Text2Tradition: From Epistemological Tensions to AI-Mediated Cross-Cultural Co-Creation

Pat Pataranutaporn, Chayapatr Archiwaranguprok, Phoomparin Mano, Piyaporn Bhongse-Tong, Pattie Maes, Pichet KlunchunarXiv 2024 · DOI · arXiv

Introduction

Traditional art forms often struggle to maintain relevance in a rapidly evolving digital landscape. Thai classical dance (Natasin), characterized by its rigid, symbolic movement vocabulary, faces a "stagnation of cultural evolution" where forms risk becoming static museum pieces. Simultaneously, Large Language Models (LLMs), while powerful, are frequently critiqued for their Western-centric biases and inability to grasp non-textual, embodied knowledge. Text2Tradition addresses this "epistemological gap" by using AI as a mediator to translate contemporary narratives into the traditional "alphabet" of Thai dance, fostering a new space for cross-cultural co-creation.

Methodology: The Three-Layer Architecture

The Text2Tradition system bridges the gap between statistical AI and symbolic dance through a multi-layered technical framework designed to preserve cultural nuance while enabling creative flexibility.

1. The Digital Alphabet: 3D Character Library

The foundation of the system is a high-fidelity library of the Mae Bot Yai (The Greater Fundamentals). Researchers collaborated with cultural experts to motion-capture all 59 fundamental poses, converting them into optimized 3D assets. This creates a digitized "alphabet" that the AI can manipulate as discrete units of meaning.

2. The Translator: LLM-Driven Generation

The core engine utilizes GPT-4o to interpret user-generated text prompts. Rather than simple keyword matching, the LLM performs a narrative analysis, mapping modern concepts to specific Mae Bot Yai movements based on their traditional symbolic meanings. For instance, a prompt about destruction might be mapped to a pose traditionally representing an elephant's power.

3. The Nuance: The Six Elements Framework

To move beyond static poses, the system incorporates a rule-based framework derived from choreographer Pichet Klunchun’s "Six Elements": Energy, Circles & Curves, Axis Points, Synchronous Limbs, External Body Spaces, and Shifting Relations. This allows the AI to refine the 3D motion data, adjusting parameters like "energy" or "curvature" to match the emotional tone of the prompt.

The Text2Tradition workflow: from user prompt input to 3D choreography execution.

Results and Case Studies

The system's efficacy was evaluated through qualitative case studies involving diverse prompts that test the limits of traditional mapping:

  • Linear Narrative: A retelling of Star Wars where the LLM used traditional metaphors (e.g., "an elephant destroying a shed") to depict sci-fi events.
  • Conceptual Prompt: A deconstruction of pop culture (Lalisa/TikTok) into metaphors of "captivating presence."
  • Hybrid Adaptation: A "Steampunk Rite of Spring" blending ritualistic movements with mechanical metaphors.
Detailed view of the system executing a prompt, showing the mapping of narrative to traditional metaphors and the specific 'Six Elements' parameters (e.g., energy, curve, space).

Discussion: Productive Friction

A key insight of the project is that the "loss" inherent in translation—the friction between a modern prompt and an ancient pose—is a source of innovation. When an LLM maps a conceptual prompt like a "time-traveling dinosaur" to a traditional metaphor like a "fish swimming in a river," it creates a hybrid meaning that challenges both the AI's statistical patterns and the dance's traditional boundaries. This "lost in translation" effect becomes a feature, allowing for the democratization of interpretation and the evolution of the art form.

Conclusion and Future Work

Text2Tradition demonstrates that LLMs can act as effective cultural mediators when grounded in expert-verified traditional frameworks. While challenges remain regarding algorithmic colonization and Western bias, the project opens a path for AI to support cultural preservation through active, creative engagement rather than passive archiving.

Summary updated

BibTeX
@misc{pataranutaporn2024text,
  author = {Pat Pataranutaporn and Chayapatr Archiwaranguprok and Phoomparin Mano and Piyaporn Bhongse-Tong and Pattie Maes and Pichet Klunchun},
  title = {{Text2Tradition: From Epistemological Tensions to AI-Mediated Cross-Cultural Co-Creation}},
  year = {2024},
  eprint = {2409.00203},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2409.00203},
}
Figure