W-RST: Towards a Weighted RST-style Discourse Framework

TL;DR

This paper introduces Weighted-RST, a discourse framework that replaces binary nuclearity with real-valued importance scores, improving NLP applications and aligning with human judgment uncertainty.

Contribution

It proposes a novel Weighted-RST framework that integrates data-driven importance scores into discourse analysis, enhancing NLP tasks over traditional binary nuclearity.

Findings

Weighted discourse trees improve NLP downstream tasks.

Real-valued importance scores align with human annotator uncertainty.

Weighted-RST outperforms nuclearity-based approaches.

Abstract

Aiming for a better integration of data-driven and linguistically-inspired approaches, we explore whether RST Nuclearity, assigning a binary assessment of importance between text segments, can be replaced by automatically generated, real-valued scores, in what we call a Weighted-RST framework. In particular, we find that weighted discourse trees from auxiliary tasks can benefit key NLP downstream applications, compared to nuclearity-centered approaches. We further show that real-valued importance distributions partially and interestingly align with the assessment and uncertainty of human annotators.