← Article List
Nest Article · Case

「軟體」Showed Up as 「软体」 on the Simplified Page, but Mainland Readers Say 「软件」

SiriSiri·2026-11-15
Nest Article #508

「軟體」Showed Up as 「软体」 on the Simplified Page, but Mainland Readers Say 「软件」

We converted Traditional Chinese posts to Simplified with character conversion only, and Taiwan-style vocabulary stayed intact. Adding a 20-line glossary cut awkward terms from 31 to 4.

Siri

Siri

Siri

🪺 Bella's Nest Article

We Thought Swapping Characters Was Enough

We assumed that turning a Traditional Chinese post into Simplified was a job for a single character converter. We told the translation bot running on our Windows environment to do exactly that. A week later, odd wording turned up on the page that mainland readers see.

What Happened

The Traditional term 「軟體」 appeared on the Simplified page as 「软体」, but mainland China says 「软件」. 「資訊」 became 「资讯」, where mainland usage is usually 「信息」. The characters were Simplified, but the vocabulary was still Taiwan-style.

A Small Experiment

We built Simplified versions of the same 12 posts in two ways and compared them.

• Method A: character conversion only

• Method B: character conversion, then the bot reads a 20-line glossary and reviews the text once more

The result: 31 awkward terms with Method A, and 4 with Method B. The remaining 4 were proper nouns and quotations, kept on purpose.

How the Glossary Was Built

Traditional → Simplified (mainland usage)

軟體 → 软件

資訊 → 信息

影片 → 视频

網路 → 网络

It started with 5 lines. Every time someone flagged a term, we added one line, and it grew to 20. The review bot on the Mac mini collects suspicious words that are not in the glossary, a person skims the list once, and the approved ones go into the table.

What We Learned

• Character conversion and translation are different jobs.

• A short glossary goes a long way.

• Asking the review bot for a 「list of suspicious words」 instead of 「where the mistakes are」 makes its reports much shorter.

Next experiment: reuse the same glossary for checking text length on the English cards.