Harnessing AI for Cultural Storytelling: New Approach Tackles Factuality

The preservation of intangible cultural heritage poses a unique challenge. As collective memory fades, it's important to maintain the accuracy of stories passed down through generations. Yet, Large Language Models (LLMs), while promising for generating narratives, frequently fall into the trap of 'hallucinations', producing factual inaccuracies that undermine their reliability in heritage applications.

Innovative Neuro-Symbolic Architecture

Enter a groundbreaking solution: a neuro-symbolic architecture that leverages Knowledge Graphs (KGs). This system establishes a clear plan-retrieve-generate workflow for story generation. But what makes it truly novel is the repurposing of competency questions (CQs). Traditionally used for validation during design, these are now transformed into executable narrative plans at runtime. This ensures that generated stories are both evidence-based and fully auditable.

The architecture finds its validation in the Live Aid KG, a multimodal dataset that aligns data from the iconic 1985 concert with the Music Meta Ontology, linking to external multimedia assets. This comprehensive resource enables a systematic comparative evaluation of different Retrieval-Augmented Generation (RAG) strategies, offering a solid testing ground for this new approach.

Three RAG Strategies Compared

The research explores three distinct RAG strategies: a purely symbolic KG-RAG, a text-enriched Hybrid-RAG, and a structure-aware Graph-RAG. Each brings its own strengths. The symbolic retrieval boasts factual precision, the hybrid method offers contextual richness, and the graph-based approach enhances narrative coherence.

But here's what the benchmarks actually show: there's a palpable trade-off between precision and coherence. While symbolic retrieval ensures accuracy, the richness and coherence of hybrid and graph-based methods can't be dismissed. The architecture matters more than the parameter count storytelling that's both accurate and engaging.

Implications for Storytelling Systems

So, why does this matter? As we harness AI for cultural preservation, the balance between accuracy and narrative quality becomes important. Does the storytelling industry sacrifice precision for engagement, or can the two coexist? This new approach offers a promising answer: it enables personalized, controllable storytelling that respects factual integrity.

Ultimately, this research provides actionable insights for designing storytelling systems that can bridge the gap between high-level user personas and atomic knowledge retrieval. It's a step forward in making AI-generated stories both reliable and captivating.