Answer-based Adversarial Training for Generating Clarification Questions

TL;DR

This paper introduces an adversarial training method for generating clarification questions that are more useful and relevant by modeling hypothetical answers as latent variables within a GAN framework.

Contribution

It proposes a novel GAN-based approach that uses a utility function to generate more effective clarification questions, outperforming existing models.

Findings

Outperforms retrieval-based models in usefulness and relevance

Generates more specific and contextually appropriate questions

Effective on multiple datasets with human and automatic evaluations

Abstract

We present an approach for generating clarification questions with the goal of eliciting new information that would make the given textual context more complete. We propose that modeling hypothetical answers (to clarification questions) as latent variables can guide our approach into generating more useful clarification questions. We develop a Generative Adversarial Network (GAN) where the generator is a sequence-to-sequence model and the discriminator is a utility function that models the value of updating the context with the answer to the clarification question. We evaluate on two datasets, using both automatic metrics and human judgments of usefulness, specificity and relevance, showing that our approach outperforms both a retrieval-based model and ablations that exclude the utility model and the adversarial training.