# Open English WordNet: a lexical graph of English that contributors can actually change

> Open English WordNet is an open fork of Princeton WordNet, published yearly in LMF, JSON, RDF and WNDB formats, with a companion proper noun set derived from Wikidata. It trades the frozen Princeton release for a graph that grows through reviewed contributions, and it insists that every new synset be validated by a human.

**globalwordnet/english-wordnet** — The Open English WordNet

- Repository: https://github.com/globalwordnet/english-wordnet
- Website: https://en-word.net/
- Stars: 915 · Forks: 103
- Language: Python
- License: NOASSERTION
- Published: 2026-10-20 · Updated: 2026-10-20 · Language: en
- Canonical page: https://hysenlabs.com/projects/globalwordnet-english-wordnet

## The problem: a lexical resource that nobody can correct

WordNet is the graph most English NLP still stands on. Words are grouped into synsets, senses are defined, and the senses are linked by relations such as hypernymy, antonymy and meronymy, so that a system can walk from a word up to its category or across to its opposite.

The difficulty is that the classic resource is a closed, frozen artefact. Princeton WordNet 3.1 is a fixed release, and if you want a new sense, a new compound or a correct relation, there is no route through which to add one. The Global WordNet Association's answer is to build an open version of the English wordnet under an open source methodology, with contributions reviewed rather than published by committee.

Open English WordNet is that fork. The project states plainly that the quality and veracity of the resource may differ from Princeton WordNet, and it invites contributions. It also states that contributions may eventually be incorporated into future releases of Princeton WordNet, which is the interesting part of the arrangement: the fork is a staging ground for changes to the closed resource rather than a permanent alternative to it.

The audience is an NLP engineer or researcher who needs a lexical graph, and specifically one who cares about the licensing, the release cadence or the ability to propose a correction. The data is presented as deep lexical information about English as a graph, and it is intended for natural language processing applications.

The interface is deliberately open. Correspondence to previous versions and to wordnets in other languages is provided through the Collaborative Interlingual Index, so an identifier such as oewn:2025+ means something across the family.

## The structure: synsets, lemmas and a relation graph

The data model is the familiar one, and the value of the fork is in the graph rather than in the file format. A synset is a set of words that share a sense, each sense has a definition, and the synsets are connected by typed relations.

The query example in the project documentation shows the two directions you need. Looking up a word gives you its synsets, and each synset gives you a definition:

```python
import wn

wn.download("oewn:2025+")

oewn = wn.Wordnet("oewn:2025+")

# look up a word
for synset in oewn.synsets("bank"):
    print(synset.definition())
```

Walking the graph in the other direction starts from a word, takes its synsets, and then follows the hypernym relation upwards:

```python
# find hypernyms
word = oewn.words("dog")[0]
for synset in word.synsets():
    for hypernym in synset.hypernyms():
        print(hypernym.lemmas())
```

The interesting bit in the second example is that words returns a list, and the code takes the first entry. A word like bank has several senses, so which one you traverse from is a decision the caller has to make. The library hands you the candidates and does not pick for you.

Note the identifier in both examples. The wn library is keyed by a registry identifier and an edition, here oewn:2025+, which is the Plus edition. The same library handles other wordnets, so the edition string is the thing you have to get right.

## Three versions, and what Plus adds to the core

Since the 2025 release, the project publishes three related resources rather than one, and choosing between them is the first decision a user makes.

Open English Wordnet is the core: the common nouns, verbs, adjectives and adverbs. Open English Wordnet Plus adds a selection of proper nouns taken from Open English Namenet that have been manually validated, and it is based both on the proper nouns in Princeton WordNet 3.1 and on additional entries added by the project team. Open English Namenet itself is the full set of proper nouns derived from Wikidata, published as a companion repository.

The distinction between the core and the proper noun sets is not cosmetic. Namenet is derived from Wikidata, which means it is enormous and machine generated, while the core is curated and comparatively small. The Plus edition is the interesting middle case: it is the part of the proper noun data that a person has checked.

That checking is the policy of the whole project. The documentation is explicit that automatically generated results that have not been manually validated cannot be accepted, because the resource is described as a gold standard for natural language processing. In practice that means the Namenet set is available in full, but only a validated slice of it is folded into the graph that the edition identifiers refer to.

The release table shows the cost of that policy in graph size. The 2025 Plus edition lists 161,875 words, 120,564 synsets and 419,226 relations, while the 2025 core edition is 135,969 words, 107,519 synsets and 355,064 relations. So the proper nouns add roughly twenty five thousand words and sixty four thousand relations, which is a large proportion of the graph for a set of nouns.

## The editions, the formats and how the graph has grown

Releases are published through the project website, and each edition ships in several formats. The 2025 edition is available as LMF, JSON, RDF and WNDB, the 2025 Plus edition likewise, and the earlier editions from 2019 to 2024 are offered as LMF, RDF and WNDB, with the 2025 edition the first to publish a JSON bundle.

The formats are not interchangeable in practice. LMF is the Global WordNet Association schema, and the project documentation asks contributors to read the format guidelines in FORMAT.md before submitting, because the LMF representation is the canonical one. RDF is there for the linked data world, WNDB keeps the Princeton database format for tools that still read it, and JSON is the convenience format for people who want to load the data into something without an RDF parser.

The release dates show a steady yearly cadence with one gap: 2025 edition on 31st December 2025, 2024 edition on 1st November 2024, 2023 edition on 31st October 2023, 2022 on 31st December 2022, 2021 on 9th November 2021, 2020 on 17th April 2020 and 2019 on 17th April 2019.

The growth across editions is modest and steady rather than dramatic. Words went from 160,051 in 2019 to 161,875 in 2025 Plus, and synsets from 117,791 to 120,564. The relation count grew more, from 378,201 to 419,226, which tells you that most of the work between editions went into adding and correcting links rather than new vocabulary. Princeton WordNet 3.1, shown in the same table for comparison, has 159,015 words, 117,791 synsets and 378,203 relations.

## Three ways to use it: browser, JSON API and the wn library

There are three entry points, and they suit different stages of a project.

The web interface at en-word.net is the easiest way to explore the resource. You can search for words, browse synsets and navigate the relation graph without installing anything, which makes it the right place to sanity check a word before you build a query around it. There is also a JSON API for programmatic access, documented separately at the api/docs path on the same site.

For code, the project points at the wn library, a Python package for working with wordnets programmatically. The install is:

```bash
pip install wn
```

and the download is a call in the code, not a manual file fetch, as the first example shows with wn.download("oewn:2025+"). 

The identifier in that call is worth noting, because it is the whole selection mechanism. The same library can hold several wordnets at once, and the edition string decides which graph you are looking at. A project that wants the core wordnet, the validated proper nouns and the raw Namenet set holds three identifiers.

Full documentation for the library lives at wn.readthedocs.io, and it is a separate project from this one, so the API surface you rely on is maintained on its own schedule.

## Building from source: YAML compiled into a single LMF file

The repository holds the resource as source, not as the released artefacts. The build step compiles the YAML source files into one LMF XML file:

```bash
    python scripts/from_yaml.py
```

The result is a file at wn.xml containing the complete wordnet. So the pipeline is that the source lives in YAML in the repository, the script converts it to the LMF XML form, and the release formats are then produced from that. Further conversions are handled by a separate converter hosted by the Insight Centre at server1.nlp.insight-centre.org.

The project metadata is small and gives the toolchain away: the distribution is named english-wordnet at version 0.1.0, it requires Python 3.13 or newer, and its two dependencies are pyglossary and pyyaml. PyYAML is the parser for the source files and pyglossary and pyyaml are the project's two declared dependencies for the build pipeline. A version of 0.1.0 on the packaging is another signal that this repository is a data pipeline with a thin tool wrapper, not an application.

The repository also has a working directory of documentation that tells you how to contribute properly: CONTRIBUTING.md for the process, EDITING.md, NEW_SYNSETS.md for adding synsets, DICTIONARIES.md, FORMAT.md for the LMF guidelines, TOOLS.md, RELEASING.md, and AGENTS.md alongside CITATION.cff, citation.bib and a settings.toml.

## Licence, review policy and the relationship to Princeton

The licence deserves a careful read, because more than one document is involved. The project documentation states that Open English Wordnet is released under CC-BY 4.0, and the LICENSE.md file at the repository root is that licence. The repository also ships a separate WNDB_License.txt, which covers the Princeton format and database, and the licence field on the repository itself is recorded as a custom licence that GitHub cannot classify. So the licence of the fork, the licence of the format it also ships in, and the metadata classification are three different things, and a commercial deployment should read all three.

The review policy is the second thing to understand. The documentation says the project cannot accept automatically generated results that have not been manually validated. That is why Namenet, which is derived from Wikidata and is very large, is published in full but only a validated selection is folded into the Plus edition. It also means the growth rate of the resource is bounded by human review, which is the honest trade in an open lexical project.

For citation, the canonical reference is the GWC 2019 paper by McCrae, Rademaker, Bond, Rudnicka and Fellbaum, English WordNet 2019, An Open-Source WordNet for English, published in the proceedings of the 10th Global WordNet Conference. More recent papers cover the 2020 edition and its extensions, and the CITATION.cff and citation.bib files at the repository root are there so that citation metadata is picked up automatically.

On maintenance, the repository is not archived and the last push was in October 2026. The most recent release is the 2025 edition, published on 2025-12-31, so the tree is being worked on between editions, and the annual release is the unit you can pin.

## Conclusion

Adopt Open English WordNet if you are building English NLP and want a lexical graph that is openly licensed, openly updated and convertible into the format your pipeline already reads. Do not adopt it if you need the exact Princeton 3.1 relations for a baseline comparison, since this resource is a fork with different coverage and it is the authors, not the tool, who guarantee its quality. Before you build on it, pick the edition deliberately, since 2025 Plus, 2025 and the Namenet proper noun set answer different questions, read LICENSE.md together with WNDB_License.txt, and check that the wn library can fetch the identifier you need.

## FAQ

### What is Open English WordNet and how does it differ from Princeton WordNet?

It is a fork of Princeton WordNet developed under an open source methodology, where contributions are reviewed through pull requests and may eventually feed into future Princeton releases. The project states that its quality and veracity may differ from Princeton WordNet, and the 2025 Plus edition lists 419,226 relations against 378,203 for Princeton 3.1.

### How do I query Open English WordNet from Python?

Install the wn library with pip install wn, then call wn.download("oewn:2025+") and open it with wn.Wordnet("oewn:2025+"). From there, oewn.synsets("bank") gives the synsets for a word, and synset.hypernyms() walks up the relation graph.

### What is the difference between Open English Wordnet, Wordnet Plus and Namenet?

Open English Wordnet is the core of common nouns, verbs, adjectives and adverbs. Open English Wordnet Plus adds a selection of proper nouns from Open English Namenet that have been manually validated. Open English Namenet is the full set of proper nouns derived from Wikidata, published as a companion repository.

### In which formats is Open English WordNet released?

The 2025 edition and the 2025 Plus edition are published as LMF, JSON, RDF and WNDB, while the 2019 to 2024 editions are published as LMF, RDF and WNDB. The canonical representation is the Global WordNet Association LMF format, whose guidelines are in FORMAT.md.

### Can I build the wordnet from the repository source?

Yes. The source lives in YAML and is compiled with python scripts/from_yaml.py, which produces a wn.xml file containing the complete wordnet. The project requires Python 3.13 or newer and depends on pyglossary and pyyaml.

## Sources

- [globalwordnet/english-wordnet on GitHub](https://github.com/globalwordnet/english-wordnet)
- [Issues](https://github.com/globalwordnet/english-wordnet/issues)
- [Project website](https://en-word.net/)
- [README](https://github.com/globalwordnet/english-wordnet/blob/main/README.md)
- [Releases](https://github.com/globalwordnet/english-wordnet/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/globalwordnet-english-wordnet