GraSAME: Injecting Token-Level Structural Information to Pretrained Language Models via Graph-guided Self-Attention Mechanism

TL;DR

GraSAME introduces a novel, efficient self-attention mechanism that seamlessly integrates token-level graph structural information into pretrained language models, improving graph-to-text generation without extra pre-training or increased complexity.

Contribution

The paper presents GraSAME, a lightweight, end-to-end module that effectively bridges graph and text modalities within PLMs using a novel graph-guided self-attention mechanism.

Findings

Outperforms baseline models on WebNLG dataset.

Achieves results comparable to state-of-the-art models.

Reduces trainable parameters by over 100 million.

Abstract

Pretrained Language Models (PLMs) benefit from external knowledge stored in graph structures for various downstream tasks. However, bridging the modality gap between graph structures and text remains a significant challenge. Traditional methods like linearizing graphs for PLMs lose vital graph connectivity, whereas Graph Neural Networks (GNNs) require cumbersome processes for integration into PLMs. In this work, we propose a novel graph-guided self-attention mechanism, GraSAME. GraSAME seamlessly incorporates token-level structural information into PLMs without necessitating additional alignment or concatenation efforts. As an end-to-end, lightweight multimodal module, GraSAME follows a multi-task learning strategy and effectively bridges the gap between graph and textual modalities, facilitating dynamic interactions between GNNs and PLMs. Our experiments on the graph-to-text generation task demonstrate that GraSAME outperforms baseline models and achieves results comparable to state-of-the-art (SOTA) models on WebNLG datasets. Furthermore, compared to SOTA models, GraSAME eliminates the need for extra pre-training tasks to adjust graph inputs and reduces the number of trainable parameters by over 100 million.