18 comments

[ 3.0 ms ] story [ 51.6 ms ] thread
I’d love to see a similar interactive UI for a language model embedding layer.

The ones that I’ve seen don’t seem to let you easily navigate all the interesting properties and relationships.

As a visual dictionary, this is not as useful as conceptnet, compare vs "conceptnet" on common words such as "red", "apple".
Thanks for pointing out conceptnet. On GraphWords, "synonym" only returns "equivalent word", conceptnet adds "antonym" (as entire category!) as I'd been looking for...
WordNet has antonym and other semantic relationship information, so it should be possible to add that to GraphWords.
What is the relationship between words modelled as? Levenshtein or semantic?
Semantic. Specifically, showing the synonym relationships in the WordNet database. Those are manually constructed.
Ah - synonyms. Makes sense, thanks
If you like this, definitely check out WordNet (https://wordnet.princeton.edu/)
And Open English WordNet (https://en-word.net/) which builds on WordNet 3.0 and 3.1, improving the core data both with new words and with fixes to the hierarchy.
> And Open English WordNet (https://en-word.net/) which builds on WordNet 3.0 and 3.1, improving the core data both with new words and with fixes to the hierarchy.

This looks dope!

Do you recommend some other viewer type software to use on OS X or is everything self contained in the zip file?

The zip files contain the complete data in various formats.

For the WNDB (WordNet DataBase) format, you could use the WordNet viewer or any WordNet compatible software. This will not be able to view all the Open English WordNet data, as it has parts of speech and relationship types not in the original WordNet.

For the RDF/Turtle (Resource Description Framework) format, you could use any RDF visualization tool. I don't have any specific recommendations.

For the LMF/XML (Lexical Markup Framework XML) format, you would need an XSLT stylesheet to convert the data to something else like HTML.

You could also run https://github.com/jmccrae/wordnet-angular locally, which is the project used to run the https://en-word.net website. That takes the LMF/XML data, creates a SQLite database for it, and runs a web server on that database.

Cool tool but I don’t think I’d call this a visual dictionary. With that title I imagine an image for each word. This is more of a mind map dictionary.
This resource could be an aid to beginner ESL learners who struggle with common words. However, most words I encountered that I had to look up recently give empty or misleading results.

E.g., “aleph” results in an empty graph, and “heron” results in a false graph where all relations are to semantically unrelated “hero” except one to “Heron” the person. WordNet dataset actually does define “heron” as a bird, but somehow that was lost.

I got some bizarre results. Entering “taco” returns nothing having to do with food.