Phase Transitions in Affective Meaning Divergence: The Hidden Drift Before the Break
Napassorn Litchiowong · 2026
Read Phase Transitions in Affective Meaning Divergence: The Hidden Drift Before the Break by Napassorn Litchiowong (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-09043.txt.
Introduction
Napassorn Litchiowong, writing alone, studies how conversations between partners can break down even when both use the same words. One person says a word meaning closure; the other hears surrender. The paper gives this gap a formal measure based on how far apart the two sides' emotional readings of shared anchor words sit.
From speech-act theory and game theory, the author derives a model in which repair efforts can collapse suddenly once the gap passes a threshold, and the collapse is hard to reverse. In a dataset of online talks that went off the rails, warning signs of slowing recovery appeared before the break. A second dataset gave mixed but broadly matching support for the pattern.
Catalog record
| CREATOR | Napassorn Litchiowong |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.09043
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Phase Transitions in Affective Meaning Divergence: The Hidden Drift Before the Break Napassorn Litchiowong Abstract One partner says “Fine” meaning resolution ; the other hears surrender . The word is shared; the affective uptake is not. We formalize this as affective meaning divergence (AMD), the total-variation distance between interlocutors’ anchor-conditioned affect distributions. Building on speech-act theory, common-ground accumulation, and entropy-regularized game theory, we derive a logit best-response map whose dynamics undergo a saddle-node bifurcation : when β α > 4 \beta\alpha>4 , a monotone increase in AMD-driven load produces an abrupt, hysteretic collapse of repair coordination. On Conversations Gone Awry (CGA-Wiki; N = 652 N{=}652 ), derailing conversations exhibit critical-slowing-down (CSD) signatures across multiple levels: lexical divergence variance ( p < 0.001 p{<}0.001 , d = 0.36 d{=}0.36 ), AMD variance ( p = 0.001 p{=}0.001 , d = 0.26 d{=}0.26 ), and dialog-act repair variance ( p = 0.016 p{=}0.016 , d = 0.20 d{=}0.20 ), all significant after correction and stronger than toxicity and sentiment baselines. AMD provides a distinct temporal signature, with retrospectively measured variance peaking at the bifurcation point while toxicity variance peaks earlier, and is the only indicator grounded in the theoretical framework. Boundary-condition analysis on CGA-CMV ( N = 1,169 N{=}1{,}169 ) yields mixed but directionally consistent evidence. 1 Introduction Consider a couple at dinner. One partner says “Fine.” and the other hears surrender ; the speaker meant resolution . The word is shared; the affective uptake is not. In the terminology of speech-act theory ( 2 ; 30 ) , the locutionary act is identical, but the illocutionary force received diverges from the force intended. If this gap is detected, repair can close it ( 28 ) . If it goes undetected, it accumulates in the partners’ divergent common-ground records ( 4 ) . Eventually, a conversation that would once have repaired itself stops repairing, and the relationship tips. We call this phenomenon affective meaning divergence (AMD). The present paper asks: How can gradual, continuous drift in affective meaning produce abrupt relational rupture? Our answer is a formal phase-transition model grounded in three pillars: 1. Linguistic: AMD is defined over anchors , high-frequency lexical items whose pragmatic force depends on affective uptake rather than truth conditions ( 6 ; 21 ) . We decompose apparent meaning drift into a context divergence component and a genuine conditional affect divergence component, following the decomposition logic of appraisal theory ( 21 ) . 2. Agentic: Each interlocutor treats anchors as conditioning observations in a partially observable coordination game. Under entropy-regularized best response ( 22 ; 37 ; 10 ) , the repair probability follows a logit map. 3. Dynamical: The resulting one-dimensional map q_{t+1}=\sigma(\beta(\alpha q_{t}-\kappa)) undergoes a saddle-node bifurcation. Below a critical load, the dyad rests in a high-repair attractor; above it, repair collapses to a low-repair attractor. The transition is hysteretic: recovery requires reducing load below a strictly lower threshold than the one that triggered collapse. Contributions. (C1) We formalize AMD as a context-conditioned divergence measure with an explicit decomposition bound separating genuine affective drift from context confounds (§ 3 ). (C2) We ground AMD in an assurance game and prove that belief divergence bounds value disagreement (§ 4 ). (C3) We prove a saddle-node bifurcation theorem with closed-form thresholds and extend to a two-agent system with an interpretable loop-gain condition (§ 5 ). (C4) We estimate P_{i}(s\mid x,c) via a contextual emotion classifier rather than lexicon averages, bringing the practical estimator closer to the formal construct (§ 6 ). (C5) We provide preliminary empirical evidence on synthetic data, CGA-Wiki (primary), and CGA-CMV (boundary-condition analysis) that theory-derived CSD indicators are detectable across multiple levels of linguistic representation (lexical, affective, dialog-act), with AMD providing a temporally distinct signature among the tested indicators (§ 7 ). (C6) We study three repair proxies as complementary operationalizations and report their degree of agreement; boundary-condition analysis on CGA-CMV yields mixed but directionally consistent evidence under shorter threads and noisier labels (§ 7 ). Figure 1 provides a visual overview of the full framework and pipeline. Scope of claims. We do not claim that the current estimator perfectly observes latent affective meaning, nor that AMD alone is sufficient for reliable individual-level prediction. The empirical claim is narrower: a feasible AMD proxy exhibits critical-slowing-down behavior on CGA-Wiki, with mixed but directionally consistent evidence on CGA-CMV, consistent with the proposed dynamical account. The theoretical framework identifies a class of bifurcation models that could explain abrupt relational collapse; the experiments do not uniquely identify the saddle-node formulation over alternatives. Figure 1: Overview of the AMD framework: divergent anchor-conditioned affect distributions define AMD via TV distance (A-B) ; rising rolling-window variance before breakdown constitutes the CSD early-warning signal (C) ; the saddle-node bifurcation predicts abrupt, hysteretic collapse of repair coordination (D) . 2 Theoretical Roadmap and Motivation Roadmap. The framework rests on five theoretical components, each with a distinct and non-redundant role. Speech-act theory motivates why the same lexical form can carry different received illocutionary force. Common-ground and repair theory explain how unnoticed affective divergence accumulates silently and how repair is the control variable that can sustain or collapse coordination. Appraisal theory supplies the affective state space over which speaker meanings are formally compared. Entropy-regularized game theory converts diverging affective beliefs into repair-choice dynamics via a logit best-response map. Critical-transition theory then supplies the empirical prediction: rising variance and autocorrelation before the collapse point. Background on alignment and accommodation models is included only to situate the drift assumption, not as an additional formal commitment. Figure 1 provides a visual overview of how these components compose into the full pipeline. Affective meaning as an interactional object. In relational talk, many high-frequency items function as stance markers, commitment devices, or repair moves whose success depends on affective uptake ( 6 ; 33 ) . Appraisal theory ( 21 ) systematizes this: the ATTITUDE system (Affect, Judgment, Appreciation) classifies evaluative meaning, while the ENGAGEMENT system models how speakers open or close dialogic space. When one speaker contracts dialogic space (“Obviously, we need to…”) while the other expands it (“Well, one possibility…”), their evaluative stances become incommensurable, a state we formalize as high AMD. Common ground, silent failure, and repair. Clark’s ( 4 ) joint-action framework treats communication as cumulative grounding: each act builds on prior shared understanding. If grounding fails silently, the partners develop divergent common-ground records, and each subsequent utterance is interpreted against increasingly different backgrounds. Schegloff ( 29 ) characterizes repair after next turn as “the last structurally provided defense of intersubjectivity.” Crucially, Stivers ( 33 ) distinguishes structural alignment from affective affiliation: one can align without affiliating, providing the mechanism by which surface cooperation masks deepening affective divergence. We model repair as the control variable that can be sustained or can collapse. From alignment to divergence. Although automatic priming across linguistic levels predicts convergence ( 25 ) , speakers systematically diverge in open conversation ( 11 ) and perceived over-accommodation can trigger reactive divergence ( 9 ) . These findings motivate a model where the default trajectory is drift, and repair is the active force that maintains alignment. Why entropy-regularized game theory? Entropy-regularized decision rules yield softmax policies identical to the quantal response equilibrium