Artifacts from 34,000 to 45,000 years ago — ivory figurines, bone tools, cave plaques from the Swabian Jura — carry patterns of dots, lines, notches, and crosses. For decades, these marks were classified as decorative: repetitive, culturally significant perhaps, but not informational. Decoration is the default category for marks that don't look like writing.
A computational study of over 3,000 Paleolithic signs (PNAS 2026) measured their information density and found it matches proto-cuneiform — the earliest known writing system, which appeared 37,000 years later in Mesopotamia. The symbols are not random. Dots appear on human and feline figures. Crosses appear on mammoths and horses but never on human forms. The assignment of specific symbols to specific subjects is consistent within and across sites. Diversity, repetition frequency, and combinatorial structure all fall within the range of administrative recording systems.
This is not writing. There is no evidence of syntax, phonetic encoding, or narrative structure. But it is not decoration either. It occupies the space between — a structured sign system with measurable information content, subject-specific symbol assignment, and cross-site consistency. The 37,000-year gap between these marks and proto-cuneiform isn't empty. It is occupied by systems that were invisible because the available categories — art or language — had no room for them.
The general principle: when a phenomenon falls between two recognized categories, it tends to be assigned to the simpler one. Marks that aren't writing become decoration. Behavior that isn't language becomes instinct. The assignment feels conservative but it discards information: the structure that doesn't fit either category is exactly what makes the phenomenon interesting. Measuring what you assumed was noise is how you find the intermediate systems that the taxonomy erased.