Extended Moral Foundations Theory

• Extended Moral Foundations Theory is a comprehensive framework that refines Haidt’s model by formalizing moral intuitions and incorporating evolutionary cooperation dynamics.
• It employs a lexicon-based methodology coupled with statistical and topic modeling techniques to quantify moral content across dimensions like virtue and vice.
• The framework integrates Bayesian trust models to elucidate how moral priors influence authority selection, polarization, and epistemic processes in complex environments.

Extended Moral Foundations Theory (EMFT) is a theoretical and operational framework for analyzing the latent structure and functional role of moral intuitions in human cognition, communication, and trust. Building on Jonathan Haidt’s original Moral Foundations Theory (MFT), EMFT systematizes, expands, and formalizes the “moral taste buds” that underpin pre-rational ethical appraisals, incorporating additional foundations, refined distinctions, and explicit mappings to evolutionary cooperation domains. Recent research has demonstrated its utility as both an empirical coding framework for natural language and as a central pillar in comprehensive models of epistemic processes, particularly for explaining polarization, authority selection, and trust formation in complex information environments (Mutlu et al., 2020, Schwabe, 2 Dec 2025, Schwabe, 20 Dec 2025).

1. Theoretical Genesis and Extensions Beyond Classical MFT

Haidt’s MFT posits five foundational domains—Care/Harm, Fairness/Cheating, Loyalty/Betrayal, Authority/Subversion, and Purity/Sanctity—subsequently expanded to include Liberty/Oppression. EMFT explicitly augments and differentiates this space in two key respects:

• Addition of Foundations and Facets:
  • Liberty/Oppression as a distinct foundation capturing resistance to coercion and the valuation of individual autonomy.
  • Subdivision of Fairness into two facets:
  • Fairness-as-Equity (concern for equal outcomes) and
  • Fairness-as-Proportionality (concern that rewards track individual contribution or merit).
• Integration with Evolutionary and Cognitive Models:
  • Each foundation is linked to a specific class of cooperative problems (Morality-as-Cooperation, MAC), providing both psychological immediacy and an evolutionary rationale (Schwabe, 20 Dec 2025).

2. Lexical Operationalization and Quantification Schemes

The Extended Moral Foundations Dictionary (EMFD), developed by Araque et al., provides a lexicon-based implementation of EMFT for natural language processing. Each term in the EMFD is tagged with a foundation $d \in \{$Care/Harm, Fairness/Reciprocity, In-group/Loyalty, Authority/Respect, Purity/Sanctity$\}$ and a valence $v \in \{$virtue, vice$\}$.

• Lexicon Structure:

| Foundation | Virtue Tokens | Vice Tokens | |--------------------|--------------|-------------| | Care/Harm | 95 | 85 | | Fairness/Reciprocity | 69 | 57 | | In-group/Loyalty | 99 | 72 | | Authority/Respect | 160 | 101 | | Purity/Sanctity | 97 | 161 |

• Pipeline Overview (Mutlu et al., 2020):
  1. Tweets are aggregated by cascade into pseudo-documents.
  2. Preprocessing removes non-semantic artifacts and tokenizes the text.
  3. For document $D$, the count $n_{d,v}(D)$ of foundation-valence tokens is scored.
  4. Normalized loading $$M_{d,v}(D) = n_{d,v}(D) / \sum_{d',v'} n_{d',v'}(D)$$ defines the share per foundation-valence.
  5. Topic modeling (LDA) partitions narratives so that moral loading can be analyzed by context.
  6. Statistical tests (ANOVA, z-test) and time-series analyses (CRQA) characterize stability, valence dominance, and cross-foundation co-variation.

3. Formal Role in Bayesian Trust and Epistemic Models

EMFT underpins latent, agent-specific priors in Bayesian-style models of reasoning and trust (Schwabe, 2 Dec 2025). For $F = \{$Care, Equity, Prop, Liberty, Loyalty, Authority, Purity$\}$, let $w = (w_1,\ldots,w_7)$ be an agent’s normalized weights and $f(c)$ the moral-content vector for claim $c$:

$$P_0(c)\;=\;\frac{\exp(\sum_{i=1}^7 w_i f_i(c))}{\sum_{c'} \exp(\sum_{i=1}^7 w_i f_i(c'))}$$

After new evidence $E$ is acquired through an agent’s “trust lattice” (a recursive graph of claims and support), a Bayesian update is performed:

$P(c|E) \propto P(E|c)P_0(c)$

Here, $P_0(c)$ reflects the EMFT-weighted “semantic prior,” typically pre-setting the cognitive plausibility of $c$ before any evidentiary elaboration.

4. Integration in the MEVIR and MEVIR 2 Frameworks

The MEVIR and MEVIR 2 frameworks synthesize EMFT as one of three pillars (moral, virtue, procedural):

• Moral Pillar (EMFT + MAC): Determines which authorities, “truth makers,” and types of evidence are perceived as legitimate. Ideological/political differences are attributed to divergent EMFT weight vectors and thus distinct moral priors, even when exposed to identical propositional information (Schwabe, 20 Dec 2025).
• Virtue Pillar: Mediates whether agents defer to authority responsibly or succumb to overconfidence, groupthink, or confirmation bias.
• Procedural Pillar: Constructs trust lattices—networked structures of supported claims, where foundational beliefs anchored by EMFT influence admissible evidence and ultimate trust anchors.

MEVIR 2 introduces the concept of “Truth Tribes”—stable epistemic clusters whose members are synchronized across moral, virtue, and procedural pillars, producing internally coherent but mutually unintelligible epistemic worlds.

5. Empirical Applications: Textual Moral Content and Polarization

Applied EMFT analyses, such as quantifying moral rhetoric in large-scale tweet corpora (Mutlu et al., 2020), reveal the following:

• Virtue over Vice: Virtue-framed rhetoric predominates in shared content, contrary to some sentiment-analysis assumptions.
• Dimension Stability: The relative prevalence of moral dimensions is stable, though the absolute strength pulses in response to events.
• Dimension Profiles: Care/Harm exhibits low tweet count but highest rhetorical strength and virtue–vice polarity gap; Purity/Sanctity is consistently weakest.
• Cross-Foundation Dynamics: Fairness and In-group foundations covary most tightly, as shown via CRQA entropy metrics.
• Mechanistic Implications: EMFT-determined priors block or admit entire chains of evidential reasoning, as shown in case studies of vaccination and climate politics (Schwabe, 2 Dec 2025, Schwabe, 20 Dec 2025).

6. Limitations and Proposed Extensions

Empirical implementations of EMFT currently face several methodological constraints:

• Lexicon-based approaches are limited to surface matching, lack contextual or negation sensitivity, and are bounded to English corpora.
• No validated integration with contextual embeddings (e.g., BERT); absence of human-coded validation.
• Focus remains on single-topic corpora; multi-domain generalization is not routinely tested.
• Failure to detect implicit or negated moral content restricts comprehensiveness.

Proposed improvements include fusing EMFD with embedding models, expanding to heterogeneous corpora, incorporating tf-idf weighting, and modeling causality in polarization time series (Mutlu et al., 2020).

7. EMFT’s Explanatory Power and Prospects

Extended Moral Foundations Theory enables formal modeling of how moral priors filter reality, frame trust, and drive polarization by pre-selecting:

• Compelling “truth bearers” and “truth makers” (ontological unpacking),
• Authority figures and evidentiary standards,
• Which procedural search paths are even traversed during deliberation.

In its current evolutionary-psychological instantiation, EMFT provides both descriptive and predictive leverage for understanding political, cultural, and epistemic schisms. Ongoing refinements, such as moral reframing and metacognitive training, are proposed within MEVIR 2 to bridge divides by making underlying moral configurations visible and actionable (Schwabe, 20 Dec 2025).

---

Key References:

(Mutlu et al., 2020) Mutlu et al., "Quantifying Latent Moral Foundations in Twitter Narratives: The Case of the Syrian White Helmets Misinformation" (Schwabe, 2 Dec 2025) "The MEVIR Framework: A Virtue-Informed Moral-Epistemic Model of Human Trust Decisions" (Schwabe, 20 Dec 2025) "The MEVIR 2 Framework: A Virtue-Informed Moral-Epistemic Model of Human Trust Decisions"