Open-domain clarification question generation without question examples

TL;DR

This paper introduces a framework for generating clarification questions in dialogue using an information gain approach, without relying on supervised question-answer data, to improve communication success.

Contribution

It presents a novel visually grounded question-asking model that generates polar questions to resolve misunderstandings without requiring question-answer training data.

Findings

Model produces informative questions that enhance dialogue clarity.

Questions improve success rate in a 20 questions game with synthetic and human answers.

Framework operates without supervised question-answer pairs.

Abstract

An overarching goal of natural language processing is to enable machines to communicate seamlessly with humans. However, natural language can be ambiguous or unclear. In cases of uncertainty, humans engage in an interactive process known as repair: asking questions and seeking clarification until their uncertainty is resolved. We propose a framework for building a visually grounded question-asking model capable of producing polar (yes-no) clarification questions to resolve misunderstandings in dialogue. Our model uses an expected information gain objective to derive informative questions from an off-the-shelf image captioner without requiring any supervised question-answer data. We demonstrate our model's ability to pose questions that improve communicative success in a goal-oriented 20 questions game with synthetic and human answerers.