NLP-Notes

Some fun NLP projects based on Pytorch.

• Pytorch: 1.0.1
• Python: 3.6

1. LSTM part-of-speech tagger with character-level features

Train and test

• The following command starts training.

python lstm_tag_plus.py

Refs

• Exercise: Augmenting the LSTM part-of-speech tagger with character-level features
• pytorch-discuss
• Understanding LSTM Networks

2. 普通话-四川话翻译器

Train and test

• The following command starts training.

python main.py

• Examples:
  • 普通话> 你想做什么 四川话: 你想做啥子
  • 普通话> 你怎么这么那个 四川话: 你啷个这么那个
  • 普通话> 我明天不上班 四川话: 老子明天不上班
• A jupyter-notebook version can be found in pt-sc/pt_sc.ipynb.

Refs

• pytorch-official-tutorials
• jieba-introduction

3. compare the sum embeddings and average embeddings based on BOW(bag of words) model

4. Dive-into-DL-PyTorch

• 只截取了其中第10章NLP部分，包括word2vec, glove, seq2seq, machine translation..., 详见Dive-into-DL-PyTorch