A temporal framework modeling dialogue semantics as four interacting forces — not static labels. DSFT-TD detects semantic transitions 7 turns before they occur.
Traditional NLP classifiers assign labels to individual utterances in isolation. Human dialogue, however, is defined by temporal dynamics — gradual shifts, emerging tensions, persuasive drift. DSFT models these dynamics as an evolving field of four interacting semantic forces.
Meaning is not a point in space. It is the dynamics of interaction between opposing forces. The observer is not neutral — it actively modifies the field it measures.
Each dialogue generates a field of four operational forces. Forces are not independent — they interact via an asymmetric coupling matrix, creating emergent temporal dynamics.
Forces evolve under a discrete-time equation incorporating inertia, inter-force coupling, momentum, and hysteresis resistance. Early detection uses the precursor probability function.
| α | 0.20 | Inertia — memory of past state |
| β | 0.25 | Coupling strength |
| γ | 0.50 | Momentum coefficient |
| λ | 0.10 | Hysteresis resistance |
| Mᵢ(t) | ΔF | Fᵢ(t) − Fᵢ(t−1) |
| Rᵢ(t) | persist×0.15 | Accumulated dominance penalty |
Controlled synthetic benchmark results. All five canonical transition types detected with identical 7-turn early warning latency.
| Transition Type | Precursor | Dominance | Latency | Status |
|---|---|---|---|---|
| Analytical → Affective | Turn 0 | Turn 7 | 7 turns | ✓ Early |
| Analytical → Persuasive | Turn 0 | Turn 7 | 7 turns | ✓ Early |
| Affective → Persuasive | Turn 0 | Turn 7 | 7 turns | ✓ Early |
| Persuasive → Exploratory | Turn 0 | Turn 7 | 7 turns | ✓ Early |
| Exploratory → Analytical | Turn 0 | Turn 7 | 7 turns | ✓ Early |
| Average | — | — | 7.0 turns | All ✓ |
Four observer modes alter how the semantic field is sampled. This is a configurable architectural choice, not a claim about quantum measurement.
DSFT was not conceived in one step. Each predecessor system failed in a specific, instructive way — and each failure generated the next design insight.