Exploring the Limitations of Large Language Models in Compositional Relation Reasoning

TL;DR

The Multilingual Composition Relation (MCR) benchmark aims at investigating the robustness and adaptability of LLMs in handling composition relation reasoning across diverse linguistic contexts.

Abstract

We present a comprehensive evaluation of large language models(LLMs)' ability to reason about composition relations through a benchmark encompassing 1,500 test cases in English, designed to cover six distinct types of composition relations: Positional, Comparative, Personal, Mathematical, Identity, and Other. Acknowledging the significance of multilingual capabilities, we expanded our assessment to include translations of these cases into Chinese, Japanese, French, and Korean. Our Multilingual Composition Relation (MCR) benchmark aims at investigating the robustness and adaptability of LLMs in handling composition relation reasoning across diverse linguistic contexts.

Figures (7)

• Figure 1: The performance of ChatGPT across the two types of questions.
• Figure 2: Category breakdown accuracy in English zero-shot cot (ZSC).
• Figure 3: Accuracy(%) regarding #Relations across languages in GPT-4 using zero-shot (ZS).
• Figure 4: Category breakdown accuracy, Zero-shot (ZS) prompt
• Figure 5: Category breakdown accuracy, Zero-shot Chain-of-Thought (ZSC) prompt