This BERT based POS tagger is derived almost entirely from https://github.com/Kyubyong ipython notebook
https://github.com/Kyubyong/nlp_made_easy
Requirements
- Pytorch (conda install -c pytorch pytorch)
- Pytorch pre-trained BERT . (pip install pytorch-pretrained-bert)
- nltk (pip install nltk)
Data
- Automatically fetches treebank training data using nltk (when running first time, it will prompt to install treebank. This can be done from within python prompt in command line)
Usage
-
To train. python bert_post_train.py < model dir to save. e.g. out > . (this will result in training with accuracy ~98%)
-
To test. python bert_post_test.py < model dir to load >
Example input: The bird flew over the house and perched on a tree
Output: [('The', 'DT'), ('bird', 'NN'), ('flew', 'VBD'), ('over', 'IN'), ('the', 'DT'), ('house', 'NN'), ('and', 'CC'), ('perched', 'VBD'), ('on', 'IN'), ('a', 'DT'), ('tree', 'NN')]
License
MIT License