The Mystery of Compositional Generalization in Graph-based Generative Commonsense Reasoning

TL;DR

This paper introduces a new challenge for testing compositional generalization in graph-based commonsense reasoning, revealing that even large language models struggle with unseen relation combinations and that structured training can improve their performance.

Contribution

It presents the CGGC challenge for graph-based reasoning, analyzes factors affecting generalization, and demonstrates that curriculum learning enhances LLMs' compositional reasoning.

Findings

Large language models struggle with unseen relation combinations.

Structured demonstration order improves generalization.

Different graph structures vary in difficulty for models.

Abstract

While LLMs have emerged as performant architectures for reasoning tasks, their compositional generalization capabilities have been questioned. In this work, we introduce a Compositional Generalization Challenge for Graph-based Commonsense Reasoning (CGGC) that goes beyond previous evaluations that are based on sequences or tree structures - and instead involves a reasoning graph: It requires models to generate a natural sentence based on given concepts and a corresponding reasoning graph, where the presented graph involves a previously unseen combination of relation types. To master this challenge, models need to learn how to reason over relation tupels within the graph, and how to compose them when conceptualizing a verbalization. We evaluate seven well-known LLMs using in-context learning and find that performant LLMs still struggle in compositional generalization. We investigate potential causes of this gap by analyzing the structures of reasoning graphs, and find that different structures present varying levels of difficulty for compositional generalization. Arranging the order of demonstrations according to the structures' difficulty shows that organizing samples in an easy-to-hard schema enhances the compositional generalization ability of LLMs.