Skip to content

Latest commit

 

History

History

train

CTC: A Unified Framework for Evaluating Natural Language Generation

Requirements

Our python version is 3.7. Run these commands before you start:

cd data/
unzip topical_chat.zip
unzip yelp.zip
pip install -r requirements.txt
python -m spacy download en

Alignment Model Training

The constructed data and trained models can be directed downloaded here. If you want to run the process, see below.

Data Construction

The script to construct data for our discriminative (D) and regression (R) model: scripts/construct_data.sh.

Constructed data will be saved in constructed_data/.

Training Discriminative Model (D)

The script to train the discriminative models: scripts/train_discriminative.sh.

Checkpoints will be saved in ckpts/.

Training Regression Model (R)

The script to train the regression models: scripts/train_regression.sh.

Checkpoints will be saved in xxx/.

Test Correlation

The script to test alignment models: scripts/test_correlation.sh (instance-level) & scripts/test_correlation_system.sh (system-level).

Correlation scores will be printed in the running terminal, and evaluation scores for all examples will be saved in eval_results/.

Results

All result numbers can be found in the appendix of our paper.
(exactly the same numbers would be got with our trained models.)