Playing log(N)-Questions over Sentences

TL;DR

This paper introduces a two-agent game where a questioner learns to ask discerning questions about sentences, integrating responses to refine its hypothesis, highlighting challenges in balancing game accuracy and question quality.

Contribution

It presents a novel two-agent game framework for sentence-based questioning and demonstrates the complexities of training agents to optimize both accuracy and question quality.

Findings

Agents can learn to play the game end-to-end

High game accuracy and meaningful questions are a challenging trade-off

Simultaneous learning of both agents is feasible but complex

Abstract

We propose a two-agent game wherein a questioner must be able to conjure discerning questions between sentences, incorporate responses from an answerer, and keep track of a hypothesis state. The questioner must be able to understand the information required to make its final guess, while also being able to reason over the game's text environment based on the answerer's responses. We experiment with an end-to-end model where both agents can learn simultaneously to play the game, showing that simultaneously achieving high game accuracy and producing meaningful questions can be a difficult trade-off.