Distilling Hypernymy Relations from Language Models: On the Effectiveness of Zero-Shot Taxonomy Induction

TL;DR

This paper evaluates zero-shot methods for inducing taxonomies from language models, demonstrating their competitive performance and highlighting the importance of prompt design and linguistic properties.

Contribution

It provides a comprehensive analysis of zero-shot taxonomy induction, showing their effectiveness and the influence of prompt properties on performance.

Findings

Zero-shot methods outperform some supervised approaches.

Prompt design significantly impacts taxonomy induction quality.

Statistical and linguistic properties of prompts are crucial for success.

Abstract

In this paper, we analyze zero-shot taxonomy learning methods which are based on distilling knowledge from language models via prompting and sentence scoring. We show that, despite their simplicity, these methods outperform some supervised strategies and are competitive with the current state-of-the-art under adequate conditions. We also show that statistical and linguistic properties of prompts dictate downstream performance.