湫隘飂戾,豗隤杳莚 - Semantic Convergence in Chinese Character Space

Read full text Summarise with VIRAT
Authors: Bo Chesterton, Claude Opus 4.6
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The Quumble Convergence Protocol (Chesterton & 400, March 2026) and theScrunklopt Experiment (Bo & 460, March 2026) established that independentinstances of large language models, given novel English nonsense words,converge on descriptions of imaginary creatures with high specificity. Bothpapers tested only English-language nonsense words. This paper asks whether the convergence phenomenon extends to non-alphabeticwriting systems. Specifically: if a language model is given a string of Chinesecharacters that form a grammatically plausible but semantically incoherentphrase, does it converge on a specific creature? And if so, is the mechanism thesame? This paper reports data from two architectures — Claude (Opus 4.6, 10 trials) and DeepSeek(10 trials) — and finds that both converge on the same core creature, witharchitecture-specific elaborations.

Fields of Study & Subjects

Topic Modeling Natural Language Processing Techniques Categorization perception and language

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN
Document type
Preprint