An Empirical Revisiting of Linguistic Knowledge Fusion in Language Understanding Tasks

TL;DR

This paper empirically investigates the role of explicit linguistic priors in language understanding tasks, revealing that trivial graph structures can perform as well as linguistically informed ones, emphasizing the importance of baselines.

Contribution

The study challenges the assumed necessity of linguistic priors by showing trivial graphs can achieve similar performance, urging better baseline design for knowledge fusion methods.

Findings

Trivial graphs perform competitively with linguistically informed graphs.

Performance gains may stem from feature interactions rather than linguistic priors.

Trivial graphs should be used as baselines in future knowledge fusion research.

Abstract

Though linguistic knowledge emerges during large-scale language model pretraining, recent work attempt to explicitly incorporate human-defined linguistic priors into task-specific fine-tuning. Infusing language models with syntactic or semantic knowledge from parsers has shown improvements on many language understanding tasks. To further investigate the effectiveness of structural linguistic priors, we conduct empirical study of replacing parsed graphs or trees with trivial ones (rarely carrying linguistic knowledge e.g., balanced tree) for tasks in the GLUE benchmark. Encoding with trivial graphs achieves competitive or even better performance in fully-supervised and few-shot settings. It reveals that the gains might not be significantly attributed to explicit linguistic priors but rather to more feature interactions brought by fusion layers. Hence we call for attention to using trivial graphs as necessary baselines to design advanced knowledge fusion methods in the future.