Answer-based Adversarial Training for Generating Clarification Questions
This addresses the problem of improving conversational AI and information retrieval systems by generating more effective clarification questions, though it is incremental as it builds on existing GAN and sequence-to-sequence methods.
The paper tackles the problem of generating clarification questions to make textual contexts more complete by modeling hypothetical answers as latent variables, resulting in a GAN-based approach that outperforms retrieval-based models and ablations on usefulness, specificity, and relevance metrics.
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.