Human–Machine Convergence in the TIE–Dialog Pilot Study

This work presents the first empirical validation of conversational coherence under the Theory of Informational Emergence (TIE) through the computational framework TIE–Dialog.
Six English dialogues (25–27 turns each) were annotated by five human raters marking perceived ruptures and repairs of meaning. These annotations were compared to the automated coherence trajectories (𝒞ₜ) generated by TIE–Dialog.

The results reveal strong human–machine convergence in the detection and propagation of coherence dynamics:

• Macro-scale alignment: warped correlation r₍warped₎ ≈ 0.93 (Dynamic Time Warping)

• Event-level agreement: F1 ≈ 0.70–0.88, κ ≈ 0.03–0.14

• Temporal asymmetry: model anticipates micro-ruptures (proto-coherence) by ≈0.3 turns on average

These findings indicate that coherence behaves as an emergent informational property, bridging human perception and computational structure.
This pilot thus provides the first quantitative baseline for studying shared understanding and perspectival synchronization between human and artificial systems.

Included Files

1. Human–Machine Convergence in the TIE–Dialog Pilot Study.pdf
   → Main paper containing the study, analyses, figures, and discussion.

2. Appendices_pilot.pdf
   → Full supplementary package, including computational parameters, visualizations, complete data tables, and theoretical integration within the TIE framework.

Key Contributions

• Empirical quantification of conversational coherence under the TIE model.

• Evidence for proto-coherence: anticipatory informational organization before explicit awareness.

• Demonstration of structural coupling: human–machine synchronization across informational fields.

• Establishment of the Quantum of Coherence (𝒬ₐ) as a triadic rhythm (Stability–Breakdown–Repair).

• Open reproducibility pipeline with documented scripts, metrics, and DTW parameters.

Data and Code Availability

All computational scripts, CSV datasets, and figure-generation notebooks are archived at
Zenodo DOI: https://doi.org/10.5281/zenodo.17516211

Environment: Python 3.11, NumPy 1.26, SciPy 1.13, scikit-learn 1.5, matplotlib 3.9, and sentence-transformers 2.3 (model: all-MiniLM-L6-v2).
Analyses executed in Google Colab (2025-04 build, seed = 42) to ensure full replicability.

Corresponding author / Contact

Adolfo J. Céspedes Jiménez — University of Groningen — Faculty of Arts  
Email: a.j.cespedes.jimenez@student.rug.nl

License: Creative Commons Attribution–NonCommercial 4.0 International (CC BY-NC 4.0)  
© 2025 Adolfo J. Céspedes Jiménez