alvations

@stevenbird something like a `class NLTKTokenizer` object and we create a tokenizer by picking the regexes we like from other tokenizer. That way, we're not restricted by license and fear...

Could you share where nltk is called in your code? It couldn't have been inside the LSTM cell in your keras code. Issue #947 was rather old and we haven't...

Hmmm, the `my_nltk_module` -> `nltk` is a warning sign of circular imports. What's in my_nltk_module? Can you comment away everything in my_nltk_module and see whether the `import nltk` gets you...

If it has something to do with serialization and pickle. Try this in your `my_nltk_module.py`: ``` import dill as pickle import nltk ``` See whether it's still get stuck.

@TristanJM just to check-in, did you happen to be getting the error on a Mac OSX? Which OS are you using?

Thanks @TristanJM for the information. Yes, it looks like Mac OSX and tinkter and multiprocessing aren't playing well. This is the same issue as #947. We'll track this issue and...

@villmow could you post the code with more context that caused the Attribute error? It looks like you have been looking at the wrong object to check for the `names()`....

Thanks @villmow, now it's clear =) Checking the type before accessing the attributes would be a quick fix.

At https://github.com/nltk/nltk/blob/develop/nltk/corpus/reader/wordnet.py#L197 Maybe this could fix it, the `type()` checks should be evaluated first. ```python def __eq__(self, other): return type(self) == type(other) and self._name == other._name def __ne__(self, other): return...

+1 on the training recipes doc! Esp. if there's any additional preprocessing steps to convert textfiles to fairseq2 trainable format like in fairseq 1.