When characters were simplified in the People's Republic of China, some Traditional characters that used to be distinct were merged into one Simplified shape. Learners meet this as a writing convenience: fewer forms to memorize. Software meets it as a fork in the road. Going back to Traditional Chinese requires a choice.
That choice is the main reason people say "the converter got my name wrong" even when the rest of the paragraph looks fine. The engine is not randomly dropping strokes. It is trying to undo a merger with only a few nearby words as evidence.
Why mergers happened
The mid-twentieth-century reforms had several methods: fewer strokes, popular handwritten shapes, and the elimination of variants that teachers treated as extras. Merging two rare or related characters into one classroom form reduced the list students had to learn. How Chinese characters were simplified places those methods in history.
Mergers are not a bug in Unicode. Unicode encodes both the Simplified shape and the Traditional shapes. The ambiguity lives in the meaning of the Simplified code point when you want to step back to an older distinction.
Not every Simplified character is a merger. 国 / 國 is essentially one word with two standard drawings. You can convert that pair back and forth with little drama. The dangerous cases are the ones where Traditional Chinese still uses two (or more) characters in careful writing.
Examples people actually hit
后 and 後. In Simplified Chinese, 后 can be the honorific in titles and also the word for "after" or "behind," which Traditional writing usually spells 後. A converter uses surrounding words to guess. In a personal name, a temple name, or a two-character heading, the guess can be wrong. The output still looks like real Chinese.
发, 發 and 髮. 发 is the Simplified form for both 發 (to issue, to develop, to shoot) and 髮 (hair). Phrase tables help on common words: 头发 is very likely to become 頭髮, and 发展 to become 發展. A surname or a coined product name that contains 发 has no such safety net.
面. Simplified 面 covers the face / surface word and, in many texts, the food word that Traditional writing may distinguish as 麵 (noodles). A menu is a classic failure case: the converter has little syntax to work with, and the wrong character is embarrassing in public.
干. Simplified 干 collects several Traditional writings, including 乾 and 幹 in different words. Short labels and verb lists are where this one goes wrong.
These are teaching examples, not a complete dictionary. Different house styles also treat 台 / 臺 and a few place-name characters as policy decisions rather than purely automatic ones.
| Simplified | Traditional forms you may need | Where software slips |
|---|---|---|
| 后 | 后, 後 | Names versus "after" |
| 发 | 發, 髮 | Hair words versus "develop / issue" |
| 面 | 面, 麵 | Menus and food packaging |
| 干 | 干, 乾, 幹 | Short verbs and technical labels |
| 复 | 復, 複, 覆 | Formal prose and headings |
A human who knows the word can choose in a second. A dictionary chooses by statistics and phrase length.
What this means for software
OpenCC is good on ordinary prose because ordinary prose contains common two-character and three-character words that sit in phrase tables. It is weaker on:
- rare personal names
- poetry and couplets
- mixed classical Chinese
- spreadsheet cells with a single character
- slide titles with no sentence around them
- invented brand spellings
That is a limit of dictionaries, not a reason to convert a 200-page file by hand one glyph at a time. It is a reason to search the output for the merger set you care about.
Round-tripping makes the problem worse. If you convert Traditional to Simplified, the merger happens. If you convert back, the engine must guess. Keep the Traditional original if you will need it again. Conversion vs translation is a different issue; this page is about a fork inside Chinese.
What you should do
- Pick the matching regional type, because Taiwan and Hong Kong may not even want the same Traditional character in edge cases. See OpenCC conversion types.
- Convert with the text or document tool.
- Search the output for 后, 发, 面, 干, 复 and any merger your document type uses often (food files need 面; biography files need 后 and 发).
- Fix names by hand. Do not run a second full conversion to repair one cell.
- Use the wider proofreading checklist before you publish.
If you maintain a company glossary, keep a column for "do not auto-convert." Paste those strings back from the source after the automatic pass.
Frequently asked questions
Is this why people say conversion is irreversible? It is the main linguistic reason. Encoding damage is the main technical reason. They are different failures.
Can a smarter model solve every merger? Better context helps on sentences. Isolated names will always need a person or a trusted list.
Do I need to worry when converting Traditional to Simplified? You need to worry less about picking the wrong split, and more about destroying a distinction you may want later. Keep the source.
Why did a common word convert correctly but a name next to it fail? Because the common word was in a phrase table and the name was not.
Are merged characters the same as simplified stroke reduction? No. Stroke reduction can be one-to-one. A merger is many-to-one, and that is the hard direction to undo.