9314JIUsi
Office tools · browser-local conversion

Chinese Script Desk: Phrase-First, Variants Normalized

Character-by-character converters turn 内存 into 內存 - the glyphs are right, the word is wrong, and colleagues in Taiwan pause for three seconds reading it. This desk ships a mapping of 2,300+ single characters and 2,900+ phrases, replacing whole phrases first with single-character fallback: 鼠标→滑鼠, 打印机→印表機, every changed character highlighted, round trips designed to close in both directions. 1MB of text converts locally in a blink.

SourceSimplified or Traditional
ResultConverted

Change highlighting: characters altered by the conversion are marked here (brown background)
Direction

Text you paste is converted in this page's memory only - nothing uploaded, no account needed. Interface options like direction and the highlight toggle live in this device's browser and are lost when you switch devices or clear site data. To remove them, use "Clear my data" in the footer.

How to use it

  1. Paste text into the left pane (leave direction on auto-detect if unsure - it judges by the ratio of Simplified to Traditional characters) and hit "Convert".
  2. The result appears on the right, with the brown highlight block underneath marking every changed character - one glance tells you what moved.
  3. Hit "Round-trip check" to verify: convert over and back, compare character by character against the source, and list any differences (variant normalization counts as a normal difference).
  4. Copy the result or download it as .txt; for a reverse check, move the result back to the source pane and convert again.

Conversion strategy: phrases first, characters as fallback

The converter scans text left to right and tries the longest phrase first at each position (up to 5 characters): a hit on 打印机 swaps the whole word for 印表機; only when no phrase matches does it fall back to the single-character table. The phrase tables are two independent, directional maps - one for Simplified→Traditional (内存→記憶體) and one for Traditional→Simplified (記憶體→内存) - not a simple inversion.

Polyphonic characters are locked by phrases: 皇后 (where 后 must not become 後), 太后 and 王后 all sit in the identity-phrase table; 头发→頭髮 versus 发廊→髮廊 - the two Traditional forms of 发 are decided by the word. This is exactly where character-by-character tools crash most often.

Variant normalization: 里→裡 (Taiwan-preferred), while Traditional→Simplified maps both 裏 and 裡 back to 里 - both Traditional input conventions are accepted. This normalization is the main intentional edit in round trips, fully visible in the diff highlight.

When it earns its keep

Sending documents to clients in Hong Kong or Taiwan, shipping a Traditional-Chinese game localization, preparing dual-language subtitle tracks - the nightmare is half-converted trash like 內存. The phrase dictionary aligns high-frequency IT and everyday vocabulary (内存/鼠标/硬盘/软件/打印机/网络...) as whole words, so the output reads naturally to Taiwanese readers.

The reverse works the same: archive an incoming Traditional contract as Simplified - paste, click, copy, with variant characters normalized along the way.

InputAny mixed Chinese/English text (direction auto-detected by the Simplified/Traditional character ratio); ≤1MB
OutputFull text in the target script + a diff view highlighting changed characters
Key parametersDirection auto by default; ?dir=s2t or ?dir=t2s deep-links; highlight toggle on by default
ProcessingBrowser-local conversion (dictionaries embedded in the page, zero requests, zero uploads), works offline; interface options in localStorage
Limits1MB per run (~300k Chinese characters, milliseconds); the diff view renders only changed segments to stay fast
Known limits~2,900 high-frequency phrases covered; rare words beyond the table fall back to per-character conversion, which may not match Taiwan conventions; variants (裏/裡) normalize on round trips; no localization beyond vocabulary (currency and date formats untouched); classical Chinese and archaic characters not covered

FAQ

Why should 内存 become 記憶體 instead of 內存?

Character-by-character conversion only swaps glyphs: 内→內, 存→存, yielding 內存. But Taiwan does not use that word - they write 記憶體. It is a vocabulary difference, not a glyph difference. Same family: 鼠标→滑鼠, 软件→軟體, 打印机→印表機. This tool matches phrases first - the phrase table replaces whole words on hit, single-character fallback only when nothing matches - so these words never embarrass themselves.

If I convert Simplified to Traditional and back, do I get the original verbatim?

For the vast majority of everyday text, yes: the dictionaries are designed for closed Simplified→Traditional→Simplified round trips, and polyphonic characters like 皇后, 头发, 面条 are locked by phrases. Variant characters are the known loss point: Simplified 里 converts to Traditional 裏 by default, and back converts both 裏 and 裡 into 里 - if your source mixed variants, the round trip normalizes them. Rare characters beyond dictionary coverage pass through untouched.

How are variant characters handled?

The same Traditional character often has multiple forms: 里 has 裏 (the mainland standard) and 裡 (common in Taiwan). This tool converts Simplified→Traditional with the Taiwan-preferred form (里→裡), while Traditional→Simplified maps both 裏 and 裡 back to 里, accepting input from both conventions. This normalization is the main intentional edit in round trips, and the diff highlight marks every instance.

What is the size limit for conversion?

Up to 1MB of text (roughly three hundred thousand Chinese characters) converts in milliseconds in the browser. Beyond 1MB the page asks you to split. The diff highlight renders only changed segments on large texts, so it never stalls.

Is my text uploaded anywhere?

No. Conversion happens inside this device's browser with the dictionaries built into the page - no requests, works offline. Only interface options live in this browser's localStorage; Clear my data in the footer removes them.

Where do the dictionaries come from? How big are they?

Programmatically distilled from the open-source OpenCC dictionaries: 2,300+ common single characters and 2,900+ high-frequency phrases selected by GB2312 frequency weighting (including Taiwan vocabulary substitutions and polyphonic locks), inlined into the page as "simplified=traditional" line text with no external requests. The generation script and criteria live in the delivery notes. OpenCC is Apache-2.0 licensed; the distilled dictionaries follow the same license.

Related tools