DM Doppelgangers: Implicit Connectives as eRST Signals

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing & Computational Linguistics · Depth: Advanced, quick

Summary

The paper "DM Doppelgangers: Implicit Connectives as eRST Signals" introduces an algorithm to bridge the gap between two major discourse relation frameworks: eRST (Extended Rhetorical Structure Theory) and PDTB (Penn Discourse Treebank). While eRST models hierarchical discourse and identifies explicit connectives like "because" or "although," it lacks mechanisms for implicit connectives, a core feature of PDTB. Conversely, the relationship between PDTB's implicit connectives and eRST's discourse graphs has been largely unexplored. This research proposes and evaluates a novel algorithm designed to align eRST data, which already includes explicit connective annotations, with PDTB-style implicit connective annotations. Furthermore, it presents the first comprehensive evaluation of how hierarchical RST-style relations correlate with PDTB's implicit connectives, advancing understanding of discourse structure representation.

Key takeaway

For NLP Engineers developing advanced discourse parsers, this research highlights a critical gap in eRST's ability to handle implicit connectives, a feature central to PDTB. You should consider integrating mechanisms for implicit discourse relation detection, potentially utilizing the proposed alignment algorithm, to build more comprehensive and robust discourse analysis systems. This work suggests that combining hierarchical RST-style relations with PDTB's implicit connective insights can lead to richer linguistic understanding.

Key insights

An algorithm aligns eRST discourse graphs with PDTB implicit connectives, bridging two major discourse relation frameworks.

Principles

• eRST lacks implicit connective devices.
• PDTB postulates unexpressed connectives.
• Hierarchical RST relations can align with PDTB implicits.

Method

The paper proposes and evaluates an algorithm to align eRST data, which indicates explicit connectives, to implicit connective annotations following PDTB guidelines.

Topics

• Discourse Relations
• eRST
• PDTB
• Implicit Connectives
• Discourse Parsing
• Natural Language Processing

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Discourse Relations and Connectives in Higher Text Structure — Local discourse analysis can reveal clues about higher text structure and global coherence.
• BiCon-Gate: Consistency-Gated De-colloquialisation for Dialogue Fact-Checking — BiCon-Gate improves dialogue fact-checking by consistently de-colloquializing claims with contextual semantic gating.
• A Neural Approach to Discourse Relation Signal Detection — A neural network and \"delta-softmax\" metric quantify discourse signal strength and ambiguity, improving relation identification.
• The Pragmatic Persona: Discovering LLM Persona through Bridging Inference — LLM personas are deeply encoded in discourse coherence via implicit conceptual links, not just surface-level lexical patterns.
• Relational Rank Geometry in Transformers: Detecting and Steering Hidden-State Relation Frames — Transformers encode multi-argument relations as detectable and steerable rank-indexed hidden-state geometries.
• Revisiting Semantic Role Labeling: Efficient Structured Inference with Dependency-Informed Analysis — A new SRL framework offers explicit predicate-argument structure and 10x faster inference with modern encoders.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.