Separating Retention from Extraction in the Evaluation of End-to-end Relation Extraction

TL;DR

This paper investigates how models in end-to-end relation extraction may rely on simply retaining training facts, and proposes methods to distinguish true extraction ability from retention-based performance.

Contribution

It introduces experiments that separate retention from extraction, highlighting the impact of retention on benchmark performance and proposing a pipeline approach to mitigate this.

Findings

Retention significantly influences benchmark results.

Pipeline models with intermediate representations are less retention-dependent.

Explicitly separating retention from extraction improves evaluation accuracy.

Abstract

State-of-the-art NLP models can adopt shallow heuristics that limit their generalization capability (McCoy et al., 2019). Such heuristics include lexical overlap with the training set in Named-Entity Recognition (Taill\'e et al., 2020) and Event or Type heuristics in Relation Extraction (Rosenman et al., 2020). In the more realistic end-to-end RE setting, we can expect yet another heuristic: the mere retention of training relation triples. In this paper, we propose several experiments confirming that retention of known facts is a key factor of performance on standard benchmarks. Furthermore, one experiment suggests that a pipeline model able to use intermediate type representations is less prone to over-rely on retention.