From Semiotic Fitting to Semiotic Collapse: A Multi-Level Process Model of Extinction

Extinction is usually analysed as a terminal event at the population level: populations decline, fall below their viability thresholds, and disappear. We propose an alternative framework that reconceives extinction as the gradual collapse of semiotic systems — the recognition-based bonds through which organisms interpret their environment. Drawing on Kull's theory of semiotic fitting (2020, 2022), we introduce Semiotic Overlap (SO) as a measurable index of the structural alignment between an organism's interpretive system and its ecological environment. SO breaks down into three components: SO_I (informational alignment between cue and outcome), SO_B (within-population behavioural dispersion), and SO_C (structural coherence of the cue network). From these we derive five formal propositions, each operationalised so that it can be tested against open-access ecological datasets. The framework was applied to nine independent systems spanning six biological kingdoms and three continents: a prokaryote (the Baltic Sea bacterial community), a plant ( Boechera stricta ), an insect ( Operophtera brumata ), a fish ( Oncorhynchus nerka ), an amphibian ( Triturus cristatus , across five populations), a reptile (marine turtles), a bird ( Parus major ), and two mammals ( Lepus americanus and Lutra lutra ). Composite SO ranged from 0.35 to 0.68 across the systems, ordering them by phase of collapse rather than by taxonomy. In O. brumata , SO acted as a leading indicator of population state (Granger F = 4.54, p = 0.049, ΔR² = +0.108), but it did not in P. major , which is buffered demographically. Boechera stricta , where phenotypic plasticity accounts for roughly 80% of the phenological change against about 20% for evolution, provided direct support for Kull's claim that fitting comes before selection. Oncorhynchus nerka , subject to artificial hatchery selection since 1991, showed a 62% weakening of the cue–outcome coupling and a drop in escapement from roughly 213,000 to 45,000 — a picture of fitting overridden by imposed selection. Behavioural entropy in L. americanus fell below the 1.5-bit clustering threshold (0.27 to 0.63 bits; synchronisation r̄ = +0.86 to + 0.99), whereas T. cristatus showed the opposite pattern: high individual diversity together with low regional synchrony — a local fitting that shields the metapopulation. Partial hysteresis appeared in L. lutra as a persistent expected-minus-observed heterozygosity gap of 0.125, still present after two decades of recovery. Most tellingly, SO carried predictive information that classical variance-based early-warning signals did not (ΔR² = +0.109 for SO against + 0.015 for EWS). The framework moves Kull's semiotic fitting from qualitative description to quantitatively testable prediction, and shows across six kingdoms that the informational dimension of ecological systems is not redundant with their dynamical or demographic dimensions.