More Than a Dictionary: How AI Scaffolds the Journey from Digital Outsider to Insider

Paper Title

More Than a Dictionary: How AI Scaffolds the Journey from Digital Outsider to Insider

Publication Info

• Topic area: AI-mediated cultural integration in online communities
• Keywords: cultural learning, AI scaffolding, Chain-of-Thought, Retrieval-Augmented Generation, digital outsiders, absurd language, Chinese social media, socio-technical systems, cultural mediation, distributed cognition

Background and Problem

• Problem / challenge: Online communities develop symbolic vocabularies that marginalize newcomers, and existing AI systems fail to scaffold cultural integration beyond translation.
• Significance: Effective cultural integration enables meaningful participation in online communities, fostering social inclusion and reducing barriers for outsiders.
• Motivation and related work: Prior work on cultural mediation and distributed cognition highlights the importance of socio-technical systems in bridging interpretive gaps. However, AI systems often struggle with implicit cultural norms and fail to support dynamic, situated learning processes.

Solution

• Proposed approach: An AI mediator integrating Chain-of-Thought (CoT) reasoning and Retrieval-Augmented Generation (RAG) to scaffold cultural integration in Chinese social media.
• Novelty:
  1. A five-stage model of AI-mediated cultural integration derived from user interaction.
  2. A conversational system combining CoT and RAG to illuminate cognitive and social dynamics in cultural learning.
  3. Design principles for socio-technical systems that support cultural integration rather than simple information retrieval.
• Procedure and key techniques:
  • Curated a 120,000-item corpus of "absurd language" from Chinese social media platforms.
  • CoT module deconstructs cultural texts into stylistic features, inferred contexts, and motivations.
  • RAG module retrieves authentic examples from the corpus to ground explanations.
  • Conducted a mixed-methods user study comparing the system to a baseline large language model.

Results

• Concrete findings:
  • System B (CoT+RAG) achieved 93.37% comprehension accuracy, outperforming System A (baseline) by 11.95 percentage points.
  • User ratings for System B were significantly higher across Text Understanding (p = 0.007), System Proficiency (p = 0.021), and AI Understanding (p = 0.045).
• Advantage over baselines: System B provided structured scaffolding, enabling better comprehension and cultural integration, especially for digital outsiders.
• Experiments / evaluation:
  • Within-subjects design with 14 participants completing six comprehension tasks.
  • Quantitative measures included task accuracy and Likert-scale ratings; qualitative analysis traced a five-stage cultural integration process.
• Limitations and future work: Small sample size (n = 14), focus on a single cultural phenomenon, and lab-based setting. Future work should explore longitudinal studies, platform-specific affordances, and AI’s role in co-creative processes.

Summary

This study introduces a five-stage model of AI-mediated cultural integration, charting users' progression from peripheral observation to confident participation in Chinese social media communities. By integrating Chain-of-Thought reasoning and Retrieval-Augmented Generation, the proposed system significantly improves comprehension accuracy and user confidence compared to a baseline model. The findings highlight the importance of adaptive scaffolding, dynamic agency, and trust in human-AI partnerships. Future systems should focus on supporting creative participation and ethical representation of cultural practices, positioning AI as a socio-technical catalyst for community integration.