Neural RST-based Evaluation of Discourse Coherence

TL;DR

This paper introduces RST-Recursive, a neural model leveraging RST features for discourse coherence evaluation, achieving state-of-the-art accuracy with fewer parameters on the GCDC benchmark.

Contribution

It presents a novel RST-based neural model that improves coherence classification accuracy and efficiency, outperforming existing methods.

Findings

Achieves new state-of-the-art accuracy on GCDC benchmark.

Uses 62% fewer parameters than previous models.

Ensembling with current SOTA further boosts performance.

Abstract

This paper evaluates the utility of Rhetorical Structure Theory (RST) trees and relations in discourse coherence evaluation. We show that incorporating silver-standard RST features can increase accuracy when classifying coherence. We demonstrate this through our tree-recursive neural model, namely RST-Recursive, which takes advantage of the text's RST features produced by a state of the art RST parser. We evaluate our approach on the Grammarly Corpus for Discourse Coherence (GCDC) and show that when ensembled with the current state of the art, we can achieve the new state of the art accuracy on this benchmark. Furthermore, when deployed alone, RST-Recursive achieves competitive accuracy while having 62% fewer parameters.