Contextual Structure Expression: A Unified Modeling Framework for Semantic Contextual Markup
Liwei Zhao · 2025
With the development of large language models (LLMs), knowledge has gradually replaced information as the core object of processing, and the focus of structured expression is quietly shifting. Traditional structured expressions, represented by JSON, YAML, RDF, etc., are limited by static structures and deterministic assumptions, making them difficult to meet the needs of a semantics-driven era. In intelligent systems, the actual semantics often depend on changes in contextual background. The preset schema approach becomes extremely rigid when dealing with dynamically changing contexts. Moreover, because LLMs are better at processing natural language, using JSON, YAML, or RDF to exchange semantics leads to frequent conversions between these formats and natural language. During the linearization process, a certain degree of semantic loss occurs, ultimately leading to contextual inconsistencies and various issues in multimodal collaboration, dynamic context binding, and explainability. This is fundamentally a generational disparity issue, as these formats were created for information exchange rather than for expressing semantic relationships. To address this, this paper proposes a minimalist expression paradigm for semantic modeling-Contextual Structure Expression (CSE), and based on this paradigm, designs Context Mark Language (CML), which prioritizes semantic structure expression and can be embedded within natural language. It demonstrates significant advantages in the marking, transmission, storage, sharing, and computation of context, showing vast potential in the standardization and generalization of semantic relationship expression.