Rhetoric-Based Strategies in Persuasion

• Rhetoric-Based Strategies are systematic methods for organizing and delivering persuasive discourse using formal frameworks, dualities, and quantified measures.
• They integrate classical appeals with computational annotation techniques, enabling dynamic analysis and precise modeling of argumentation and persuasion.
• Key applications span persuasive dialogue, explainable AI, and curriculum design, with ethical implications for mitigating manipulative uses in media.

Rhetoric-based strategies are systematic methods for shaping, organizing, and delivering discourse with the explicit aim of persuasion, influence, or communicative effectiveness. These strategies are grounded in frameworks that formalize rhetorical modes, argument structures, and stylistic operations, and are increasingly instantiated computationally in natural language processing, argument mining, and artificial intelligence systems. Their use spans text, speech, visual media, and interactive platforms, underpinned by a rich set of dualities, taxonomies, and formal methods for reasoning about, generating, and evaluating persuasive communication.

1. Formal Typologies and Dualities of Rhetorical Strategies

Rhetoric-based strategies are conceptualized via typologies that classify the surface and functional properties of discourse. One major system is the duality-based mode operations framework, which defines four key operations over a set of atomic rhetorical modes (e.g., Narration, Argumentation, Exposition):

• Split–Unite Duality: Decomposes compound (diatomic) rhetorical modes into atomic constituents (e.g., "Comparison-Contrast" → "Comparison" + "Contrast") and re-composes them.
• Forward–Backward Duality: Reverses logical or temporal directions (e.g., “Cause→Effect” vs. “Effect→Cause”, “Deduction↔Induction”).
• Expansion–Reduction Duality: Adjusts the granularity of exposition (e.g., expanding a definition to a survey, reducing exposition to a summary).
• Orthogonal Duality: Pairs modes that operate in independent cognitive dimensions (e.g., "Narration" vs. "Description").

Combinatorics on these operations yield a large expressivity space (e.g., with K atomic modes, there are $\binom{K}{2}$ diatomic compounds and $2^K - 1$ non-empty subsets) (Wu, 10 Nov 2025).

Further, strategies can be classified into function-oriented types:

Type	Definition	Example
Causal	Cause-effect, predictive reasoning	"Mandatory vaccination could result in rich countries hoarding vaccines, making them inaccessible"
Empirical	Reliance on concrete evidence (statistics, examples, citations)	"Research has estimated that many death row inmates were wrongly convicted"
Emotional	Linguistic techniques aimed at arousing or expressing emotion	"Imagine a deer, wasting away, its ribs showing through…"
Moral	Appeals to duty, virtue, rights, justice	"It is a duty of the state to protect its citizens from disease"

This typology bridges classical Aristotelian appeals and contemporary social-psychological persuasion theory (Ji et al., 16 Oct 2025).

2. Rhetorical Strategies in Argumentation, Discourse, and Persuasion

Rhetoric-based strategies manifest as distinct tactics in argumentation frameworks, persuasion dialogues, and computational annotation schemes.

• Sentence-level strategies in loan requests can be classified as Concreteness, Reciprocity, Impact, Credibility, Politeness, or Other. The effectiveness of these strategies depends not only on their presence but on their sequencing—e.g., politeness at the close of a request (Po–Po–EOS) is associated with high success, while overuse of Concreteness (Co–Co–Co) correlates with low effectiveness (Shaikh et al., 2020).
• Debate and deliberation rely on structured appeals to ethos (credibility), pathos (emotion), and logos (logic). Operationalization involves multi-criteria scoring rubrics for evidence quality and causal reasoning explicitness (Wu et al., 14 Dec 2025).
• Counter-speech in persuasion: Effective counter-arguments against hate speech incorporate Walton argumentation schemes (e.g., Means-for-Goal, From Consequence), speech acts (Denouncing, Facts, Hypocrisy, Praise, Questioning), and human-centric qualities (Big Five personality, moral values). Experimentally, combining argument-schemes with personality models increases both BLEU and ROUGE performance and argumentative effectiveness (Saha et al., 2024).
• Dynamic argument frameworks: In abstract argumentation, persuasion acts—modeled via relations $R_p$—modify visibility and status of arguments dynamically, with defense, attack, and persuasion all modeled via transition logic (CTL). This allows formal verification of the conditions under which a persuasion strategy succeeds or fails, and how collaborative or adversarial dynamics play out across multi-agent discourse (Arisaka et al., 2017).

3. Computational Annotation and Detection of Rhetorical Strategies

Advances in annotation and detection have enabled scalable analysis and modeling of rhetorical strategies.

• Automated rhetoric annotation: Large LLM-based frameworks automatically generate and annotate synthetic debate datasets labeling utterances on the causal, empirical, emotional, and moral dimensions. Continuous [0,1] scoring enables fine-grained regression modeling with high domain and cross-domain generalization (Ji et al., 16 Oct 2025).
• Rhetorical figures and ontologies: The GRhOOT ontology structures 110 German rhetorical figures as formal classes with compositional relations for operation type (e.g., repetition, inversion), affected element, and operation form. The "Find your Figure" application grounds LLM retrieval-augmented generation (RAG) over this ontology, improving annotation quality, limiting hallucination, and creating reusable, provenance-annotated resources for downstream NLP (Kühn et al., 2024).
• Epistemic stance detection: Rhetorical strategies in political texts are robustly modeled via epistemic stance (asserted, denied, hedged, attributed, or underspecified), with models such as RoBERTa-large achieving macro-F1 ≈ 77.6% for clause-level multi-source stance annotation, enabling large-scale, ideology-sensitive analysis of belief holder citation and commitment (Gupta et al., 2022).
• Function-aware paraphrase analysis: Multi-agent LLM systems grounded in argument theory dramatically improve the detection of D-I-S-G-O rephrase functions (Deintensification, Intensification, Specification, Generalisation, Other), with retrieval-augmented agents more than doubling macro F1 score in detecting rhetorical reformulation strategies in political debates (Uberna et al., 14 Feb 2026).

4. Rhetoric in Explainable AI, Visualization, and Scientific Writing

Rhetoric-based strategies increasingly structure explanation, communication, and user trust in computational systems.

• Explainable AI (XAI): Rhetorical design principles organize explanation types (feature, example, rule-based, counterfactual, concept) around logical reasoning (logos), credibility (ethos), and emotional resonance (pathos). Explanations as recommenders ("what next?"), authority framing, numerical association, and expectation management are all mapped to rhetorical appeals, with empirical and design trade-offs analyzed (Liu et al., 14 May 2025).
• Visualization Rhetoric: Five families—information access, provenance, mapping, linguistic, procedural—capture the means by which visualizations persuade. LLMs remain less sensitive than human experts in correctly attributing these rhetorical moves, but frameworks now enable explicit scoring of models’ error sensitivity (ESS) and behavioral similarity (MBS) to expert annotation. Guidelines recommend explicit provenance reporting, zero-based axes, neutral linguistic framing, and avoidance of default interactions that steer narrative (Blasilli et al., 1 Apr 2026, Lin et al., 16 Jul 2025).
• Scientific Style Quantification: Counterfactual LLM-based frameworks disentangle rhetorical style from substantive content by generating persona-driven rewrites and scoring via Bradley–Terry aggregation. Rhetorical strength (sensationalized/promotionally bold language) is shown to be both robustly measurable and predictive of downstream citations and media attention, independent of peer-review scores (Qiu et al., 22 Dec 2025).

5. Quantitative and Algorithmic Measures for Strategic Rhetoric

The operationalization of rhetoric-based strategies involves combinatorial, information-theoretic, and statistical tools.

• Expressive diversity: Binomial combinatorics ($C(K,k)$) and total rhetorical capacity ($K_{RC}(K) = \log_2(2^K-1) \approx K$ bits) quantify the configuration space of rhetorical modes and strategies.
• Entropy reduction via layering: Mapping from rhetorical ($R$) to cognitive ($C$) to epistemic ($E$) layers via a pyramid multilayer framework reduces cognitive and search entropy from $H_{\mathrm{flat}} = \log_2(K_R)$ in flat selection to $H_{\mathrm{hier}} = H_E + H_{C|E} + H_{R|C}$, with empirical reductions by an order of magnitude (Wu, 10 Nov 2025).
• Marginal Rhetorical Bit (MRB): Each newly introduced mode increases expressive space by one bit, formalizing the scaling of rhetorical expressiveness.

These measures support layered planning in NLG and policy design, curriculum sequencing in pedagogy, and dynamic adaptation of AI-generated discourse to match user expertise or communicative intent.

6. Applications and Implications for System Design and Analysis

Rhetoric-based strategies inform both system design and analytical pipelines:

• Pedagogy and Discourse Design: Rhetorical mode planning supports sequenced curriculum and lesson design, scaffolding for rhetorical maturity, and explicit separation of content from style (Wu, 10 Nov 2025, Wu et al., 14 Dec 2025).
• Persuasion and Counter-Persuasion: In automated dialogue, explicit representation of argument schemes, speech acts, and personality traits or values enables the conditioning of generated responses to mitigate hate, promote engagement, or shift viewpoints (Saha et al., 2024).
• Robust Model Evaluation: In XAI and visual analytics, rhetorical metrics track whether explanations or visual encodings persuade, inform, mislead, or manipulate, with quantitative diagnostics enabling benchmarking and improvement (Liu et al., 14 May 2025, Blasilli et al., 1 Apr 2026).
• Sociotechnical and Ethical Considerations: Rhetoric-based strategies can be exploited for manipulation as a service (MaaS), necessitating transparency, contestability, cultural adaptation, and ethical meta-controls over rhetorical system outputs (Williamson, 8 Apr 2026).

The landscape of rhetoric-based strategy research demonstrates a progressive formalization and operationalization of persuasion across modalities and domains, equipped with quantitative metrics, computational models, and validated annotation schemes. These advances enable more principled, measurable, and adaptable design of both communicative agents and analytic methodologies.