Languages simplify over time. This is the received view — inherited from centuries of philological tradition and reinforced by common examples. Latin’s six cases became French’s none. Old English’s gendered nouns became Modern English’s genderless ones. Inflectional complexity decays, and what takes its place is simpler: word order, prepositions, auxiliaries.

Hariharan and Mortensen (arXiv:2512.05364) analyze 2,000 years of Sanskrit using transformer-enabled diachronic methods on a 1.47-million-word corpus. The finding: morphological complexity does not decrease. It redistributes. Verbal features decline cyclically — the inflectional system for verbs simplifies — but compounding expands dramatically. Nominal complexity shifts. New philosophical terminology emerges, carrying fresh morphological structure. The total complexity budget stays roughly constant; it just moves from one part of the grammar to another.

The through-claim: simplification and complexification can occur simultaneously in the same language, and what looks like a decrease depends entirely on where you measure. Track verbs and the language simplifies. Track compounds and it complexifies. Track the whole system and it migrates. The “simplification” narrative is a measurement artifact — the consequence of monitoring one subsystem while the complexity moves to another.

This is a conservation law for linguistic complexity, or at least a near-conservation. The analogy to physical systems is not metaphorical: when one degree of freedom is damped, the energy redistributes to other degrees of freedom. The language doesn’t become simpler; it becomes differently complex. The verbal flexibility that Sanskrit loses reappears as nominal compounding, as semantic precision in new vocabulary, as syntactic patterns that weren’t present in the earlier period.

The lesson is general: any system measured at one scale will appear to simplify if the complexity is migrating to another scale. To see the conservation, you need the whole system. And “the whole system” is always larger than the subsystem that gave you the impression of decline.