How WordFor ranks candidate words
When you describe a meaning, WordFor scores hundreds of thousands of dictionary entries and returns the best matches in order. Here's what goes into that ranking, and why the list looks the way it does.
Watch it in action
1. Lexical match
First, literal overlap between the words in your query and the words in each entry's definition and synonyms. If you type "wordy", entries whose definitions literally contain "wordy" get an immediate boost. This is fast and precise when your wording happens to line up with the dictionary's.
2. Semantic similarity
Most reverse-dictionary queries don't share words with the answer — you describe a concept in your own words. So WordFor converts both your query and every entry into embeddings (numeric meaning-vectors) and measures how close they are. This is what lets "lasting a very short time" find ephemeral and fleeting even with no shared words.
The embeddings are computed in your browser with a small transformer model — no query is ever sent to a server.
3. Source confidence
Each word is defined by one or more public-domain dictionaries (WordNet, Webster 1913, the Century Dictionary, Chambers 1908, and others). Entries that are well attested across reliable sources, with richer and more specific definitions, are weighted up; thin or obscure entries are weighted down. This keeps genuinely useful words above noise.
4. Lemma family grouping
English words come in families: run / runs / running / ran. Showing all of them would clutter the list, so at build time WordFor groups inflected forms under a single lemma and shows one row per family. Crucially, it only collapses forms that are genuinely the same word — it never merges words that just share a prefix (unhappy is not happy; invaluable is not valuable).
5. Reranking (fast first pass, precise second pass)
Scoring every entry at full precision for every keystroke would be slow, especially on a phone. So WordFor uses two passes:
- First pass — binary. A compact binary representation scores the whole dictionary in milliseconds and shortlists the best candidates. On mobile this can be the only pass, which is why search feels instant.
- Second pass — reranking. On desktop (and after the background download finishes) a higher-precision representation re-scores just the shortlist, sharpening the order at the top of the list.
This is why the page is usable immediately in "fast mode" and then quietly upgrades to "best ranking" once the reranker has loaded — you never wait for the big download to start searching.
Putting it together
The final score is a blend: lexical and semantic similarity decide the candidates, source confidence breaks ties toward trustworthy words, lemma grouping tidies the list, and reranking polishes the top results. For vague, feeling-based queries the semantic signal dominates; for precise definitional queries the lexical signal carries more weight.
See it from the user's side in how to find a word from a meaning, and the data side in how we use public-domain dictionaries.