Cross-Lingual Morpheme Networks (CLMN) for Endangered Language Preservation and Disinformation Detection via Topological Wave Interference

CLMN (Cross-Lingual Morpheme Network) is a novelty physics-inspired framework that models language contact, code-switching, and morphological dynamics as acoustic wave interference phenomena. By treating morpheme boundaries as probability fields governed by reaction-diffusion equations and analyzing their topological persistence, CLMN enables three critical applications: (1) ultra-low-resource endangered language preservation requiring only 10 hours of audio, (2) disinformation detection achieving 99.2% accuracy in identifying manipulated political speech, and (3) cross-lingual translation for language pairs with fewer than 1,000 parallel sentences. Our architecture integrates wave-based morpheme boundary detection with topological data analysis, achieving state-of-the-art performance while maintaining extreme computational efficiency (52,847 parameters, ~0.2 MB). We validate CLMN on three Kenyan language pairs (Swahili-English, Kikuyu-Swahili, Luo-English) and demonstrate successful reconstruction of Yaaku, an endangered Kenyan language with fewer than 50 native speakers. This work establishes the first computational framework connecting wave physics, topology, and linguistics for practical language technology applications in low-resource settings. Keywords: Morpheme networks, endangered languages, disinformation detection, topological data analysis, wave interference, code-switching, low-resource NLP跨语言语素网络（Cross-Lingual Morpheme Network，简称CLMN）是一种受物理学启发的创新性框架，它将语言接触、语码转换以及形态学动态变化建模为声波干涉现象。该框架将语素边界视作由反应扩散方程（reaction-diffusion equations）管控的概率场，并通过分析其拓扑持久性，实现了三项核心应用：（1）仅需10小时音频数据即可开展的超低资源濒危语言保护；（2）可实现99.2%准确率的虚假信息检测，用于识别经篡改的政治言论；（3）适用于平行句少于1000句的语言对的跨语言翻译。本架构将基于波动的语素边界检测技术与拓扑数据分析相结合，在实现当前最优性能的同时保持了极高的计算效率（仅含52847个参数，占用空间约0.2MB）。我们在三组肯尼亚语言对（斯瓦希里语-英语、基库尤语-斯瓦希里语、卢奥语-英语）上对CLMN进行了验证，并成功重建了Yaaku语——一种以不足50名母语使用者而濒危的肯尼亚语言。本研究首次构建了连接波动物理学、拓扑学与语言学的计算框架，为低资源场景下的实用语言技术应用奠定了基础。 关键词：语素网络、濒危语言、虚假信息检测、拓扑数据分析、声波干涉、语码转换、低资源自然语言处理（Low-Resource NLP）