Did the Models Understand Documents? Benchmarking Models for Language Understanding in Document-Level Relation Extraction

TL;DR

This paper evaluates whether document-level relation extraction models truly understand documents by comparing their decision rules to human rationales, revealing significant discrepancies and proposing new evaluation metrics for understanding and robustness.

Contribution

It introduces a comprehensive evaluation framework including human rationales, decision rule analysis, and MAP metric to assess model understanding in DocRE.

Findings

SOTA models differ from human decision rules

Models are vulnerable to RE-specific attacks

MAP effectively measures model understanding

Abstract

Document-level relation extraction (DocRE) attracts more research interest recently. While models achieve consistent performance gains in DocRE, their underlying decision rules are still understudied: Do they make the right predictions according to rationales? In this paper, we take the first step toward answering this question and then introduce a new perspective on comprehensively evaluating a model. Specifically, we first conduct annotations to provide the rationales considered by humans in DocRE. Then, we conduct investigations and reveal the fact that: In contrast to humans, the representative state-of-the-art (SOTA) models in DocRE exhibit different decision rules. Through our proposed RE-specific attacks, we next demonstrate that the significant discrepancy in decision rules between models and humans severely damages the robustness of models and renders them inapplicable to real-world RE scenarios. After that, we introduce mean average precision (MAP) to evaluate the understanding and reasoning capabilities of models. According to the extensive experimental results, we finally appeal to future work to consider evaluating both performance and the understanding ability of models for the development of their applications. We make our annotations and code publicly available.