Evidence from counterfactual tasks supports emergent analogical reasoning in large language models

TL;DR

This paper defends previous findings that large language models can perform analogical reasoning in zero-shot settings, demonstrating their ability to generalize to counterfactual tasks despite critiques.

Contribution

It clarifies misunderstandings about the original tests and provides evidence that language models can generalize to counterfactual analogy tasks.

Findings

Language models solve counterfactual analogy tasks.

Models generalize beyond training data.

Clarification of previous misunderstandings.

Abstract

We recently reported evidence that large language models are capable of solving a wide range of text-based analogy problems in a zero-shot manner, indicating the presence of an emergent capacity for analogical reasoning. Two recent commentaries have challenged these results, citing evidence from so-called `counterfactual' tasks in which the standard sequence of the alphabet is arbitrarily permuted so as to decrease similarity with materials that may have been present in the language model's training data. Here, we reply to these critiques, clarifying some misunderstandings about the test materials used in our original work, and presenting evidence that language models are also capable of generalizing to these new counterfactual task variants.