Grammatical gender isn't random. Hidden inside the world's languages is an ancient memory of what could be owned — and what belonged to everyone.
Explore the pattern →This isn't coincidence. The neuter gender in Proto-Indo-European languages consistently encodes resources that were shared — things no one could own. The pattern survives across thousands of years and dozens of languages. And where it disappeared, something was lost.
Click any language family to explore its relationship with commons-encoding grammar. The pattern holds — and the exceptions are just as revealing.
Languages don't assign gender randomly. The neuter gender in Proto-Indo-European languages consistently marks resources that were commons — things shared, unownable, available to all: water, fire, bread, salt, gold as raw material, sky, sea.
Masculine and feminine genders, by contrast, mark things that exist within property relations — persons, animals, tools, roles. Things that can be owned, inherited, controlled.
This pattern holds across the Indo-European family: Germanic, Slavic, Baltic, Greek, Latin, Sanskrit. Where neuter was preserved, the commons/property distinction survived in the grammar. Where it was lost — as in the Romance languages after Rome's collapse — everything became masculine or feminine. Everything became property.
The implications reach into AI. Language models trained on text alone struggle most with exactly the semantic domains where commons-encoding matters most — physical relations, spatial relations, shared resources. The grammar was trying to tell us something that text alone cannot capture.
Active independent research. Cross-linguistic dataset in development. Peer-reviewed publication in preparation.
Core paper documenting the neuter=commons pattern across 13+ language families with falsifiable predictions and cross-linguistic evidence. Target: peer-reviewed linguistics journal.
Structured dataset mapping gender assignments for 200 core vocabulary items across language families, coded for commons/property status. Open release planned.
Analysis testing whether commons-category semantic domains predict LLM benchmark difficulty across languages. Builds on EWoK-core-1.0 evaluation framework.
Sign up below to be notified when the preprint is available.
Selected works that inform and contextualise this research.
Darryl Lopes is an independent researcher based in Mallorca, Spain. Of South African and Portuguese origin, he pursues self-directed inquiry at the intersection of historical linguistics, cognitive science, and artificial intelligence — outside institutional structures and without disciplinary constraint.
This research began not in an academic institution but at a kitchen table. While learning the very basics of German, Lopes noticed something that the textbooks treated as arbitrary: why was das Brot neuter? Why was bread — of all things — grammatically neither masculine nor feminine? The question refused to leave. Systematic investigation followed: cross-linguistic comparisons, archival research, and extended conversational research with AI systems that helped stress-test, refine, and ultimately confirm a pattern that had gone untheorised for centuries.
What began as a beginner's puzzlement became a hypothesis spanning 13+ language families, signed languages, and neuroimaging research. The origin outside academia is not incidental — it may be precisely the kind of question that institutional training conditions one not to ask.
A cross-linguistic dataset and peer-reviewed publication are in preparation. He can be reached at research@grammaticaleconomics.org.
Be notified when the preprint is published, the dataset is released, and new research is available.
No spam. One email when something significant is published.
Thank you — you will be notified when the preprint is available.
For media inquiries, collaboration proposals, or academic correspondence.