Eristic Dialogue

A Walton-Krabbe dialogue type whose shared goal is to vent emotion or display superiority — the dialogue type in which “winning” is the participant goal and there is no shared epistemic or practical aim. Pre-modern rhetoric and many social-media exchanges fit this type. In MAS the relevant question is not how to engage in eristic dialogues but how to detect and exit them — a multi-agent system inadvertently conducting eristic dialogues at scale is a MAST-style failure.

In this vault

Backlinks

Linked Pages

Dialogue Typology

Walton & Krabbe’s (1995) classification of argumentative dialogue into six basic types: persuasion (resolve a conflict of opinion), negotiation (resolve a conflict of interest), deliberation (decide on a course of action), inquiry (prove a hypothesis), information-seeking (transfer of information from informed to uninformed), eristic (vent emotion). Each is characterised by initial situation × individual goals × shared goal. Real argumentation often shifts between types — licit shifts broaden the dialogue, illicit ones constitute fallacies.

In this vault

Commitment in Dialogue

Commitment in Dialogue: Basic Concepts of Interpersonal Reasoning

Reference: Walton, D. N. & Krabbe, E. C. W. (1995). Commitment in Dialogue: Basic Concepts of Interpersonal Reasoning. SUNY Series in Logic and Language. State University of New York Press, Albany. Internet Archive borrow (book; institutional access via SUNY.)

Summary

Walton and Krabbe synthesise two decades of work on formal dialectic (Hamblin 1970 onward) and informal logic into a typology of dialogue and a corresponding theory of how participants’ commitments evolve as a dialogue proceeds. Their central observation is that argumentation in natural settings is not a single uniform activity but a structured family of dialogue types, each characterised by its initial situation, the individual goals of the participants, and the shared goal of the dialogue. The six basic types they identify — persuasion, negotiation, deliberation, inquiry, information-seeking, eristic — are mutually distinguished by these three dimensions and by their characteristic legitimate moves. Across all types, every utterance updates each participant’s commitment store: the public set of propositions to which they have committed, are committed by inference, and may be challenged on. The book gives semi-formal rules for each dialogue type, analyses dialogue shifts (moves that licitly or illicitly transition between dialogue types — e.g. a persuasion shifting to a negotiation), and uses the framework to give principled accounts of fallacies as illegitimate dialogue moves rather than logical errors. Its influence on multi-agent systems is enormous: every commitment-based ACL semantics (Singh, Fornara & Colombetti) inherits the Hamblin/Walton commitment-store concept; protocol-design work (e.g. ACRE, An Interaction-oriented Agent Framework for Open Environments) uses dialogue-type analysis to specify legitimate conversation policies.

Key Ideas

Six basic dialogue types: persuasion (resolve a conflict of opinion), negotiation (resolve a conflict of interest), deliberation (decide on a course of action), inquiry (establish proof of a hypothesis from agreed premises), information-seeking (transfer of information from informed to uninformed), eristic (vent emotion / display superiority). Each is characterised by initial situation × individual goals × shared goal.
Commitment store as the engine of dialogue: every participant has a public set of commitments updated by their moves; speech acts add, retract, or qualify entries. The store is open to challenge by other participants under the dialogue’s rules.
Dialectical / propositional commitments: Walton & Krabbe distinguish explicit commitments (asserted propositions) from inferred commitments (entailed by, or presupposed by, what was said) — a distinction crucial for capturing implicit communication.
Dialogue shifts: most real argumentation moves between dialogue types. Some shifts are licit (a persuasion may legitimately broaden into a deliberation); others are illicit (an information-seeking dialogue exploited as a covert persuasion is a fallacy of digression).
Fallacies as dialogue-rule violations: rather than treating fallacies as defects of single arguments (the syllogistic tradition), they are analysed as moves that violate the rules of the dialogue type they occur in — a circular argument is fine in inquiry but fallacious in persuasion.
Profiles of dialogue: a sequence diagram of typical turns within a dialogue type — the precursor of Conversation Policy / interaction-protocol diagrams in the ACL literature.

Connections

Conceptual Contribution

Claim: Argumentation is not a single uniform activity but a structured family of dialogue types characterised by the participants’ goals; within each type, every utterance updates each participant’s public commitment store under type-specific rules; fallacies are best analysed as illegitimate dialogue moves rather than as defective arguments.
Mechanism: Six basic dialogue types defined by initial situation × individual goals × shared goal; dialogue rules per type; commitment-store update semantics; analysis of dialogue shifts and mixed dialogues; classification of fallacies as type-specific rule violations.
Concepts introduced/used: Dialogue Typology, Commitment Store, Persuasion Dialogue, Deliberation Dialogue, Inquiry Dialogue, Information-Seeking Dialogue, Eristic Dialogue, Negotiation, Dialogue Shift, Profile of Dialogue.
Stance: synthesis monograph / theoretical framework.
Relates to: Provides the upstream typology that the commitment-based ACL programme (Singh, Fornara & Colombetti, Commitment Machines - Yolum and Singh) takes as given when it makes commitments first-class. The Hamblin commitment-store mechanism (Fallacies - Hamblin) is the operational heart; Walton & Krabbe extend it with the dialogue-type framing. Conceptually adjacent to Grice’s cooperative principle but more granular: where Grice supplies a single cooperative norm, Walton & Krabbe show that the relevant cooperation is type-specific (persuasion ≠ negotiation ≠ inquiry). Coordinating Agents Using ACL Conversations (Cost et al.) and ACRE Agent Conversation Reasoning Engine are direct engineering descendants — Coloured Petri Nets and Dooley graphs are formal specifications of the kinds of dialogue Walton & Krabbe describe semi-formally. In the LLM-agent era the dialogue-type taxonomy is the obvious diagnostic for MAST-style failures (e.g. multi-agent systems unwittingly conducting eristic dialogues at scale, or persuasion-shifts hijacking deliberations) and a natural specification language for conversation-policy enforcement.

Why Do Multi-Agent LLM Systems Fail?

Reference: Mert Cemri, Melissa Z. Pan, Shuyi Yang, Lakshya A. Agrawal, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Dan Klein, Kannan Ramchandran, Matei Zaharia, Joseph E. Gonzalez, Ion Stoica (2025). arXiv:2503.13657v2 (UC Berkeley). Source file: 2503.13657v2.pdf. URL

Summary

First empirically grounded taxonomy of failure modes in Multi-Agent LLM Systems (MAS). The authors analyse 200+ execution traces from seven popular MAS frameworks (MetaGPT, ChatDev, HyperAgent, AppWorld, AG2, Magentic-One, OpenManus), annotated by six human experts via grounded theory and reaching Cohen’s κ ≈ 0.88, and distil 14 fine-grained failure modes grouped into three categories: Specification Issues (42%), Inter-Agent Misalignment (37%), and Task Verification (21%).

They release MAST (Multi-Agent System failure Taxonomy), a validated LLM-as-judge pipeline for automated failure diagnosis, and two intervention case studies showing that architectural/prompt fixes inspired by MAST improve success rates modestly — demonstrating that MAS failures are system-design problems, not merely model-capability problems.

Key Ideas

Three failure categories: specification, inter-agent misalignment, verification
14 fine-grained failure modes including step repetition, information withholding, task derailment
Grounded-theory methodology with rigorous inter-annotator agreement (κ=0.88)
LLM-as-judge pipeline (MAST) achieves κ=0.77 vs humans for scalable evaluation
Insight: better specifications and verification beat bigger models

Connections

Conceptual Contribution

Claim: Multi-Agent LLM System (MAS) failures are predominantly system-design problems — specification, coordination, and verification — rather than base-model capability problems; and these failures have an empirically discoverable, reproducible taxonomy.
Mechanism: Grounded-theory analysis of 200+ execution traces across seven MAS frameworks (MetaGPT, ChatDev, HyperAgent, AppWorld, AG2, Magentic-One, OpenManus) with six human expert annotators; iterative refinement to Cohen’s κ≈0.88; yield the 14-mode MAST taxonomy grouped into Specification Issues (42%), Inter-Agent Misalignment (37%), Task Verification (21%); validate an LLM-as-judge annotator (κ≈0.77); intervention case studies showing prompt/architecture fixes provide only modest gains, motivating deeper redesign.
Concepts introduced/used: MAST Taxonomy, Grounded Theory, Inter-Agent Misalignment, LLM-as-judge, Specification Issues, Task Verification, Multi-Agent Systems, LLM Agents, Cohen’s Kappa, Standard Operating Procedures (SOPs)
Stance: empirical / evaluative
Relates to: Supplies empirical grounding for the design-quality concerns in Agents Framework - Zhou et al (SOPs attempt to mitigate FC1 specification issues) and Multi-Agent Collaboration in AI - Wasif Tunkel. Inter-agent misalignment mirrors the formal pathologies in Are Multiagent Systems Resilient to Communication Failures. Motivates richer communication protocols like A Scalable Communication Protocol for Networks of LLMs and commitment-style ACLs (ACL Rethinking Principles).