数学有没有母语？

TL;DR — A long-form essay arguing that no natural language is inherently better or worse for mathematics: all mathematical traditions independently "escaped" natural language into symbols, and what matters is the deliberately constructed register (语域), not the language itself. It matters as a grounded critique of the popular claim that Chinese suits poetry while English suits science.

• Convergent escape: al-Khwārizmī (Arabic, VSO inflectional), Li Ye/Zhu Shijie (Chinese, isolating), and Seki Takakazu (Japanese, SOV agglutinative) all abandoned prose for symbolic notation — analogous to convergent evolution, implying environmental pressure rather than lineage.
• Only three verified language effects: regular number words (Miura et al. 1988, but erased by 3 minutes of instruction per Saxton & Towse 1998; Welsh comparison by Dowker shows reading/comparison gains, not arithmetic), phonological-loop capacity (Ellis & Hennelly 1980 — affects digit span/mental buffering only, nullified once you use paper), and terminology transparency (explicitly flagged as untested conjecture).
• Syntax argument: math's logical load sits on quantifier sequences (∀∃∀ — where Chinese markers 使得/当…时/有 delimit scope cleanly) and numbered references, not nested relative clauses; Chinese left-branching cost effectively forces A-normal-form writing, which Bourbaki did deliberately. Within-language variance (Wen Tingyun vs. Liu Hui's 263 CE commentary on the Nine Chapters) swamps between-language variance.
• Register is built, not inherited: scholastic Latin, the Royal Society's anti-rhetoric program (Sprat 1667), Meiji Japanese coinage, and Xu Guangqi/Li Shanlan's translations all engineered math registers. German (peak prestige) and Russian (closed market) registers collapsed; French survived only because EGA/SGA remain untranslated. Two recent Chinese works — Li Wenwei's 《代数学方法》 and Yu Pin's 1001-page 《数学分析之课程讲义》 (which mandates writing proofs in Chinese) — are offered as opposing but complementary attempts to build the missing graduate-level expository register.