What survives when a word enters the internet

Before measuring yellow, we asked a simpler question. Does visual meaning survive compression?

For one image, the local word for heart was used to retrieve thousands of images from the internet in six languages. Every result was kept: emoji, anatomy, jewellery, cartoons and noise. The field was then averaged into a single surface.

Spanish

corazón

Portuguese

coração

Arabic

قلب

Hebrew

לב

Danish

hjerte

Dutch

hart

Different languages. Different search fields. The same image returns.

A heart is a designed symbol, built for recognition. It therefore collapses cleanly into one shape. That is the easy case. The more revealing test is a word nobody designed.

Here are the six averages side by side. Thousands of images per language, compressed into one surface each. The difference is visible before it is explained.

Spanish

árbol

Portuguese

árvore

Arabic

شجرة

Chinese

树

Danish

træ

Anomaly

Dutch

boom

Spanish, Portuguese, Arabic, Chinese and Danish settle into the same vertical structure: trunk, crown, axis, green. The Dutch average is warmer, looser and broken. The tree is barely holding together. Something orange is bleeding through it.

This is not a flaw in the method.
It is the method finding something.

The Dutch anomaly does not mean that Dutch people picture trees differently. It means that the word boom enters a global image field, not a Dutch one.

In Dutch, boom means tree. Online, however, it collides with the international English boom: the comic-book sound of an explosion. One search direction produces two visual populations competing for the same space.

boom · tree

boom · blast

Two populations, one field

Drag the line. These are the two image populations behind the Dutch average. Switch to Blend and stop in the middle. That mixture is close to what the averaging produces.

ReadingHard cut

Hard cut Blend

Position

The hard cut keeps the two meanings legible. The blend shows what happens inside the dataset: two visual archetypes try to occupy one frame, and neither wins cleanly.

The result is diffuse.

Run the two halves of boom separately and the collision becomes visible. One average is still recognisably a tree. The other is an explosion. Inside it, the letters of the word that produced it begin to surface.

Population A: tree

The Dutch meaning. A diffuse but legible tree: trunk, crown, axis.

Population B: blast

The international meaning. A comic explosion. The word BOOM surfaces in the average itself.

One surface: the real average

Searched together, both populations fold into a single image: a ghost tree under a wash of explosion orange. The average does not choose. It records the collision.

This single average is a miniature version of the whole project. Words do not exist in isolation. On the internet, meanings compete for visibility. The compression keeps the score.

Heart has high density and survives intact. Tree has high density in five languages. Dutch boom has low density, because two archetypes share the word and neither dominates. The average becomes a measure of how stable a meaning remains once it passes through the internet.

Yellow Drift applies the same method to colour. The question is not whether yellow is represented correctly. It is how stable the visual field of yellow remains when thousands of references are compressed into one measurable surface, and how far that surface drifts from a fixed point.

That fixed point is #FFFF00. It is not the true yellow and not a correction. It is simply a stable place to measure from.