The Odd Characters I Found One Morning
While skimming the Simplified Chinese article, the word 「軟件」 caught my eye. In Simplified it should be 「软件」. I had asked the AI to build the Simplified version from the Traditional one, and in twelve sentences, three spots still held Traditional characters untouched. It is the kind of mistake that almost vanishes when you read with your eyes.
Why It Happens
When the AI reads Traditional text and 「converts」 it, it sometimes leaves familiar characters as they are. Words that differ only in glyphs (軟件, 软件) slip through, and words that differ entirely (滑鼠, 鼠标) slip through even more. Readers in China notice the awkwardness immediately, so this cannot be waved off.
Three Steps You Can Follow
1. Rewrite from the source instead of converting
Tell it: 「Do not convert the zh body into cn. Read the Korean original and write it fresh for Simplified readers.」 Rewriting produced fewer slips than converting.
2. Make a vocabulary table file
Open Notepad, write the lines below, and save it as `terms.txt`. Add one line every time something goes wrong.
Traditional -> Simplified
軟件 -> 软件
滑鼠 -> 鼠标
網路 -> 网络
3. One check line before publishing
Say to Claude Code: 「If any Traditional entry from terms.txt remains in the cn body, list it with line numbers. I will confirm before anything is fixed.」
After Trying It
Once the table was in place, leftovers of this kind almost stopped appearing. The cheapest insurance for multilingual publishing was not a grand translation tool, but a single memo file of the words that went wrong.