Shakespeare translations using TensorFlow

This is an example of using the new Google's TensorFlow library on monolingual translation going from modern English to Shakespeare based on research from Wei Xu.

Prepare

First download the TensorFlow library depending on your platform:

pip install https://storage.googleapis.com/tensorflow/mac/tensorflow-0.5.0-py2-none-any.whl # for mac
pip install https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.5.0-cp27-none-linux_x86_64.whl # for ubuntu

1. Grabs parallel data.
2. Gets train, dev split.
3. Builds vocabulary
4. Converts parallel data into ids

From the root directory:

python -m tensorshake.get_data
python -m tensorshake.prepare_corpus

Delete /cache to start anew.

Train

Use the example BASH script to train the model. This saves the check points in the --train_dir directory. If you run it again, the training process continues from the check point. To restart with fresh parameters, simply delete/rename the check points.

./run.sh

Results

Benchmarks from original paper. (Shakespeare -> Modern English)

Cherrypicked examples from this repo (Modern English -> Shakespeare)

Pre-Trained Models

Here is a link for an example model: https://s3-us-west-2.amazonaws.com/foxtype-nlp/tensorshake/model_cache.zip

Possible improvements

• word embeddings
• beam search
• language model reranking