Skip to content

Latest commit

 

History

History
64 lines (56 loc) · 2.69 KB

README.md

File metadata and controls

64 lines (56 loc) · 2.69 KB
Raw

E2E NLG

Code for the INLG 2018 conference paper Syntactic Manipulation for Generating more Diverse and Interesting Texts

Setup & Requirements

We assume an Anaconda installation. To install the Conda environment run:

$ conda env create -f environment.yml

This installs tensorflow-gpu==1.10.0, make sure that you have the CUDA environment installed accordingly. Refer to this for the compatibility.

Next create some output directories, inside the main directory run the following commands:

$ mkdir logging
$ mkdir outputs
$ mkdir models

Running the Code

Preprocessing

First run the preprocessing scrips:

$ sh preprocess_data.sh

This will handle the whole preprocessing pipeline:

  • Parse the Meaning Representations and transform them into vectors.
  • Parse the Utterances and transform them into vectors.
  • Store the output into "data/e2e_nlg/preprocessed_data/version1"

Experiments

To run the experiments:

$ sh run_experiments.sh

This will run all the training procedures and create models for the various synthactic manipulations:

  • The models will be stored into the "models/sc_{name_of_experiment}" folder.
  • Every 10 epochs the outputs of the models are generated and stored into "logging/sc_lstm/{experiment_name}/training_log_{time-stamp}/" folder.

Generate Outputs

To generate the final outputs run:

$ sh run_output_generation.sh

This will produce the outputs for each of the model and store it into "outputs/sc_{name_of_experiment}". Here the sampling procedure is used to select the final utterance. Thus, the generation might take a while to complete, especially for the "full" model.

Evaluation

To run the rulebased evaluation scripts:

$ sh run_evaluation.sh

This creates a file called "rule_based_eval.txt" inside the "outputs/sc_{name_of_experiment}" folder. There the incorrect utterances are displayed and the final error rate score.

References

Please cite the following paper when using this code or pretrained models for your application.

Jan Deriu and Mark Cieliebak, Syntactic Manipulation for Generating more Diverse and Interesting Texts INLG 2018 - International Conference on Natural Language Generation

@inproceedings{deriu2018nlg,
  title = {{Syntactic Manipulation for Generating more Diverse and Interesting Texts}},
  author = {Deriu, Jan and Cieliebak, Mark},
  booktitle = {INLG 2018 - International Conference on Natural Language Generation},
  address = {Tilburg, The Netherlands}
  year = {2018},
}