PunktTokenizer does not use the correct version of the pickled model on Python 3.x

[BLKSerene]

Hi, I'm trying to package my program with NLTK and nltk_data using PyInstaller. So to minimize the size of the data file, I removed the zip file and all models for Python 2.x in nltk_data/tokenizers/punkt (only the PY3 folder is left).

But it seems that the PunktTokenizer always uses the Python 2.x version of the pickled model regardless of Python version I'm using. And the error message says that it can't find tokenizers/punkt/english.pickle instead of tokenizers/punkt/PY3/english.pickle.

Removing the PY3 folder is okay, so it seems that the Python 3 version of the pickled model is never used.

OS: Windows 10 64-bit
Python version: 3.7.2 64-bit
NLTK version: 3.4