Open-Source English Dictionary

An open source English language dictionary in CSV, SQLite, MySQL, JSONL, and Parquet.

• v2 (2026): 388,587 definitions of 221,507 words. It combines the public-domain 1913 Webster's Unabridged Dictionary with modern words from Open English WordNet 2025+, a free, maintained word database. JSONL and Parquet.
• v1 (2021): 176,023 definitions of 111,573 words from the 1913 Webster's only. CSV, SQLite, and MySQL. These files are unchanged.

Getting Started

git clone https://github.com/CloudBytes-Academy/English-Dictionary-Open-Source

Or download the v2 data and its license:

curl -LO https://raw.githubusercontent.com/CloudBytes-Academy/English-Dictionary-Open-Source/main/v2/dictionary.parquet
curl -LO https://raw.githubusercontent.com/CloudBytes-Academy/English-Dictionary-Open-Source/main/v2/LICENSE-DATA.md

Find the usage instructions here.

v2

Files

File	Description
v2/dictionary.parquet	Parquet, one row per definition, sorted by word. Opens in pandas, Polars, DuckDB, and Spark. Carries the data license in its metadata, under the key license.
v2/dictionary.jsonl.gz	The same rows as gzipped JSON Lines, one JSON object per line.
v2/SHA256SUMS	Checksums of both files.
v2/LICENSE-DATA.md	Data license and required attribution.

Schema

Column	Type	Description
id	string	Stable ID. webster1913-000057 (row number in the v1 MySQL dump) or the Open English WordNet sense ID, such as oewn-email__2.32.00...
word	string	Headword. Webster's words are lowercase. Open English WordNet words keep their case, such as Paris. Compare words case-insensitively.
pos	string or null	Part of speech: noun, verb, adjective, adverb, preposition, conjunction, pronoun, interjection, prefix, suffix, or article.
wordtype	string or null	Webster's original abbreviation, such as v. t. or imp. & p. p.. Null for Open English WordNet rows.
definition	string	Definition on one line.
examples	list of strings	Example sentences. Empty for Webster's rows.
synset_id	string or null	Open English WordNet synset (group of synonyms). Rows with the same synset_id are synonyms. Null for Webster's rows.
source	string	webster1913 or oewn.

Example rows:

{"id":"webster1913-031182","word":"computer","pos":"noun","wordtype":"n.","definition":"One who computes.","examples":[],"synset_id":null,"source":"webster1913"}
{"id":"oewn-computer__1.06.00..","word":"computer","pos":"noun","wordtype":null,"definition":"a machine for performing calculations automatically","examples":[],"synset_id":"oewn-03086983-n","source":"oewn"}

Contents

Source	Rows	Notes
Webster's 1913	175,928	From the v1 MySQL dump, cleaned.
Open English WordNet 2025+	212,659	Adds 109,790 words that Webster's does not have: 42,668 single words, such as smartphone, email, and blog, and 67,122 phrases, such as domestic dog.

What v2 fixes in the Webster's data

• Restores 224 words, such as -able and -ance, that a spreadsheet had turned into #NAME? in the v1 CSV and SQLite files.
• Repairs garbled characters: 80¡ becomes 80°, and Ã2 becomes √2. Also £, ÷, and §.
• Joins definitions that had hard line breaks onto one line.
• Removes 93 duplicate rows and 2 empty definitions.
• Maps 271 of the 293 word-type spellings, including typos such as supperl. and n ., to the 11 values in pos. The original is kept in wordtype.

Known issues

• 54 Webster's headwords, and some definitions, contain / where an accented letter was lost before this data reached OPTED, such as assaf/tida for assafœtida.
• 3,919 Webster's rows have no pos: 3,882 have no word type, and 37 have one that is not a part of speech, such as obs. or See.
• About 16,000 Webster's rows are inflected forms whose definition only names the base word, such as abandoned, imp. & p. p., of Abandon.
• Open English WordNet examples belong to the whole synonym group, so an example can use a synonym instead of the headword.

Rebuild

The build is reproducible. With the pinned pyarrow version, which uv installs, rebuilds give a byte-identical dictionary.parquet (verified on Linux with Python 3.11 to 3.13). dictionary.jsonl.gz always holds the same rows, but its compressed bytes depend on the zlib library in your Python: Python 3.14 and later on Windows uses zlib-ng, which writes a different file, so its line in SHA256SUMS changes there. The build also checks the download against a pinned SHA-256. It needs uv.

uv run scripts/build_v2.py
uv run --with 'pyarrow>=15' --with duckdb --with pandas python -m unittest discover tests

The tests that read the public URLs need network, so they run separately:

uv run --with 'pyarrow>=15' --with duckdb --with pandas python -m unittest discover tests -p 'remote_*.py'

The MariaDB import test runs only when MYSQL_HOST names a server. It ignores MariaDB option files, connects over TCP, and also reads MYSQL_TCP_PORT (default 3306), MYSQL_USER (default root), and MYSQL_PWD.

Local CI

This repo has no GitHub Actions. Before you open or update a pull request, push your commit, then run:

python3 scripts/local_ci.py               # rebuild, compare v2/, run the offline tests, then sign off
python3 scripts/local_ci.py --no-signoff  # the same checks, no status

It fails if the rebuild changes anything in v2/. When every check passes, it posts a signoff status on the tested commit. A signoff covers one commit, so run it again after you push more commits. The script refuses to sign off when the worktree is dirty, when HEAD changes during the run, or when GitHub does not have the commit. The status description says without MariaDB when MYSQL_HOST is not set. It does not run the network tests.

v1

The v1 dictionary has 3 fields

1. word: In lowercase
2. word type: Abbreviations describe the type, e.g. verb, noun, etc.
3. definition: Definition of the word in sentence case

Database Schema

Both databases use a table named entries, but their column types and nullability differ.

Column	SQLite	MySQL dump
word	TEXT, nullable	varchar(25) NOT NULL
wordtype	TEXT, nullable	varchar(20) NOT NULL
definition	TEXT, nullable	text NOT NULL

SQLite has an index named ix_entries_word on word. A query on lower(word) cannot use it and scans the whole table. The SQLite file contains 3,895 rows with a NULL wordtype and 2 with a NULL definition. CSV and the MySQL dump use empty strings for these values. The MySQL dump uses MyISAM and latin1.

Dictionary Format & Repository Structure

Format & Link	Description
CSV	A single file with all the words in standard CSV format
SQLITE3	A single file formatted as a SQLITE3 database
MYSQL	MySQL dump that can be imported directly inside MySQL / MariaDB

Usage Instructions

Language	Instructions are available for
Python	Parquet, JSONL, CSV, SQLITE3
SQL	DuckDB, querying Parquet without downloading it first

History

v1 is based on the Source Forge Project: MySQL English Dictionary, which in turn is based on The Online Plain Text English Dictionary (OPTED).

OPTED is a public domain English word list dictionary, based on the public domain portion of "The Project Gutenberg e-text of Webster's Unabridged Dictionary" which is in turn based on the 1913 US Webster's Unabridged Dictionary. See Project Gutenberg.

License

• Code: MIT.
• v1 data and Webster's rows in v2: public domain in the USA. OPTED asks redistributors to acknowledge OPTED, Project Gutenberg, and the 1913 Webster's, and to contact Project Gutenberg before commercial use. See v2/LICENSE-DATA.md.
• Open English WordNet rows in v2 (source = 'oewn'): CC BY 4.0, plus the WordNet notice. If you share the v2 files, keep v2/LICENSE-DATA.md with them.

Credits

1. Project Gutenberg: For providing the original 1913 US Webster's Unabridged Dictionary. Make sure you read the Project Gutenberg's README for license and other details if you care considering using this for commercial purposes.
2. x16bkkamz6rkb78: For compiling the MySQL dump and releasing on Source Forge.
3. dumblob: For providing the extraordinarily elegant mysql2sqlite tool for converting the MySQL dump to SQLite3.
4. The Open English WordNet Team: For Open English WordNet, the source of the modern words in v2. See v2/LICENSE-DATA.md for the citation.