QSalience introduces the first application-agnostic salience predictor for inquisitive questions. This project is fine-tuned over our newly annotated dataset of 1,766 (context, question) pairs with salience scores. We explore the usefulness of question salience in downstream applications such as TL;DR expansion.
Datasets are organized as follows:
- General data: salience
- Training, validation, and testing sets: train_val_test
Models and Code will be released after decision out.
Relevant data can be found here: answerability