Neuroscience and Cognitive Science of Emotional Memory¶

This document collects the neuroscientific and cognitive foundations that inform the library's architecture. Each section ends with the operational implications for the system.

1. The Fundamental Neural Circuit: Amygdala–Hippocampus¶

LeDoux's Dual Pathway¶

Joseph LeDoux established the neural architecture of fear memory through Pavlovian conditioning.

Two parallel pathways: - Fast path (low road): Sensory thalamus → Amygdala directly. ~150–200 ms. Coarse processing, sufficient for immediate defensive responses before conscious awareness. - Slow path (high road): Sensory thalamus → Sensory cortex → Amygdala. Slower but enables detailed, nuanced evaluation. Reduces false alarms.

The lateral nucleus of the amygdala receives convergent input from both pathways and is critical for associating stimuli with threats.

Implication for encoding: The system must support dual-speed encoding: - Fast path: immediate emotional tag (coarse valence + arousal) based on surface patterns - Slow path: elaborated emotional tag with full post-processing appraisal

Hippocampal Modulation by the Amygdala (McGaugh)¶

James McGaugh demonstrated experimentally that the basolateral amygdala (BLA) mediates the modulation of memory consolidation by stress hormones.

Mechanism: After emotionally significant learning, extracellular norepinephrine in the amygdala rises to ~300% above baseline and correlates with retention performance. The BLA then modulates consolidation in downstream regions (caudate, nucleus accumbens, cortex).

The emotional tagging hypothesis (Richter-Levin & Akirav): The amygdala, activated by emotional arousal, "tags" a hippocampal memory trace by reinforcing synaptic consolidation at activated neurons.

Arousal vs. valence: Emotionally arousing memories depend on the amygdala-hippocampus network; emotionally valenced but non-arousing memories depend on the prefrontal cortex-hippocampus network (Kensinger & Corkin, 2004).

Implication: The consolidation_strength of each memory is primarily a function of arousal at encoding time, not valence. High arousal → preferential consolidation → slower decay.

2. The Somatic Marker Theory (Damasio)¶

Antonio Damasio proposed that emotions, experienced as bodily states ("somatic markers"), play a critical role in decisional reasoning under conditions of complexity and uncertainty.

Core principles: - Somatic markers are changes in bodily state (accelerated heartbeat, visceral sensation) associated with the past emotional outcomes of decisions - They are represented and regulated primarily in the ventromedial prefrontal cortex (vmPFC) and the amygdala - During decision-making, somatic markers guide choices by reactivating the emotional "tag" of past outcomes, consciously or unconsciously

Iowa Gambling Task: Patients with vmPFC damage do not develop somatic markers and make systematically disadvantageous choices despite intact logical ability.

Implication: Somatic markers are the neuroscientific precedent for our EmotionalTag. A retrieval system that uses the emotional tags of past memories as a decision-making bias computationally implements Damasio's hypothesis.

3. Panksepp's Seven Affective Systems¶

Jaak Panksepp mapped seven primary emotional command systems through electrical brain stimulation, pharmacological challenges, and lesion studies in mammalian brains.

System	Valence	Function
SEEKING (Expectancy)	Positive	Exploration, curiosity, anticipatory enthusiasm
RAGE (Anger)	Negative	Frustration, aggression when goals are blocked
FEAR (Fear)	Negative	Threat detection, freeze/flight responses
LUST (Sexual desire)	Positive	Reproductive drive
CARE (Nurturance)	Positive	Attachment, maternal behavior
PANIC/GRIEF (Panic/Grief)	Negative	Separation distress, loss of social bond
PLAY (Social play)	Positive	Social bonding, rough-and-tumble play

Each system is subcortically generated, evolutionarily conserved across species, and evaluative (approach for positive, avoidance for negative).

Implication: These 7 systems offer a taxonomy of biologically grounded emotional primitives that can serve as the basis for discrete categorization, distinct from continuous coordinates (valence, arousal). The SEEKING system corresponds functionally to Spinozian conatus — the fundamental exploratory drive.

4. Memory Consolidation and Emotional Modulation¶

Flashbulb Memory (Brown & Kulik, 1977)¶

Flashbulb memories are vivid, detailed, highly confident recollections of the circumstances under which one learned a significant, emotionally intense piece of news (e.g. September 11th, a personal traumatic event).

Mechanism: Amygdala activation triggers the release of stress hormones that potentiate hippocampal consolidation — producing unusually durable memory traces.

Critical limitation: Despite their subjective vividness, flashbulb memories are not necessarily more accurate than ordinary memories. They are more consistent and confidently held, but subject to distortion over time (Talarico & Rubin, 2003).

Implication for the library: High arousal → high consolidation_strength → slow decay → high retrieval priority. But not necessarily high accuracy. The system must not treat high-arousal memories as ground truth.

Stress Hormones and the Yerkes-Dodson Curve¶

Inverted-U relationship between stress hormone levels and mnemonic performance: - Moderate levels of cortisol and norepinephrine enhance encoding and consolidation - Extreme or chronic stress (elevated cortisol) impairs hippocampal function and retrieval

Timing matters: - Hormones released during or immediately after encoding → enhance consolidation - Stress before retrieval → impairs it

Implication: The system must model the non-linear relationship between arousal and consolidation_strength. A very high arousal value does not necessarily produce maximum consolidation strength.

Reconsolidation (Nader, Schiller)¶

Karim Nader, Glenn Schafe, and Daniela Schiller demonstrated that reactivated memories temporarily become labile and must be "reconsolidated" to persist.

Fundamental finding: When a consolidated memory is reactivated (retrieved), it enters a labile state for a limited time window. During this window: - The memory can be weakened (amnesic agents such as propranolol or anisomycin) - The memory can be updated with new information - The emotional tag can be modified

Prediction error as trigger: Prediction error (mismatch between expectation and experience) appears to be necessary to trigger reconsolidation.

Clinical application: Fear memories can be modified during reconsolidation through behavioral interventions (extinction during the reconsolidation window) or pharmacological agents.

Critical implication for the library: Retrieval is not a read-only operation — the Affective Prediction Error (APE) determines whether the emotional tag becomes labile. Only when the APE exceeds a configurable threshold does the reconsolidation window open (window_opened_at), during which the memory's EmotionalTag can be updated. This APE-gated model is also inspired by Pearce & Hall (1980): associability increases after high prediction errors, and decreases after accurate predictions.

5. Cognitive Appraisal Theories¶

Lazarus — Primary and Secondary Appraisal (1966, 1991)¶

Richard Lazarus proposed that emotions arise not from events themselves but from the cognitive evaluation of those events.

Two-phase appraisal: 1. Primary appraisal: Is this event relevant to me? Is it threatening, beneficial, or irrelevant? 2. Secondary appraisal: Can I cope? What resources do I have?

The combination determines the specific emotion generated (e.g. threat + low coping = fear/anxiety; threat + high coping = challenge).

Scherer — Component Process Model (CPM)¶

Klaus Scherer expanded appraisal theory into a sequential multi-component process model.

Five Stimulus Evaluation Checks (SECs): 1. Novelty: Is something new or unexpected happening? 2. Intrinsic pleasantness: Is it intrinsically pleasant or unpleasant? 3. Goal significance: Does it further or obstruct my goals? 4. Coping potential: Can I handle it? 5. Normative significance: Is it consistent with my norms and self-concept?

Emotion is defined as the synchronization of five components: cognitive appraisal, physiological arousal, motor expression, action tendency, and subjective feeling.

Implication for Layer 4: Our Appraisal Vector implements a computational version of Scherer's SECs. The five appraisal dimensions in our model (novelty, goal_relevance, coping_potential, norm_congruence, self_relevance) derive directly from this framework.

6. The Theory of Constructed Emotion (Barrett)¶

Lisa Feldman Barrett challenges the classical view of basic emotions with a constructionist account.

Core claims: - Emotions are not hard-wired, universal categories triggered by dedicated brain circuits. Instead, they are constructed in the moment by the brain's predictive machinery. - The brain continuously generates predictions about incoming sensory signals based on interoception (internal bodily sensing), prior experience, and culturally learned emotional concepts. - Core affect — a continuous stream of valence (pleasant/unpleasant) and arousal (activated/deactivated) — is the fundamental building block. Discrete emotions (anger, fear, etc.) emerge when the brain categorizes core affect using conceptual knowledge.

Implication: There is no single brain signature for any emotion; emotions emerge from domain-general brain networks. For the library: discrete emotion labels (joy, fear, anger) are emergent patterns in the continuous affective space, not hard-wired primitives. This justifies a hybrid approach: continuous coordinates + contextual categorization.

7. Mood-Congruent Memory and State-Dependent Learning¶

Mood-Congruent Memory (Bower, 1981)¶

People preferentially encode and retrieve information that matches their current mood state.

Theoretical basis: Gordon Bower's Associative Network Theory — affect exists as a central node in an associative memory network. When a mood is active, it spreads activation to mood-congruent nodes, facilitating encoding and retrieval of matching material.

Empirical evidence: Happy individuals better recall positive material; sad individuals better recall negative material. MCM effects are more robust for positive than negative moods.

Implication: The emotional_congruence signal in multi-signal retrieval must measure the match between the system's current Mood and the memory's mood_snapshot at encoding time.

State-Dependent Learning (Eich, 1989)¶

Retrieval is enhanced when the internal state (mood, physiological arousal, pharmacological state) at retrieval matches the state during encoding, independently of the emotional valence of the material.

Distinct from mood-congruent memory: state-dependent memory concerns the match between encoding and retrieval contexts, not the match between mood and material valence.

Implication: The Mood at encoding time must be captured as part of the tag (mood_snapshot). Retrieval with a Mood very different from the mood_snapshot may penalize the score of an otherwise relevant memory.

8. Prediction Error and Affective Learning¶

Dopaminergic Reward Prediction Error (Schultz, 1997)¶

The dopaminergic reward prediction error (RPE) signal, characterized by Wolfram Schultz, is a central mechanism linking emotion to learning: - Dopaminergic neurons encode the difference between expected and received reward - Positive RPE (better than expected) → dopamine increase → reinforces behavior - Negative RPE (worse than expected) → dopamine suppression → guides adjustment

Separable Neural Signals for RPE and Affective Prediction Error (2025)¶

Recent research (Nature Communications, 2025) has identified separable neural signals for RPE and affective prediction errors — deviations from emotional expectations that drive learning independently of reward value, through striatum-amygdala interactions.

Implication for reconsolidation: When a memory is retrieved and the current event generates a significant affective prediction error (reality differs from the emotional expectation encoded in the memory), the reconsolidation window opens. This is the computational mechanism that triggers the update of the EmotionalTag.

9. Recent Developments (2020–2026)¶

Generalizable Neural Signatures of Emotional Memory (2025)¶

A preprint (bioRxiv, 2025) identified cross-individual patterns of brain activity that predict emotional memory formation — potentially opening the path toward biomarkers for successful emotional encoding.

Affective Computing: State of the Art 2026¶

The systematic review by Fang et al. (2025) demonstrates that multimodal fusion (facial expression + speech prosody + text + physiological signals) outperforms unimodal models. Cross-modal transformers dominate.

The MemEmo gap (2026): The MemEmo benchmark demonstrated that no current LLM memory system correctly handles emotion across extraction, updating, and question-answering. This is the gap our library aims to address.

Synthesis: From Neuron to Architecture¶

Neuroscientific phenomenon	Parameter in the library
Amygdala activation → hippocampal tagging	`consolidation_strength = f(arousal_at_encoding)`
LeDoux dual pathway (fast/slow)	Fast encoding (coarse) + Slow encoding (full appraisal)
Yerkes-Dodson curve	Non-linear `consolidation_strength` with respect to arousal
Reconsolidation at retrieval	`EmotionalTag` update window post-retrieval
Somatic Marker Hypothesis	`EmotionalTag` as decisional navigation bias
Mood-Congruent Memory	`emotional_congruence` weight in retrieval scoring
State-Dependent Learning	`mood_snapshot` in `EmotionalTag` + match at retrieval
Affective Prediction Error	Trigger for opening the reconsolidation window
Panksepp 7 systems	Possible basis for discrete emotion categories
Barrett constructionism	Discrete labels as emergent patterns in continuous space
Scherer's SECs	Layer 4: Appraisal Vector (5 dimensions)

Bibliographic Notes¶

LeDoux, J.E. (1996). The Emotional Brain. Simon & Schuster.
LeDoux, J.E. (2000). "Emotion circuits in the brain." Annual Review of Neuroscience, 23, 155-184.
McGaugh, J.L. (2004). "The amygdala modulates the consolidation of memories of emotionally arousing experiences." Annual Review of Neuroscience.
McGaugh, J.L. (2013). "Making lasting memories: Remembering the significant." PNAS.
Kensinger, E.A. & Corkin, S. (2004). "Two routes to emotional memory: Distinct neural processes for valence and arousal." PNAS.
Richter-Levin, G. (2004). "The Amygdala, the Hippocampus, and Emotional Modulation of Memory." The Neuroscientist.
Damasio, A.R. (1994). Descartes' Error: Emotion, Reason, and the Human Brain. Putnam.
Damasio, A.R. (1996). "The somatic marker hypothesis and the possible functions of the prefrontal cortex." Phil. Trans. Royal Society B.
Panksepp, J. (1998). Affective Neuroscience: The Foundations of Human and Animal Emotions. Oxford University Press.
Brown, R. & Kulik, J. (1977). "Flashbulb memories." Cognition, 5(1), 73-99.
Talarico, J.M. & Rubin, D.C. (2003). "Confidence, not consistency, characterizes flashbulb memories." Psychological Science, 14(5), 455-461.
Nader, K., Schafe, G.E. & LeDoux, J.E. (2000). "Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval." Nature, 406, 722-726.
Schiller, D. et al. (2010). "Preventing the return of fear in humans using reconsolidation update mechanisms." Nature, 463, 49-53.
Lee, J.L.C., Nader, K. & Schiller, D. (2017). "An update on memory reconsolidation updating." Trends in Cognitive Sciences, 21(7), 531-545.
Lazarus, R.S. (1991). Emotion and Adaptation. Oxford University Press.
Scherer, K.R. (2009). "The dynamic architecture of emotion: Evidence for the component process model." Cognition and Emotion.
Barrett, L.F. (2017). How Emotions Are Made: The Secret Life of the Brain. Houghton Mifflin Harcourt.
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.
Bower, G.H. (1981). "Mood and memory." American Psychologist, 36(2), 129-148.
Eich, E. & Metcalfe, J. (1989). "Mood dependent memory for internal versus external events." Journal of Experimental Psychology: Learning, Memory, and Cognition.
Schultz, W., Dayan, P. & Montague, P.R. (1997). "A neural substrate of prediction and reward." Science, 275, 1593-1599.
Nature Communications (2025). "Separable neural signals for reward and emotion prediction errors."