This repository contains the code and resources for the contrapositive inference model presented in the following paper:
Omid Kashefi, Rebecca Hwa, 2021. "Contrapositive Local Class Inference." in EMNLP Workshop on Noisy User-generated Text (W-NUT).
The model is implemented as a transfer function
on top of ctrl-gen
model (Hu et al., 2017), modified with an additional conciseness loss
constraint to ensure the class transference with minimal changes to the original greater-context.
python3 model\contrapositive-tf.py --config config.[task]