← GuidesContexto

Contexto Isn't About Synonyms. Here's What It's Actually About.

Most players approach Contexto like a thesaurus game. It isn't. It's a geometry problem — and once you understand the space you're navigating, your guesses get dramatically more efficient.

What "Semantic Similarity" Actually Means

Contexto and Semantle rank your guesses using word embeddings — mathematical representations of words based on the contexts they appear in across millions of documents. Two words are "semantically similar" in this model not because they mean the same thing, but because they tend to appear near the same other words in text.

This distinction is crucial. SURGEON and SCALPEL aren't synonyms, but they appear in similar contexts — hospitals, operations, medical descriptions — so they're semantically close in most embedding models. SURGEON and DOCTOR are closer still. SURGEON and NURSE are close. SURGEON and HOSPITAL are close in a different way — not because they mean the same thing, but because they co-occur constantly.

How an embedding model ranks words relative to OCEAN
#1
OCEANThe answer
#8
SEASemantic neighbor — near-synonym
#24
MARITIMEShares context — ships, ports, trade
#67
WAVEAssociated — appears in ocean contexts
#189
BLUELoose association — ocean is blue
#412
SWIMActivity in oceans — weak link
#891
VACATIONBeach vacation — very indirect
OCEANThe answer
SEASemantic neighbor — near-synonym
MARITIMEShares context — ships, ports, trade
WAVEAssociated — appears in ocean contexts
BLUELoose association — ocean is blue
SWIMActivity in oceans — weak link
VACATIONBeach vacation — very indirect

Why Your Intuitive Guesses Keep Failing

The most common Contexto mistake is guessing synonyms. If your best guess is HAPPY at rank 45, your instinct is to try JOYFUL, PLEASED, CONTENT, CHEERFUL. These are the right semantic neighbors in a thesaurus. They're often not the right neighbors in an embedding model.

Embedding models are trained on real text — not dictionaries. In real text, HAPPY and BIRTHDAY appear together constantly ("happy birthday," "happy occasion," "happy couple"). So BIRTHDAY might rank closer to HAPPY than JOYFUL does, even though BIRTHDAY and HAPPY have very different meanings.

The co-occurrence insight: Instead of asking "what means the same as this word?" ask "what words appear in the same sentences as this word?" The answer to the second question is often surprising — and often closer to the Contexto answer.

The Three Zones and How to Navigate Them

Think of Contexto as three concentric zones around the target word:

1

Zone 1: Top 50. You're in the right semantic neighborhood. Every guess should now be a word that appears in the same narrow context as your best guesses. If WAVE and TIDE are both in the top 50, you're looking at ocean-related words — not 'water' generically, but the specific vocabulary of tides, currents, and maritime contexts.

2

Zone 2: Ranks 50–300. You're in the right broad category but approaching from the wrong angle. Lateral moves work here: if MUSIC is at rank 200, try specific music vocabulary — CHORD, MELODY, TEMPO, RHYTHM, BEAT — rather than synonyms like SONG or SOUND.

3

Zone 3: Ranks 300+. You haven't found the category yet. Use broad exploratory guesses across completely different semantic fields: emotions, objects, actions, places, professions, colors. You're mapping the space, not closing in on the answer.

Advanced Technique: The Triangulation Method

Once you have three or four guesses in the top 100, you have enough data to triangulate. Look for the overlap — the semantic intersection of all your best guesses. What concept appears in the context of all of them simultaneously?

If SAIL (#23), CAPTAIN (#41), HARBOR (#67), and WIND (#89) are all in the top 100, the answer probably isn't OCEAN (too broad) or BOAT (too specific). It might be VOYAGE, SHIP, or NAVIGATION — words that sit at the intersection of all four contexts.

The "what story would use all these words?" test: When you're stuck, mentally write the first sentence of a story that naturally uses all your top-ranking guesses. The subject of that story is often close to the answer. Four high-ranking words about sailing might produce: "The CAPTAIN guided the ship out of the HARBOR under full SAIL, with a strong WIND at her back." The subject — SHIP or VOYAGE — is worth trying.

What the Closer Tool Can and Can't Do

The Closer tool on this site analyzes your guesses and their ranks, then suggests vocabulary to explore. It's useful — but it's important to understand its limitation: Claude's semantic model is different from Contexto's embedding model. A word Claude thinks is semantically close might rank 600 in Contexto. A word Claude thinks is peripheral might rank 15.

Use Closer as a vocabulary prompt, not a prediction engine. When it suggests MARITIME after you've told it OCEAN is close, it's giving you a word worth trying — not guaranteeing it'll rank well. The suggestions are most useful for expanding your vocabulary in the right direction, not for zeroing in on the answer.

The Meta-Pattern Across Many Puzzles

Players who track Contexto answers over time notice that the answers tend to be concrete, common nouns — not abstract concepts, not proper nouns, not obscure words. The most frequent answer types are everyday objects, animals, body parts, food, professions, and places. Abstract concepts like FREEDOM or JUSTICE appear occasionally but are rarer.

This means if you're deep into a puzzle with your best guess at rank 800+, the answer is probably a word you use every day — not a word you've never heard of. Guessing simple, high-frequency words across different semantic fields is a better strategy than guessing sophisticated vocabulary.

The frequency reset: When stuck above rank 500, spend five guesses on the most common nouns in the English language: TIME, YEAR, PEOPLE, WAY, MAN, WOMAN, DAY, HAND, PART, PLACE, CASE, WEEK. These aren't glamorous guesses, but they cover a huge swath of semantic space and frequently unlock a new direction.

Semantle vs. Contexto — Key Differences

Semantle (semantle.com) uses a different word2vec-style embedding model than Contexto. The rankings are similar in principle but different in practice — a word that ranks #10 in Contexto might rank #80 in Semantle. Semantle also shows a "similarity score" as a percentage in addition to rank, which provides more granular feedback.

The strategy is the same for both, but the specific vocabulary that performs well can differ. If you play both games regularly, you'll notice that Semantle tends to reward more formal vocabulary while Contexto rewards everyday language — a reflection of the different training corpora used by each model.

Ready to put this into practice?

Open Closer← All guides