Cemantle is the French-language sibling in the family of semantic and daily word games: the Semantle mechanic, rebuilt on a model trained on French text. It launched within weeks of the original in 2022 and has kept a steady francophone following since. If you already play Contexto in English or Portuguese, Cemantle is the clearest demonstration in the whole genre that these games are made of their training data, not just their rules.
What Cemantle is
One secret French word per day. You guess in French; the game returns a similarity score, plus a rank if your guess falls inside the thousand nearest words. The scoring is the same cosine-similarity measure Semantle uses, presented the same way — a temperature reading rather than a position.
The interface keeps Semantle’s austerity: a text field, a table of your guesses sorted by score, and nothing else. There are no colour bands, no letter feedback and no cap on attempts. What differs is everything underneath — the vocabulary, the neighbourhoods, and which words the model considers adjacent.
How to play
- Guess a French noun. Concrete nouns give the model more to work with than adjectives or verbs.
- Read the score. Under 10 means you are nowhere near; French models tend to run slightly warmer than English ones, so 12–15 is worth investigating rather than discarding.
- Once a word breaks 25, work its semantic field: near-synonyms, the category above it, the objects it usually appears with.
- When you land inside the top 1,000, follow the rank rather than the score.
- Accents matter for spelling but not for scoring — the model normalises them, so cafe and café resolve to the same vector.
A new word appears at midnight Paris time, which is worth knowing if you play from outside Europe.
Why the language changes the game
This is the interesting part, and the reason Cemantle deserves its own page rather than a footnote on the Semantle one. Word embeddings encode how a culture writes, so a French model and an English model disagree about closeness in ways that are consistently surprising:
- Culturally weighted words move. Terroir, laïcité and quartier sit in dense, well-connected regions of the French model and sparse ones in English. A French answer can have a rich near-orbit that simply has no English equivalent.
- Grammatical gender leaks in. Nouns that share a gender co-occur with the same articles and adjectives, which nudges them fractionally closer together than meaning alone would put them.
- Compounds behave differently. English glues concepts into single tokens — lighthouse, seafood — while French spreads them across phrases. The model has no vector for pomme de terre as a unit, so vegetable rounds play very differently in the two languages.
The same effect is why Contexto runs a genuinely separate word list per language rather than translating one answer into three. A Brazilian Portuguese round built around saudade has no meaningful English mirror, and translating it would produce a puzzle that is either trivial or nonsensical.
Cemantle vs Contexto
Cemantle is a temperature game; Contexto is a ranking game. Against Cemantle’s raw similarity score, most guesses look identical — a run of low numbers with no direction in them. Contexto’s rank-out-of-30,000 always tells you something, because even a bad guess has a position.
The other practical difference is support. Contexto ships a tip system with three difficulty settings, a per-language archive and a visible band structure; Cemantle ships a text box. That makes Cemantle the better choice on a day when you want a long, quiet, unaided puzzle, and Contexto the better choice when you want to actually finish before lunch. Our strategy guide’s category-sweeping method transfers to Cemantle almost unchanged, since both games reward systematic exploration over clever single guesses.
If you are learning French, Cemantle is unusually good practice — it forces you to generate vocabulary in a semantic field rather than recognise it, which is the harder and more useful direction.
Where to go next
Semantle is the English original and the direct comparison point — playing the same week in both languages is the fastest way to feel how much the corpus matters. For something structurally different, Wordle remains the letter-based benchmark that all of these games were built in reaction to.
The full games-like-Contexto hub compares all twelve titles we track, including the daily geography and grid games that use the same warmer/colder idea outside of vocabulary.