Abstract
While formal debates follow explicit rules, informal arguments often appear chaotic. This paper presents a computational framework using Augmented Transition Networks (ATNs) to model the underlying structure of spontaneous discourse. By organizing conversation into hierarchical "context spaces," the model explains how participants maintain coherence and resolve complex pronominal references.
Introduction
Everyday arguments between friends, though seemingly spontaneous, are governed by implicit structural rules. Traditional linguistic models often fail to capture the "long-distance" relationships in such discourse—for instance, how a speaker can return to a topic after a lengthy digression. Rachel Reichman proposes the Context Space Theory of Discourse to bridge the gap between surface-level speech and the logical goals driving a debate.
Methods: The ATN Grammar
The core of the proposed model is an Augmented Transition Network (ATN), a graph-based formal grammar. This system segments discourse into functional units called Context Spaces. Each space acts as a "mental folder" containing the current topic, participants, and communicative goals (e.g., a "Support" space or a "Challenge" space).
The ATN operates in three primary modes:
- Organizational: Categorizing incoming messages.
- Constructive: Updating the mental model by shifting focus or creating new context spaces.
- Productive: Generating appropriate conversational responses.
Conversational Moves and Logic
Dialogue is viewed as a sequence of "moves" such as Support, Challenge, or Interrupt. Each move has specific preconditions (what must be true to make the move) and effects (how the move changes the state of the conversation).
Key Insights
Hierarchical Structure
Discourse is hierarchical rather than linear. New sub-arguments are "pushed" onto a stack, and once resolved, the conversation "pops" back to the previous level. This explains the human ability to resume a topic after a digression without losing the logical thread.
Reference Resolution and Pronominalization
A significant breakthrough of this model is its explanation of pronominalization. Pronouns follow logical structure rather than simple recency. In the analyzed "nature vs. nurture" debate, a speaker used the word "it" to refer to a claim made 18 lines earlier. Reichman’s model explains this: because the intervening lines were part of a "support" sub-space that had been "closed," the listener's brain skips back to the last "active" high-level claim in the foreground to find the referent.
Qualitative Analysis
The model was tested against a transcript of a spontaneous conversation between friends debating criminal behavior. The analysis demonstrated that by identifying "clue words" (e.g., "however," "because," "except"), the ATN could accurately track transitions between logical states and resolve ambiguous pronouns that would baffle simpler models.
Conclusion
Reichman’s framework provides a foundation for a "well-formedness" grammar of discourse. While challenges remain in automating the identification of speaker intent and managing computational complexity in long debates, the Context Space Theory offers a robust explanation for the hidden order in informal human communication.
Summary updated