Modeling Argumentation and Explanation in the Social Web

Taraneh KhazaeiAAAI 2014 · DOI

Introduction

The Social Web has evolved into a primary venue for collective intelligence, where users collaborate to solve problems, debate policies, and share knowledge. However, as computational tools for "argument mining" become more prevalent, they often overlook a fundamental distinction in human discourse: the difference between argumentation and explanation. This research proposes a computational framework to distinguish these two reasoning types, ensuring that online discourse is analyzed accurately and that valid reasoning is not misidentified as fallacious.

The Reasoning Gap: Argumentation vs. Explanation

The core of the research lies in the epistemic status of the claim being discussed. While the two modes of reasoning may appear linguistically similar, their goals are distinct:

  • Argumentation is used to settle an unsettled issue. Its purpose is to persuade an audience to accept a claim that is currently in doubt or under dispute.
  • Explanation is used to clarify why a fact is true. In this context, the claim is already accepted as a fact by the participants; the goal is to provide understanding rather than proof.

Failing to distinguish between these leads to what the author describes as a "fallacy trap." For instance, circularity is a well-known logical fallacy in argumentation. However, in an explanation, a circular reference might be a valid way to describe a recursive system or a self-evident truth. Without a model that recognizes the intent of the speaker, AI tools risk flagging helpful explanations as "bad" arguments, potentially stifling productive dialogue.

Proposed Methodology

To move beyond simple keyword matching, the author proposes a multi-disciplinary approach to classify discourse based on intent and structure:

  1. Speech Act Theory: By analyzing the "illocutionary force" of statements, the model can determine if a statement is a response to a request for information (typical of explanation) or a challenge to a claim (typical of argumentation).
  2. Rhetorical Structure Theory (RST): This involves mapping how different segments of text relate to one another. The research explores whether specific discourse relations—such as "evidence" versus "justification"—are unique to one reasoning type over the other.
  3. Sentiment Analysis: Argumentation often involves controversial or unsettled topics, which may carry different emotional markers and sentiment intensities compared to the more factual, explanatory mode of discourse.
  4. Iterative Refinement: The framework is designed to be developed first on small-scale, high-quality datasets before being adapted to handle the noise and scale of the broader Social Web.

Experimental Design and Datasets

The research utilizes two primary data sources to test the proposed framework:

  • Lab-Controlled Dataset: Transcribed dialogues from a study where participants worked in groups of five to solve a murder mystery. This provides a "clean" environment where the transition from unsettled claims (argumentation) to established facts (explanation) can be clearly observed and annotated.
  • Wikipedia "Articles for Deletion" (AfD): A large-scale, real-world dataset where community members deliberate on whether to keep or remove specific articles. This dataset offers a rich variety of reasoning styles within a structured, rule-bound environment.

Implications for Collective Intelligence

This research was presented as a Doctoral Consortium paper, focusing on the theoretical framework and experimental design. The primary contribution is the recognition that reasoning is context-dependent; one cannot identify an argument or an explanation in isolation without understanding the participants' intent and the status of the conclusion.

The ultimate goal of this work is to enhance Human-Computer Interaction (HCI). By developing tools that can visualize "reasoning maps" in real-time, the author hopes to increase participant awareness in online debates. Such tools could steer crowds toward more productive problem-solving and prevent the breakdown of civil discourse by clarifying the nature of the reasoning being employed.

Summary updated

BibTeX
@inproceedings{khazaei2014modeling,
  author = {Taraneh Khazaei},
  title = {{Modeling Argumentation and Explanation in the Social Web}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2014},
  doi = {10.1609/aaai.v28i1.8772},
}
Figure