soutsios / pos-tagger-bert-tensorflow

BERT fine-tuning for POS tagging task (google's tensorflow)

BERT POS tagger (tensorflow)

Fine-tune Google's BERT for POS-tagging task (UD Treebank as the dataset).

Folder Description:

pos-tagger-bert-tensorflow
|____ pos_tagger_bert_tensorflow.ipynb  # Notebook with all actions required to download UD Treebank dataset and BERT model, fine-tune BERT, and evaluate POS tagger
|____ bert_pos.py           # Main code
|____ data                  # Train data
|____ middle_data           # Middle data (label id map)
|____ output                # Output (final model, predicted results)
|____ uncased_L-12_H-768_A-12	# BERT model downloaded from -> https://storage.googleapis.com/bert_models/2018_10_18/uncased_L-12_H-768_A-12.zip
|____ run_pos.sh    		# Run model and evaluate results

Install packages:

pip install pyconll  # for UD treebank reading
pip install bert-tensorflow # for using bert model

Usage:

Preferably run notebook pos_tagger_bert_tensorflow.ipynb, or

bash run_pos.sh

What's in run_pos.sh:

#!/usr/bin/env bash

  python  bert_pos.py\
    --task_name="POS"  \
    --do_lower_case=False \
    --crf=False \
    --do_train=True   \
    --do_eval=True   \
    --do_predict=True \
    --data_dir=data   \
    --vocab_file=./uncased_L-12_H-768_A-12/vocab.txt  \
    --bert_config_file=./uncased_L-12_H-768_A-12/bert_config.json \
    --init_checkpoint=./uncased_L-12_H-768_A-12/bert_model.ckpt   \
    --max_seq_length=220   \
    --train_batch_size=16   \
    --learning_rate=2e-5   \
    --num_train_epochs=4.0   \
    --output_dir=./output/result_dir

perl conlleval.pl -o '[SEP]' -r -d '\t' < ./output/result_dir/label_test.txt

References:

[1] https://arxiv.org/abs/1810.04805

[2] https://github.com/google-research/bert

Acknowledgement

Natural Language Processing course is part of the MSc in Computer Science of the Department of Informatics, Athens University of Economics and Business. The course covers algorithms, models and systems that allow computers to process natural language texts and/or speech.