About this dictionary
The TKG English Learner's Dictionary is an original English-English learner's dictionary written by language models under the direction of Tom Gally, a lexicographer, and published as a static website from the repository github.com/tkgally/eex-dict. It is meant for intermediate and advanced learners: the definitions are plain enough for an intermediate reader, and the grammar, collocations, and notes are complete enough for an advanced one.
Who wrote it
Every entry was written by language models and reviewed by other language models; no human has checked most entries. The entries are drafted by a language model from its own knowledge and a written house style, then checked field by field by two other language models; a further model adjudicates their objections and records every decision. Pronunciations and inflected forms are checked by agreement between several models. Nothing in an entry is copied from another dictionary. Errors remain possible on every page: if a definition or example looks wrong, it may well be wrong.
Frequency bands and levels
Each entry shows a frequency band in words, from very common to rare. The bands are the dictionary's own editorial estimates: several models were asked how basic, common, and important each word is, and the answers were settled editorially. No corpus counts, external frequency lists, or published word lists were used, and the bands are not CEFR levels. The defining vocabulary mark shows a word that belongs to the about 2,500 words the definitions are written with.
Counts
- 493 entries published
- 459 headwords
- 469 phrases
- 455 defining-vocabulary entries
| Band | Entries |
|---|---|
| very common (1) | 460 |
| common (2) | 7 |
| fairly common (3) | 26 |
| less common (4) | 0 |
| rare (5) | 0 |
From the project's run records
- 73 runs recorded; the latest on 2026-09-26.
- Entries in the repository at that run: 493 (493 reviewed, 0 draft).
- Defining vocabulary: 422 of 2394 words have an entry (18%).
- Headword queue: 4360 pending, 493 done.
Licences
The dictionary data (every entry, in JSON and as rendered here) is released under CC0 1.0: it is in the public domain and may be copied, changed, and reused for any purpose without permission. The code that builds and runs this site is released under the MIT licence. The repository contains only the project's own work: no word list, frequency list, pronouncing dictionary, or other data set is included or derived from, whatever its licence.
Resources consulted
The register below lists every outside resource the project's tools consult while checking a fact. None of them is included in the dictionary; a resource is read at run time and only the verdict is recorded.
Register of resources consulted at run time
This directory holds markdown only. No data file is ever stored here or anywhere in the repository. A resource listed below may be fetched to a temporary directory outside the repository (.tmp/, gitignored) to check a single fact, and the entry then records only the verdict (for example checked_by: "agreement:3/3;cmudict:agree"). Nothing is copied from any of them into an entry, a list, or a table.
| Resource | Used for | How | Verified reachable |
|---|---|---|---|
CMU Pronouncing Dictionary (cmudict.dict, raw file on GitHub, cmusphinx/cmudict) | one more vote on an American transcription in tools/pronounce_check.py | fetched to .tmp/, ARPAbet converted to IPA by a fixed mapping in the tool, compared, discarded | 2026-09-16 (HTTP 200, about 3.6 MB) |
English Wiktionary (en.wiktionary.org API, action=parse) | a hand check of a sample of British transcriptions in the pronunciation experiment only; not part of the pipeline | read in the session, the verdict recorded per word | 2026-09-16 (HTTP 200) |
OpenRouter model list (openrouter.ai/api/v1/models) | confirming that the model slugs in config/models.md exist | fetched, read, discarded | 2026-09-16 (443 models) |
Not consulted for anything: frequency lists (wordfreq, Google Books Ngrams), published word lists (the New General Service List, the Oxford lists, the General Service List), WordNet, and any commercial or copyrighted dictionary. The defining vocabulary, the headword queue, and the frequency bands were built from the models' own judgment alone; the method is in ../wiki/decisions/defining-vocabulary.md.
Originality check: about every tenth Routine run, tools/originality_check.py samples definitions for exact-phrase web searches and a reviewer question; the results go to ../reviews/originality/.
Reading an entry
Grammar codes are explained in the grammar key, usage labels in the labels key, and the phonetic symbols in the pronunciation key. Every word in a definition, explanation, example, or note that has an entry of its own is a link; resting the pointer on it (or tapping it once on a phone) shows a preview. In an explanation or a note, a word in bold is being talked about as a word, and a phrase in italics is quoted as an illustration of use; in an example sentence the headword itself is in bold. The translator's view button at the top reveals notes written for anyone adapting the dictionary into another language: how the senses split across languages, grammar traps, cultural background, false friends, and pronunciation traps.
Site built on 2026-09-26.