# compute pmi using seq2seq method.
# base on google nmt.

## train 
    train.sh
    log.src2tgt     # the log of training
    log.tgt2src     # the log of training

    ### also nedd
        train/dev/test data
        vocab       # a word-character vocabulary to avoid unknown words problem
        embedding   # the embedding size mush be same as the units num of model

## compute pmi
    compute_pmi.py
    ./scripts/pmi.sh    # the shell to run compute_pmi.py

    ### also need
        remain_words.txt    # used when recutting sentences