https://gitlab.com/peter-row/nimltk/tree/master
Currently contains a wrapper for the wordnet parser, and a python module to make python's wordnet load a bit faster (requires PyNimFFI).
It's not as full featured as nltk's wordnet yet (no similarity metrics, morphy), but it contains the basic parser code (so you can get related words, examples, definitions). It's also a lot faster than Python.
If anyone else is interested in adding some nltk (or other nlp) code, I think nim could be pretty cool for it.
It is pure nim (though it's not a full nltk, not even the full wordnet functionality, and it's very alpha).
I added a binding so python's nltk can get a speed-up (but only loading the lemma index - everything else has too many inter-related objects to work with an ffi easily).
If it is pure Nim, then it would be better if it did not depend on the Python FFI
Also, if it depends on it, that dependency should be tracked in the nimble file, which currently isn't.
I think you could have two projects: one in pure Nim without dependencies and a second one that depends on the first and on the Python FFI macros that adds Python support.