Go to ACL ARR 2025 May homepageExploring the Role of Discourse Structure and Tropes in Detecting Enthymemes in Social Media Posts
Abstract: This study investigates the linguistic characteristics signaling enthymemes—arguments with implicit premises or conclusions—in social media texts, focusing on their detection using computational methods. We address two primary research questions: (1) How effective are Rhetorical Structure Theory (RST) discourse features and online tropes in detecting enthymemes? (2) What micro-level rhetorical strategies characterize enthymemes in short texts? We augment a dataset of 313 tweets from the 2019 British electoral campaign, annotated for tropes and enthymeme presence, with automatically generated RST trees. Predictive models, including classical machine learning and transformer-based approaches (e.g., RoBERTa), are trained for enthymeme detection. Findings reveal that RST structural features, such as nucleus-satellite ratio, tree depth, and particular patterns of coherence relations, enhance model capacity to discern enthymeme presence. A rhetorical strategy involving \textsc{joint}, \textsc{background}, and minimal argumentative relations is identified as a key pattern in enthymeme encoding. While certain tropes (e.g., hidden\_motives) correlate with implicit arguments, others are less reliable. Contributions include: (1) a novel dataset annotated for enthymeme presence, (2) an analysis of RST and trope feature efficacy, (3) a signal-based approach to discern enthymeme types, and (4) insights into micro-level discourse-driven rhetorical strategies for enthymeme detection.
Paper Type: Long
Research Area: Discourse and Pragmatics
Research Area Keywords: rhetorical structure theory, rst, implicit communication, tropes, enthymemes detection, persuasion
Contribution Types: Model analysis & interpretability, Data resources, Data analysis
Languages Studied: english
Submission Number: 7360
Loading