populations? A meta-analytic investigation of autonomic features of emotion categories. Psychological Bulletin, 144(4), 343–393. https://doi.org/10.1037/bul0000128
- van Heijst, K., Kret, M. E., & Ploeger, A. (2025). Basic emotions or constructed emotions: Insights from taking an evolutionary perspective. Perspectives on Psychological Science, 20(3), 377–391. https://doi.org/10.1177/17456916231205186
- Wilson-Mendenhall, C. D., Barrett, L. F., Simmons, W. K., & Barsalou, L. W. (2011). Grounding emotion in situated conceptualization. Neuropsychologia, 49(5), 1105–1127. https://doi.org/10.1016/j.neuropsychologia.2010.12.032
- Wilson-Mendenhall, C. D., Barrett, L. F., & Barsalou, L. W. (2013). Neural evidence that human emotions share core affective properties. Psychological Science, 24(6), 947–956. https://doi.org/10.1177/0956797612464242
- Wilson-Mendenhall, C. D., Barrett, L. F., & Barsalou, L. W. (2013). Situating emotional experience. Frontiers in Human Neuroscience, 7, Article 764. https://doi.org/10.3389/fnhum.2013.00764