SR-LLM: Rethinking the Structured Representation in Large Language Model

TL;DR

This paper introduces SR-LLM, a novel framework that effectively integrates structured representations into large language models through natural language prompts and fine-tuning, leading to significant performance improvements.

Contribution

It presents the first demonstration that structured representations can substantially enhance LLM inference by using natural language descriptions and fine-tuning methods.

Findings

Performance improvements on downstream datasets

Notable gains of 3.17% and 12.38% in PAWS

First to show structural data enhances LLM reasoning

Abstract

Structured representations, exemplified by Abstract Meaning Representation (AMR), have long been pivotal in computational linguistics. However, their role remains ambiguous in the Large Language Models (LLMs) era. Initial attempts to integrate structured representation into LLMs via a zero-shot setting yielded inferior performance. We hypothesize that such a decline stems from the structure information being passed into LLMs in a code format unfamiliar to LLMs' training corpora. Consequently, we propose SR-LLM, an innovative framework with two settings to explore a superior way of integrating structured representation with LLMs from training-free and training-dependent perspectives. The former integrates structural information through natural language descriptions in LLM prompts, whereas its counterpart augments the model's inference capability through fine-tuning on linguistically described structured representations. Performance improvements were observed in widely downstream datasets, with particularly notable gains of 3.17% and 12.38% in PAWS. To the best of our knowledge, this work represents the pioneering demonstration that leveraging structural representations can substantially enhance LLMs' inference capability. We hope that our work sheds light and encourages future research to enhance the reasoning and interoperability of LLMs by structure data.