01 model laboratory
Meaning is not
one number.
Put two words under the microscope. Compare vector similarity, corpus association, lexical relations, semantic axes, sound, spelling, and soft categories in real time.
THE SAME PAIR, DIFFERENT QUESTIONS
coffee ↔ tea
A high score means something different in every lens. The useful design question is which difference creates a good decision for the player.
DATAMUSE + WORDNET
Means like
Broad semantic similarity: paraphrases, related concepts, and neighbors.
ml=coffeeNo explicit edge.
That absence is information too: this lens is sparse and typed, unlike a dense vector model.
SOFT GROUPS
Recharts radarCategory fingerprint
coffee tea
INTERPRETABLE DIRECTIONS
prototype axesSemantic compass
LOCAL GEOGRAPHY
D3 forceNeighbor constellation
- espresso
- tea
- caffeine
- cafe
- mug
- breakfast
- bitter
- drink
TYPOGRAPHIC DENSITY
Motion wordsAssociation field
espressoteacaffeinecafemugbreakfastbitterdrink
HOW TO READ THIS
Dense models always answer
MiniLM places any text in a continuous 384-dimensional space. Its score is useful, but its reason is latent.
Typed models can stay silent
WordNet relations are precise and explainable, but many pairs have no direct edge. Sparse can be a feature.
Axes are designed interpretations
Each compass direction is built from opposing anchor sets. Change the anchors and you change the game.
Maps are interfaces
The constellation is a force layout of real neighbor links—not a claim that the screen is the true geometry of meaning.