We use cookies to understand how you use this site and improve your experience.
In 1940, a fire insurance inspector named Benjamin Lee Whorf published a paper that would haunt linguistics for the next century.
Whorf had spent years studying the Hopi language of northeastern Arizona, and he'd noticed something strange: Hopi grammar didn't handle time the way English did. Where English forces every verb into past, present, or future tense, Hopi used a different system entirely. One that distinguished between events that are manifesting, events that are manifested, and events that are expected. Not when something happens, but how it relates to the process of becoming real.
Whorf made a leap. If Hopi grammar structures time differently than English grammar, then maybe, maybe, Hopi speakers experience time differently than English speakers. Not because the Hopi are mysterious or exotic, but because the grammar you use every day creates the default pathways through which you conceptualize reality.
His mentor, Edward Sapir, had said it more carefully: "The 'real world' is to a large extent unconsciously built up on the language habits of the group."
Together, their ideas became the Sapir-Whorf hypothesis: the proposal that language doesn't just express thought, but shapes it.
The hypothesis has been controversial. In its strongest form, language determines thought, making certain ideas literally unthinkable, it's been mostly rejected. Humans are resourceful. We coin words, borrow from other languages, use metaphor to reach beyond what grammar provides. We have spatial intuition, embodied experience, emotional knowledge that exists before and beneath language.
But in its weaker form, language influences thought, creating default pathways and making certain concepts more accessible, the evidence has become compelling. I want to share three experiments that changed how I think about this.
Lera Boroditsky, a cognitive scientist at UC San Diego, has spent two decades demonstrating these effects. Russian speakers have two mandatory words for blue: goluboy (light blue) and siniy (dark blue). English speakers use one word and qualify when needed. Winawer and colleagues showed that Russian speakers discriminate between light and dark blue than English speakers. Not because Russian eyes see differently, but because the linguistic boundary creates a perceptual boundary. When your language forces a distinction, your brain automates that distinction.
Stephen Levinson documented that speakers of Guugu Yimithirr, an Aboriginal Australian language, use absolute cardinal directions instead of relative ones. They say "pass the cup to the north" rather than "pass the cup to the left." The result: they maintain a constant internal compass, an awareness of cardinal direction that persists even in enclosed spaces, even after being spun around.
Boroditsky herself showed that English speakers, who use horizontal metaphors for time ("looking forward to the weekend"), think about time differently from Mandarin speakers, who additionally use vertical metaphors. Prime Mandarin speakers with vertical spatial cues and they respond faster to temporal questions.
The mechanism is always the same: not determinism, but grooves. Channels in cognition through which thought flows more easily. You can think against the grain of your language, but it takes effort. The default is to think with it.
I didn't start with Whorf. I started with a frustration.
I was building behavioral contracts for AI agents, files that told a model how to interact with a specific part of a codebase. I tried Markdown first. Natural language instructions. They worked sometimes. The model would read the instructions, follow some, forget others, interpret a few in ways I never intended.
Then I noticed: when I structured the same instructions differently, gave them explicit types, enforcement levels, scoped sections, the model's compliance improved dramatically. Not because I'd added information. I'd added structure. The same content, organized differently, produced different reasoning.
That's when Whorf's insight hit me. The structure of the notation was creating cognitive grooves. Default paths through which the model's reasoning flowed. The content was the same. The medium was different. And the medium was doing the real work.
For a long time I thought of notation as a container, something that holds information. Somewhere along the way I started seeing it as a landscape. The surface creates valleys and ridges, and thought flows downhill.
If this is true, if notation structure shapes AI cognition the way language structure shapes human cognition, then notation design is not a formatting question. It's a cognitive architecture question.
Every choice in a notation, what gets named, what gets typed, what gets enforced, what gets compressed, creates grooves. Default paths that make certain thoughts easier and certain thoughts harder.
The question isn't whether this happens. Every prompt engineer who's seen better output from well-structured prompts already knows it happens. The question is whether we design those grooves deliberately, or leave them to accident.
That question is what led me to build YON. And the rest of this series explores the intellectual lineage behind it, from McLuhan's media theory to Clark's extended mind, from Alexander's pattern languages to Borges's constructed realities.
It starts with grooves. Everything starts with grooves.
Next in the series: The Medium, on why form matters more than content.
Loading comments…