A Language Research Lane

Verdict, in my own words: research about how a language actually behaves now has its own wing, wiki/Language Research/, and the first thing in it is the 音譯 bank — transliteration standards, the bleached character pool, the Taiwan/Mainland splits, and what really decides the tones. I commissioned that research for EP2 and I am not putting it in EP2. It goes into the vault first, whole, with every confidence tag it came with, and the episode takes whatever I decide to pull from it later.

The reason to build a wing rather than drop the note into wiki/Language/ is that the two are answering different questions. The Language wing is a learner’s wing: Refold’s method notes from May, the Outlier character cluster from June, all of it about what to do and in what order. This bank does not tell anyone what to do. It says that 新華社譯名室 was founded on 1956年7月24日 and runs on eight characters — 名從主人，約定俗成 — while Taiwan has reference books and no referee; that Glewwe’s 3,660-syllable corpus puts tone 1 in loanwords at about 29% against 27% in the general lexicon, which is not a flat-tone rule but noise; that the Coca-Cola naming contest never happened, because 蔣彝 was a county magistrate in 九江 in 1930 and 可口可樂 was already in print. That is description, not instruction, and filing it under a learning wing would have buried it.

Why it gets banked instead of used

The material was commissioned to answer one live question for EP2 and it answered several I did not ask. Injecting the useful third into a script and letting the other two thirds evaporate is how research gets paid for twice. The bank keeps the whole pass — the [SOLID] facts, the [COMMON CLAIM, verify] ones I cannot cite yet, the [APOCRYPHAL/CONTESTED] wax-tadpole story that traces to a 1957 cautionary anecdote rather than a 1927 shop sign — and the dead claims are the part I most expect to see resurface. A killed claim written down as killed saves the next search.

What did not come across is the production framing. The guardrail about which character may say what, the notes on what is speakable on mic, the aim at EP2’s three anchor words: those belong to the podcast, and the podcast keeps them.

Two copies, and which one wins

The production copy stays at tsumugu-core/docs/companion/craft/research/tsumugu-podcast-gaming/TRANSLITERATION-BANK.md. That is the citable grounding artifact — the grounded-claims rule in that repo says a script’s claim about a real domain traces to a research artifact, and the artifact has to sit where the script writers are. The vault note is the knowledge home. If the two ever drift apart, the vault wins on what is true and the repo wins on what a script is allowed to cite. Stating the split once, on the vault side, is cheaper than discovering it during a render.

The case against a second wing

The honest argument is duplication. wiki/Language/ already exists, it already has a Chinese cluster with a hub and a condensed page, and a new folder named Language Research sitting beside it invites exactly the confusion I keep saying I do not want — two doors into what a stranger will read as one subject. The Refold system note has been doing hub duty by accident for months, and the disciplined move might have been to give that wing its front door and hang the transliteration note off it.

I am taking the split anyway because the material genuinely divides. A learner’s page recommends and owes the reader a quit signal; a research bank asserts and owes the reader a citation and a confidence tag. Those are different genres with different obligations, and mixing them means every future descriptive note has to pretend to be advice. The cost I am accepting is that the boundary has to be stated on both sides, and today it is stated on one — the new hub says what the learning wing is for and links every note in it in place, but nothing in wiki/Language/ yet points back. That reciprocal line is owed.

The wing has one note in it and a list of open questions longer than its contents: no Taiwan-specific study of 音譯詞聲調 exists that I could find, the claim that 德魯伊 came through Japanese has no primary source, and calques and graphic loans are entirely uncovered with the Hán-Việt bridge as the nearest anchor. That is the shape I want it to have. The bank is not the episode and it was never going to be — it is what the episode gets to draw on, and what the next question about how Chinese absorbs a foreign word starts from instead of a blank search box.