Chainle terminology · glossary
Semantic similarity
Definition: A measure of how close two words are in meaning, derived from word-use patterns and expressed here as a strength from 0 to 100.
Examples
- 'car' and 'automobile' score very high; 'car' and 'banana' score near zero.
Semantic similarity is the scoring engine behind Chainle: a number that says how close two words are in meaning. It is derived from word-use patterns — the company each word keeps in real text — which is why it captures associations and categories as well as strict synonyms. In practice, each word is represented as a vector (GloVe vectors), and how closely two vectors align becomes a strength from 0 to 100.
In a Chainle round, your submitted word is compared with every clue separately, and each comparison yields one of these strengths. The round score then combines them: 80% from the geometric mean of the strengths and 20% from the weakest link, scaled into the 0-1,000 per-round range. That is why a word can be excellent for one clue and still score poorly overall — weak similarity to the other clues pulls the combined number down.
Why it matters
Understanding that the game measures meaning, not spelling or dictionary definitions, explains most scoring surprises: associations, shared materials, and familiar phrases all count, and lopsided answers are punished. The same underlying idea powers similar word games such as Semantle, so learning it here transfers elsewhere.