Rhetorical and Discourse Structure Analysis

Overview

Analyze how an author constructs arguments and organizes text beyond the word level: argument ordering (claim-first vs. evidence-first vs. dialectical), rhetorical devices (hedging, concessions, rhetorical questions, analogies), discourse marker usage by functional type, and paragraph/post structural habits (length distributions, list vs. prose ratios, summary positions). The core principle: discourse structure reveals argumentation habits that are more stable across topics than word choice -- a person who leads with evidence before stating conclusions does so whether discussing technology, politics, or cooking. These structural patterns produce replicable constraints for voice matching that complement but do not duplicate lexical or sentiment analysis.

Research foundation: Rhetorical Structure Theory (Mann & Thompson, 1988) established that texts have hierarchical discourse relations between nucleus (core content) and satellite (supporting content) units. The enhanced RST framework (eRST; Zeldes, 2024, Computational Linguistics 51(1)) extends this to discourse relation graphs with non-projective and concurrent relations plus explicit signaling. Argument mining research (Lawrence & Reed, 2019, Computational Linguistics 45(4)) formalized the pipeline of component identification, component classification (claim vs. premise), and structure identification (support/attack relations). Fraser's (1999) taxonomy of discourse markers classifies them by function: contrastive ("but", "however"), elaborative ("moreover", "also"), inferential ("therefore", "thus"), and topic-management ("anyway", "so"). Hedge detection research (Medlock & Briscoe, 2007; Vincze et al., 2008) established lexicon-based and syntactic approaches to identifying epistemic uncertainty markers.

When to Use

• Characterizing how an author orders arguments (do they lead with claims or evidence?)
• Quantifying rhetorical device frequency: hedges, concessions, rhetorical questions, analogies
• Profiling discourse marker usage patterns by functional type (contrastive, elaborative, inferential)
• Analyzing structural habits: paragraph length distributions, list-vs-prose ratios, summary positioning
• Building structural constraints for voice replication that go beyond word choice and sentiment
• Comparing argumentation styles across authors or across time periods for the same author

When NOT to use: