Introduction
Human conversation is rarely a random sequence of words. Instead, it often follows invisible "scripts" or discourse routines—predictable patterns of interaction that help participants navigate social goals. For instance, in decision-making contexts, a question typically prompts an answer, and a suggestion often leads to an agreement or a counter-proposal. While these patterns are intuitively understood, quantifying the degree to which a conversation adheres to these routines has historically been a challenge.
This research introduces a quantitative method using Markov models to measure conformity to discourse routines. By comparing face-to-face (FtF) interactions with computer-mediated communication (CMC), the study explores how the medium of exchange and task complexity influence our adherence to conversational "ruts."
Methodology: The Markovian Framework
To measure discourse management strategies, the researchers developed a multi-step analytical framework centered on utterance coding and sequence analysis.
Utterance Coding
Every sentence within a dialogue was manually categorized into functional groups:
- Moves: Actions such as making a suggestion (SA) or requesting information.
- Responses: Reactions like agreeing (AS) or disagreeing (DS) with a previous move.
- Other: Categories including metalanguage (talking about the conversation itself) or social pleasantries.
The Decision Routine Model
The study defined a "standard" decision-making routine as a three-step path: Orientation $\rightarrow$ Suggestion $\rightarrow$ Agreement. This model serves as the theoretical benchmark for measuring conformity.
Markov Analysis
To quantify how strictly participants followed this routine, the researchers employed Markov chain analysis at three levels:
- 0-order: Analyzing the frequency of individual functional categories (e.g., the total proportion of suggestions).
- 1st-order: Examining the frequency of pairs (e.g., how often a suggestion immediately follows an orientation).
- 2nd-order: Measuring the frequency of triples (e.g., the full O-S-A sequence).
By calculating the probability of these sequences occurring in actual talk versus a random distribution, the authors derived a "conformity score" for different communication environments.
Data and Experimental Design
The analysis utilized two primary corpora of dyadic (two-person) interactions:
- The Original Corpus: 32 interactions involving simple planning tasks, such as organizing a weekend getaway.
- The MTV Corpus: 68 interactions involving a complex task: planning a televised music video awards ceremony.
Both datasets included variations in communication medium (FtF vs. CMC). In the MTV corpus, the researchers also manipulated the "message-window size" in the digital interface to determine if visual constraints impacted discourse structure.
Key Findings
The study revealed several critical insights into how humans adapt their communication strategies to their environment.
Digital Constraints and Efficiency
The most significant finding was that participants in CMC environments conformed to discourse routines much more strictly than those in FtF settings. In the complex MTV task, the proportion of 2nd-order sequences following the ideal decision routine was 0.12 to 0.17 for CMC groups, compared to just 0.04 for FtF groups.
This suggests that the limitations of digital media—such as slower typing speeds and the absence of non-verbal cues—force users to rely on "predictable defaults." By sticking to a rigid script, participants minimize cognitive load and maximize the efficiency of the interaction.
Task Complexity and Metalanguage
As tasks became more complex, the nature of the "Other" category shifted. Participants engaged in significantly more metalanguage, or "talk about talk." Rather than jumping straight to suggestions, they spent more time organizing the structure of the conversation (e.g., "Let's decide the host first").
Resilience of the Routine
Interestingly, while CMC participants sometimes used longer turns to address multiple points at once (parallel processing), the underlying Markovian conformity to the decision routine remained high. Furthermore, the size of the text window (ranging from 4 to 18 lines) had a negligible effect on the functional proportions of the discourse, suggesting that the drive toward routine is a fundamental adaptation to the medium itself rather than specific interface constraints.
Conclusion
This research demonstrates that Markovian models provide a robust, stable metric for measuring the "scriptedness" of human interaction. The finding that digital environments increase conformity to routines has significant implications for the design of collaborative software and natural language processing systems. By understanding these predictable patterns, developers can create tools that better support the natural—and the digitally adapted—flow of human decision-making.
Future work aims to automate the labor-intensive coding process using machine learning and to apply these measures to other interaction types, such as negotiation and pedagogical dialogue.
Summary updated