What question did this study set out to answer?

The thesis aims to assess if large language models can effectively collaborate in interpreting historical Chinese philosophical texts.

June 3, 2026Open Access

Cultural Interpretation Through Computational Collaboration With AI: Evaluating Rag-Enhanced LLMS as Research Partners in Song-Ming Moral Philosophy

Key Points

The thesis aims to assess if large language models can effectively collaborate in interpreting historical Chinese philosophical texts.
Employs a Retrieval-Augmented Generation (RAG) methodology to analyze Neo-Confucian texts.
Constructs a specialized knowledge base from primary sources, dynastic histories, and urban studies.
Evaluates seven large language models using a four-dimensional framework.
Models showed significant variation in contextualizing philosophical concepts, with some outperforming others.
AI demonstrated the ability to synthesize spatial-political evidence alongside textual analysis.
Findings highlight a 'conceptual translation problem' in how AI navigates historically embedded concepts.

Abstract

This thesis investigates whether large language models (LLMs) can function as meaningful collaborators in the interpretation of historical Chinese philosophical texts, specifically examining the evolution of moral concepts across the Song (960-1279) and Ming (1368-1644) dynasties. Employing a Retrieval-Augmented Generation (RAG) methodology, I construct a specialized knowledge base from primary sources representing Neo-Confucian rationalist (理學) and idealist (心學) schools, alongside official dynastic histories and urban structure studies. The study focuses on tracking semantic shifts in three cardinal virtues—忠 (zhōng, loyalty), 禮 (lǐ, ritual propriety), and 仁 (rén, benevolence)—to assess how competing philosophical schools and sociocultural transformations influenced moral discourse during this pivotal period. Seven LLMs (Qwen-3:8B, DeepSeek-R1:8B, Gemma-3:12B, Mistral:7B, Claude-Opus-4.5, Claude-Sonnet-4.5, GPT-5.2) are evaluated using a four-dimensional framework measuring: (1) document-grounded reasoning, (2) cultural-temporal competence, (3) interpretive sophistication, and (4) collaborative capacity. Results demonstrate significant variation in LLM capabilities, with models showing differential aptitude for contextualizing philosophical concepts within their historical frameworks. The integration of urban studies materials reveals AI's capacity to synthesize spatial-political evidence with textual analysis, demonstrating how abstract moral concepts became materially instantiated in city planning and architectural regulations. A central finding addresses the "conceptual translation problem"—how AI systems trained on modern frameworks navigate historically-embedded concepts when provided with period-appropriate textual evidence. Rather than positioning LLMs as autonomous interpreters, the thesis conceptualizes them as "cultural participants" capable of mobilizing documentary evidence under human guidance. The findings advance theoretical understanding of machine participation in humanistic inquiry while providing practical methodologies for integrating computational approaches into traditional sinological research.

Cultural Interpretation Through Computational Collaboration With AI: Evaluating Rag-Enhanced LLMS as Research Partners in Song-Ming Moral Philosophy

Key Points

Abstract

Cite This Study