Understanding Semantic Space

Understanding Semantic Space

How word embeddings create a geometry of meaning that AI uses to understand language

Updated: January 25, 2026

Understanding Semantic Space

What If Meaning Had a Shape?

Here's something remarkable: when you teach a machine to read billions of sentences, it discovers that words have positions.

Not positions on a page—positions in a space of meaning. "Happy" sits near "joyful." "Cat" neighbors "dog." "Ocean" and "sea" are almost touching. Meanwhile, "refrigerator" and "existentialism" are very, very far apart.

Nobody programmed this. The machine figured it out by reading how humans use words together. And when you visualize it, something magical happens: the structure of human thought becomes visible.

Word Space 3D lets you explore this structure. You're not just playing a guessing game—you're navigating the actual geometry of meaning that AI has learned from us.

How Words Become Coordinates

When an AI reads the sentence "The cat sat on the mat," it doesn't see words. It sees patterns. It notices that "cat" often appears near "dog," "pet," "fur," and "meow." Over billions of sentences, these patterns become stable. The machine builds an internal map.

This map is called an embedding. Each word gets a list of numbers—512 of them in our case—that encode everything the machine learned about that word's meaning. Think of it as GPS coordinates, but for concepts instead of locations.

"telescope" → [0.023, -0.156, 0.891, 0.044, ..., -0.234]  (512 numbers)
"microscope" → [0.019, -0.148, 0.877, 0.051, ..., -0.229]  (very similar!)
"happiness" → [-0.445, 0.712, -0.033, 0.298, ..., 0.156]  (completely different)

Words with similar meanings get similar numbers. This isn't a trick—it's a discovery. The AI found structure in language that linguists suspected but couldn't quantify.

The Problem of 512 Dimensions

There's a catch: you can't visualize 512 dimensions. Your brain maxes out at three on a good day.

So we compress. We use an algorithm called UMAP that squeezes 512 dimensions down to 3, while fighting to preserve what matters: neighborhood relationships. Words that were close in the full space should stay close in 3D.

Think of it like a map projection. Greenland looks enormous on a flat map because you're squishing a sphere onto a plane. Something similar happens here. The 3D view you see is a useful approximation—a window into a space too vast to see directly.

Some distortion is inevitable. Two words might appear close in 3D but have only moderate similarity. Two words might look far apart but actually be quite related. The 3D visualization is a guide, not gospel.

That's why we show you the similarity score—a percentage calculated in the full 512-dimensional space. When the map and the score disagree, trust the score. It's measuring in the space we can't see.

What Clusters Reveal

Spend time in Word Space and you'll notice something: words clump together.

Animals form a neighborhood. Emotions cluster. Tools gather. Colors, foods, weather, professions—each finds its region. This isn't something we designed. It emerges naturally from how humans use language.

Here's what's fascinating: the boundaries aren't always where you'd expect.

"Bat" (the animal) sits with mammals, but also reaches toward "ball" and "baseball." The word carries multiple meanings, and the embedding reflects all of them. It's pulled in different directions.

"Cold" (the temperature) neighbors "hot," "warm," and "freezing." But "cold" (the illness) has connections to "flu," "sick," and "sneeze." Context shapes meaning, and the embedding captures this ambiguity.

When you guess a word and see the cluster it lands in, you're learning something about how language organizes concepts. The game teaches you to think in semantic neighborhoods. Family resemblances.

Similarity: The Fourth Dimension

The percentage you see isn't calculated from the 3D positions. It's the actual similarity between your guess and the mystery word, measured in the full 512-dimensional space.

Think of it this way:

  • X, Y, Z tell you where a word is (its semantic address)
  • Similarity tells you how close it is to the target (your temperature reading)

They're related but not the same. You can be in the right neighborhood (similar X, Y, Z) but still far from the target. Or you can be pointing the right direction despite appearing distant in 3D.

The colors help:

  • Blue means cold—you're far away
  • Red means hot—you're getting close

The arrows point toward the mystery word, growing longer as you warm up.

The Related Words: Your Semantic Radar

When you guess a word, we show you nearby words from our vocabulary. Here's the crucial detail:

  • Selected by proximity to YOUR GUESS: These are words similar in meaning to what you typed
  • Scored by similarity to the MYSTERY WORD: Each one shows how warm that region is

This gives you a radar sweep. If you guess "ocean" and see "sea" (62%), "water" (58%), "wave" (55%), and "beach" (51%), you know this whole region runs from the mid-50s to low-60s. Not freezing, not hot. Worth exploring nearby, but probably not the exact answer.

If you guess "chair" and see "table" (23%), "furniture" (21%), "seat" (19%)—everything's cold. This neighborhood isn't it. Move on.

The related words help you think in clusters, not just individual guesses. You're mapping the territory.

What This Teaches About AI

Word Space isn't just a game. It's a window into how modern AI understands language.

Every chatbot, every search engine, every translation system—they all use embeddings like these. When you ask ChatGPT a question, it converts your words into coordinates and navigates a space much like this one.

Playing Word Space gives you intuition for:

Why AI sometimes makes strange mistakes. If "bank" (financial) and "bank" (river) occupy similar regions, the AI might confuse them. You can see how polysemy creates ambiguity.

Why context matters so much. A word's meaning isn't fixed—it depends on neighbors. The embeddings capture this, which is why "cold beer" and "cold personality" land in different places.

Why AI is both impressive and limited. The geometry of meaning is real and the machine found it. But the map is compressed, approximate, and sometimes misleading. Understanding this helps you understand AI's capabilities and failures.

The Wonder of Discovered Structure

Here's what still amazes us:

Nobody told the AI that "king" relates to "queen" the same way "man" relates to "woman." It figured this out from patterns in text. The famous equation king - man + woman ≈ queen actually works in embedding space.

Nobody told the AI that colors form a spectrum, or that emotions have opposites, or that tools cluster by function. These structures emerge from the statistics of human language.

When you play Word Space, you're exploring a map that was drawn by reading. Every position encodes something about how humans think and communicate. The clusters you see are clusters in our collective understanding.

That's worth a moment of wonder.

Playing as Learning

The best way to understand semantic space is to get lost in it.

Guess wildly at first. See where things land. Notice which words neighbor each other and which sit far apart. Pay attention to surprises—the words that land somewhere unexpected teach you the most.

Over time, you'll develop intuition. You'll start thinking in regions: "I'm in the animal neighborhood, but I need something more domestic." You'll learn to triangulate: "These three guesses are all 40-50%—the answer must be somewhere between them."

This intuition transfers. Next time you use a search engine and it returns strange results, you'll have a sense of why. Next time an AI translation seems off, you'll understand the ambiguity it faced.

Word Space is a puzzle game. It's also a tutorial in how machines understand meaning.

Welcome to semantic space.

Going Deeper

Want to understand the technical details?