Our English lexical resources, ready to use.
These are the resources behind OMEGA's English corrector, dictation and Hangman — built the way the French ones were, but for English: a spelling lexicon of ~124,000 surfaces with IPA and subtitle frequency, homophone groups, inflected forms, character and phoneme n-grams, and a data-driven grapheme→phoneme table. Small, plain files — free to reuse for any project.
Served as individual files (link each below). No personal data: nothing comes from a private corpus.
What's available
| File | Content | Source |
|---|---|---|
lex_en.tsv.gz | Spelling lexicon — ~124,000 surfaces, columns surface·pos·ipa·lemma·tags·gender·freq. | Wiktionary · SUBTLEX-US · CMUdict |
homophones_en.json | ~5,500 symmetric homophone groups (their↔there↔they're, right↔rite↔write). | Wiktionary |
forms_en.tsv.gz | ~91,000 lemmas with inflected forms (go→went→gone, big→bigger→biggest). | Wiktionary |
ngrams_ortho_en.json.gz | Character n-grams (uni/bi/tri) with word boundaries — graphotactics. | Wiktionary · SUBTLEX-US |
ngrams_phon_en.json.gz | Phoneme n-grams over a normalised 40-phoneme General-American inventory — phonotactics. | Wiktionary · CMUdict |
g2p_en.json | Data-driven grapheme→phoneme table (context-dependent IPA + surprisal), aligned by EM over 30,000 word↔IPA pairs. | derived from lex_en |
verbmorph_en.json | Regularised irregular verbs (runned→ran, teached→taught) — a common dyslexia/L2 error. Built so that forms the data attests as valid (betted, dreamed, hanged) and US/GB doubling variants (traveled/travelled) are never flagged. | AGID · lex_en |
Plain formats: TSV and JSON (the .gz files decompress with any tool, gunzip included).
The OMEGA Dys font — install it in Word, LibreOffice, Pages…
A research font for written-language disorders, designed blind from our own measurements and then checked against the literature: voiced consonants are heavy (b d g v z j), voiceless ones are light (p t c k q f s), mirror letters b/d/p/q each get their own mark, accents are enlarged, spacing is wide. Three weights (Regular, Light, Heavy), free licence (SIL Open Font License).
Installed in Word it encodes voicing per letter. The full "sound font" — the s of poison rendered voiced, silent letters in vermilion, syllables — needs our engine: it lives in the dictation, the corrector and the extension. Research prototype: not yet measured on real readers.
Word add-in (beta): the full sound font inside Word — every word re-inserted character by character as styled runs (voiced heavy, voiceless light, silent vermilion, syllables). Download the manifest · install (sideload) & limits. Not yet tested inside Word by us: the logic is tested against the real engine, the first real run is still to come.
Free to reuse
We did the heavy lifting — extracting, cross-referencing, normalising the IPA to a clean 40-phoneme inventory, learning the grapheme→phoneme alignment. It's that version we're sharing, ready to use, not the raw material.
Built from the English Wiktionary (via kaikki.org), SUBTLEX-US subtitle frequencies, the CMU Pronouncing Dictionary, and AGID (Automatically Generated Inflection Database, Copyright 2000-2016 by Kevin Atkinson — permission to use, copy, modify, distribute and sell this database and the output created from it is granted without fee). Wiktionary/kaikki content is CC BY-SA 4.0: keep derivatives open and credit OMEGA-Ω + these sources if you reuse.