Delimiters in Languages
We propose that languages employ a universal duality in which the same set of symbols can serve two distinct roles. First-order expressions, such as directives or computations intended for immediate execution, contrast with second-order expressions, such as quoted text or stored information that is meant to be represented without being interpreted. This distinction enables complex cognitive processing through the use of delimiters: explicit markers that signal a shift in interpretation between these layers.
| Article | Description | Revisions |
|---|---|---|
| First-Order, Second-Order Expressions, and Delimiters in Languages | This article demonstrates how this structural pattern manifests across diverse symbolic systems: human speech (quotation marks), programming languages (string literals), and genetic code (palindromic DNA sequences that function like quotation marks, isolating viral signatures for immune memory rather than triggering protein synthesis). | July 6, 2024: initial publication |
| LLM Delimiters and Higher-Order Expressions | This article explores the fundamental role of delimiters in enabling LLMs to distinguish between standard text and higher-order expressions, such as system instructions or reasoning processes. By examining these specialized markers across different contexts, we illustrate how particular tokens allow models to transition seamlessly between distinct semantic layers. | February 22, 2025: initial publication |
| Formulaic Delimiters in The Iliad and The Odyssey | In this article, we examine how Homer utilized specific linguistic patterns to signal shifts between narration and character dialogue in the Iliad and the Odyssey. Since ancient oral performances lacked modern punctuation such as quotation marks, we argue that formulaic delimiters served as essential structural cues for the audience. These delimiters fall into three categories: those that open a speech, those that close it, and those that facilitate transitions between speakers. We demonstrate how these recurring expressions provided a predictable framework that enabled listeners to follow complex dialogic exchanges in performance. | December 26, 2025: initial publication |
| Why XML Tags Are so Fundamental to Claude | Claude's designers made a surprising choice: they built their most advanced AI around XML tags, a technology from 1998. But this article argues that the choice may be far more significant than it appears. By making Claude "aware" of delimiters through XML tags, Anthropic may have inadvertently tapped into this deeper principle, and given Claude something that other models, which use delimiters only mechanically and opaquely, lack: a genuine capacity to interpret layered meaning. | March 1, 2025: initial publication / March 2, 2025: updated to address the inference framework v. training ambiguity |