What Does My QA Model Know? Devising Controlled Probes using Expert Knowledge

TL;DR

This paper introduces systematic, expert-knowledge-based probes to evaluate what factual and taxonomic knowledge state-of-the-art QA models truly possess, revealing their strengths and limitations in lexical and hierarchical reasoning.

Contribution

It presents a novel methodology for automatically creating controlled knowledge probes from expert sources, enabling comprehensive evaluation of QA models' knowledge understanding.

Findings

QA models recognize some lexical knowledge but struggle with hierarchical reasoning.

Performance drops with increased complexity and distractor answers.

Models show room for improvement in cluster-based semantic evaluations.

Abstract

Open-domain question answering (QA) is known to involve several underlying knowledge and reasoning challenges, but are models actually learning such knowledge when trained on benchmark tasks? To investigate this, we introduce several new challenge tasks that probe whether state-of-the-art QA models have general knowledge about word definitions and general taxonomic reasoning, both of which are fundamental to more complex forms of reasoning and are widespread in benchmark datasets. As an alternative to expensive crowd-sourcing, we introduce a methodology for automatically building datasets from various types of expert knowledge (e.g., knowledge graphs and lexical taxonomies), allowing for systematic control over the resulting probes and for a more comprehensive evaluation. We find automatically constructing probes to be vulnerable to annotation artifacts, which we carefully control for. Our evaluation confirms that transformer-based QA models are already predisposed to recognize certain types of structural lexical knowledge. However, it also reveals a more nuanced picture: their performance degrades substantially with even a slight increase in the number of hops in the underlying taxonomic hierarchy, or as more challenging distractor candidate answers are introduced. Further, even when these models succeed at the standard instance-level evaluation, they leave much room for improvement when assessed at the level of clusters of semantically connected probes (e.g., all Isa questions about a concept).