Credits
Stonele is built on other people’s generous work.
Word data
- ConceptNet Numberbatch 19.08 by Robyn Speer, Joshua Chin and Catherine Havasi (Luminoso Technologies, Inc.), github.com/commonsense/conceptnet-numberbatch, licensed CC BY-SA 4.0. How related two words are.
- ConceptNet 5.7 by the Commonsense Computing Initiative and contributors, conceptnet.io, licensed CC BY-SA 4.0. Part, material and whole relations (PartOf, HasA, MadeOf), reviewed by hand into extra links such as human → arm. ConceptNet itself includes data from Wiktionary, WordNet, Open Mind Common Sense and other sources: see Copying and sharing ConceptNet.
- WordNet 3.0, Princeton University, wordnet.princeton.edu. Base forms of words, part-of relations, and which nouns can start or end a puzzle. WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved. Full WordNet 3.0 license.
- GloVe by Jeffrey Pennington, Richard Socher and Christopher D. Manning, nlp.stanford.edu/projects/glove, Public Domain Dedication and License (PDDL) 1.0. Word-frequency order only.
- ENABLE word list by Alan Beale and M. Cooper, public domain. Which words count as English words.
- Names Corpus 1.3 by Mark Kantrowitz, with additions by Bill Ross, via NLTK. Used to keep first names out of the word list.
- List of Dirty, Naughty, Obscene, and Otherwise Bad Words, github.com/LDNOOBW, licensed CC BY 4.0. Used only to remove words from the word list; the list itself isn’t published here.
What I changed
I kept about 31,000 base forms of the 50,000 most common English words. I centered the Numberbatch vectors, reduced them to 96 dimensions and stored them as 8-bit integers, worked out each word’s closeness thresholds, removed names and blocked words, and turned hand-reviewed ConceptNet and WordNet part-of relations into a list of extra links.
My adapted data is CC BY-SA 4.0
The word data files this site serves are adapted from the material above, and I license them under CC BY-SA 4.0. You may copy and adapt them under that license:
- /data/vectors.bin, /data/bars.bin, /data/links.bin, /data/words.txt, /data/lemmas.json and /data/meta.json
These files are provided as is, without warranties of any kind; see the license for the full disclaimer. The Commonsense Computing Initiative, Luminoso, Princeton University and the other authors above don’t endorse Stonele.
Fonts
- Schibsted Grotesk, Copyright 2023 The Schibsted-Grotesk Project Authors, github.com/schibsted/schibsted-grotesk, SIL Open Font License 1.1 (license).
- Source Serif 4 by Frank Grießhammer for Adobe, github.com/adobe-fonts/source-serif, SIL Open Font License 1.1 (license).
Both are served unmodified from this site, as built by Fontsource.
The game
Code, design and puzzle selection © 2026 Stonele.