Evaluating Theory of Mind in Question Answering

TL;DR

This paper introduces a new dataset to evaluate question answering models' ability to reason about beliefs, inspired by theory-of-mind experiments, revealing current models' limitations in handling inconsistent world states.

Contribution

The paper presents a novel dataset for belief reasoning in question answering and assesses neural models, highlighting their failure to handle belief inconsistencies.

Findings

Models fail on belief reasoning tasks

Model accuracy drops with random sentences

Current neural models struggle with inconsistent states

Abstract

We propose a new dataset for evaluating question answering models with respect to their capacity to reason about beliefs. Our tasks are inspired by theory-of-mind experiments that examine whether children are able to reason about the beliefs of others, in particular when those beliefs differ from reality. We evaluate a number of recent neural models with memory augmentation. We find that all fail on our tasks, which require keeping track of inconsistent states of the world; moreover, the models' accuracy decreases notably when random sentences are introduced to the tasks at test.