BYVoid/OpenCC

Open Chinese Convert 開放中文轉換

Introduction 介紹

Open Chinese Convert (OpenCC, 開放中文轉換) is an open source project for high-quality conversion between Traditional Chinese, Simplified Chinese, Japanese Shinjitai, and regional wording across Mainland China, Taiwan, and Hong Kong. It provides dictionaries, a reusable library, conversion tools, and dictionary generation tools.

Open Chinese Convert（OpenCC，開放中文轉換） 是一個開源中文轉換項目，支持繁體中文、簡體中文、日文新字體，以及中國大陸、臺灣、香港等地區習慣用詞之間的高品質轉換，並提供詞典、可重用庫、轉換工具及詞典生成工具。

Discussion (Telegram) 討論區（Telegram）: https://t.me/open_chinese_convert

Features 特點

• 基於詞庫的確定性轉換，不使用大語言模型，結果穩定可預期，可離線快速運行。 Deterministic, dictionary-based conversion without large language models (LLMs): results are stable and predictable, and conversion runs fast and fully offline.
• 詞庫和函數庫可分離，允許自定義修改和擴展。 Dictionaries are decoupled from the library, allowing custom modification and extension.
• 嚴格區分簡繁對應與異體字對應關係，嚴格審校一簡對多繁詞條，OpenCC標準的用字原則爲「能分則不合」。 Strictly distinguishes Simplified-Traditional mappings from character variant mappings, with rigorously reviewed one-to-many Simplified-to-Traditional entries; the OpenCC standard follows the character usage principle of "separate whenever distinguishable".
• 支持中國大陸、臺灣、香港地區慣用異體字（如「裏」「裡」），以及習慣用詞轉換（如「鼠標」「滑鼠」）。 Supports conversion of preferred character variants (such as 「裏」/「裡」) and regional wording (such as 「鼠標」/「滑鼠」) for Mainland China, Taiwan, and Hong Kong.
• 詞庫內容均以官話（普通話、國語）詞彙爲基準，不支持語言之間的翻譯，例如官話與粵語、閩南語或日語之間的翻譯。 Dictionaries are based on Mandarin (Putonghua / Guoyu) vocabulary; translation between languages, such as between Mandarin and Cantonese, Southern Min, or Japanese, is not supported.
• 另外，有限支持日文新字體與舊字體之間的轉換。 In addition, conversion between Japanese Shinjitai (new character forms) and Kyujitai (old character forms) is supported to a limited extent.

詳情參閱 OpenCC 設計思想及地區詞收錄標準。 For details, see Design Principles and Regional Phrase Criteria.

Installation 安裝

Package Managers 包管理器

• Debian
• Ubuntu
• Fedora
• Arch Linux
• macOS (Homebrew)
  • 使用 brew install opencc
• WinGet
  • 使用 winget install opencc
• Bazel
• Node.js
  • 使用 npm install -g opencc 命令可安裝 OpenCC Node.js CLI
  • 使用 npm install -g opencc opencc-jieba 命令可同時安裝 OpenCC Node.js CLI 及 Jieba 分詞插件
• Python
  • 使用 pip install opencc 命令可安裝 Python API 及 Python CLI
• Repology

Prebuilt binaries 預編譯二進位檔

OpenCC 1.4.2 大幅加速了轉換熱路徑（純文字語料的整體轉換時間最多降至原本的 1/7），修復了大端序平台載入 legacy .ocd 字典得到空字典的問題，並包含一批詞庫 修正；C++ ABI 與 1.4.1 相同（SOVERSION 1.4），自 1.4.0／1.4.1 升級的下游 C++ 程式無需重新連結。

• Windows (x86_64): OpenCC-1.4.2 (SHA-256)
  • This Windows release is available from WinGet. For details, see doc/windows-winget-release.md.
  • The executables are Authenticode-signed; see Code Signing 程式碼簽章.
  • Requires Microsoft Visual C++ Redistributable for Visual Studio 2015-2026. Download the latest version from Microsoft.
• Debian/Ubuntu:
  • opencc-1.4.2-1-deb-amd64.zip
  • opencc-1.4.2-1-deb-arm64.zip
  • Each zip bundles the opencc, opencc-jieba, and libopencc* deb packages for one architecture, with a SHA256SUMS file.

Usage 使用

Online 線上轉換

https://opencc.js.org/converter?config=s2t

Node.js

npm install opencc

The npm package supports Node.js >=20.17. The native addon is installed through prebuilt @opencc/opencc-<platform>-<arch> packages covering macOS (x64/arm64), Linux (x64/arm64), and Windows (x64). On platforms without a prebuilt binary package, npm install builds from source with Bazel — compiling the addon and regenerating the dictionaries — which requires a C++ toolchain and network access; Bazel is found on PATH (bazel or bazelisk) or fetched automatically via npx, and Bazel downloads its own hermetic Python toolchain for the dictionary generation.

Bun and Deno can also use the npm package through their npm compatibility support.

To install the npm CLI:

npm install -g opencc
opencc -c s2t.json -i input.txt -o output.txt

The npm CLI supports basic text conversion. Plugins, --inspect, --segmentation, and --ambiguities require the native OpenCC CLI.

import { OpenCC } from 'opencc';
async function main() {
  const converter: OpenCC = new OpenCC('s2t.json');
  const result: string = await converter.convertPromise('汉字');
  console.log(result);  // 漢字
}

Inline configurations can be passed with OpenCC.fromConfig:

import { OpenCC } from 'opencc';
const converter = OpenCC.fromConfig({
  name: 'Demo Inline Config',
  conversion_chain: [{
    dict: { type: 'inline', entries: { '鼠标': '滑鼠' } },
  }],
});
console.log(converter.convertSync('鼠标'));  // 滑鼠

See demo.js and demo.ts.

Python

pip install opencc (Windows, Linux, macOS)

import opencc
converter = opencc.OpenCC('s2t.json')
converter.convert('汉字')  # 漢字

The Python package also installs a basic CLI:

pip install opencc
opencc -c s2t.json -i input.txt -o output.txt

The Python CLI supports basic text conversion, --include-tofu-risk-dictionaries, and --resource-zip. Diagnostic modes such as --inspect, --segmentation, and --ambiguities still require the native OpenCC CLI.

C++

#include "opencc.h"

int main() {
  const opencc::SimpleConverter converter("s2t.json");
  converter.Convert("汉字");  // 漢字
  return 0;
}

Full example with Bazel

When OpenCC is embedded in a server binary or self-contained application, the JSON config can stay small while dictionary resources are loaded from explicit resource directories:

#include <memory>
#include <vector>

#include "SimpleConverter.hpp"

int main() {
  auto resources = std::make_shared<opencc::FilesystemResourceProvider>(
      std::vector<std::string>{
          "/opt/my-app/opencc",
          "/opt/my-app/plugins/opencc-jieba",
          "/usr/share/opencc",
      });
  const opencc::SimpleConverter converter("s2t.json", resources);
  converter.Convert("汉字");
  return 0;
}

FilesystemResourceProvider searches directories in order. Existing SimpleConverter("s2t.json") and CLI behavior continue to use the config file location, current directory, explicit paths, and installed OpenCC data directory as before.

C

#include "opencc.h"

int main() {
  opencc_t opencc = opencc_open("s2t.json");
  const char* input = "汉字";
  char* converted = opencc_convert_utf8(opencc, input, strlen(input));  // 漢字
  opencc_convert_utf8_free(converted);
  opencc_close(opencc);
  return 0;
}

Full Document 完整文檔

Command Line

Unless otherwise noted, this section describes the native OpenCC CLI built from the C++ toolchain. The Python and npm CLIs support basic file/stdin conversion only, plus --include-tofu-risk-dictionaries; the Python CLI also supports --resource-zip.

• opencc --help
• opencc_dict --help

Segmentation and Inspection Modes

OpenCC CLI supports two diagnostic modes that output JSON instead of converted text:

--segmentation — Output segmentation result only (no conversion):

echo "他只看了几行日志，就一叶知秋，猜到整个系统是数据库连接池出了问题" | opencc -c s2twp.json --segmentation
# {"input":"他只看了几行日志，就一叶知秋，猜到整个系统是数据库连接池出了问题","segments":["他","只看","了几行","日志","，就","一叶知秋","，猜到","整个","系统","是","数据库","连接池","出了","问题"]}

--inspect — Output full inspection result (segmentation + per-stage conversion + final output):

echo "他只看了几行日志，就一叶知秋，猜到整个系统是数据库连接池出了问题" | opencc -c s2twp.json --inspect
# {"input":"他只看了几行日志，就一叶知秋，猜到整个系统是数据库连接池出了问题","segments":["他","只看","了几行","日志","，就","一叶知秋","，猜到","整个","系统","是","数据库","连接池","出了","问题"],"stages":[{"index":1,"segments":["他","只看","了幾行","日誌","，就","一葉知秋","，猜到","整個","系統","是","數據庫","連接池","出了","問題"]},{"index":2,"segments":["他","只看","了幾行","日誌","，就","一葉知秋","，猜到","整個","系統","是","資料庫","連線池","出了","問題"]},{"index":3,"segments":["他","只看","了幾行","日誌","，就","一葉知秋","，猜到","整個","系統","是","資料庫","連線池","出了","問題"]}],"output":"他只看了幾行日誌，就一葉知秋，猜到整個系統是資料庫連線池出了問題"}

# Pretty-print with jq:
echo "他只看了几行日志，就一叶知秋，猜到整个系统是数据库连接池出了问题" | opencc -c s2twp.json --inspect | jq .

--ambiguities — Convert while marking every output span whose dictionary match is one-to-many (e.g. 文丑 may be either 文丑 or 文醜), as a stream of JSONL records:

printf '大战文丑的时候，他的头发很干燥' | opencc -c s2t.json --ambiguities
# {"def":"文丑"}
# {"lit":"大戰"}
# {"amb":{"t":"文丑","s":0}}
# {"lit":"的時候，他的頭髮很乾燥"}
# {"end":{"output_bytes":45,"ambiguities":1,"sources":1}}

Record kinds: {"def"} defines the next source index (each distinct input word is defined once, before its first reference); {"lit"} is a literal run of unambiguous output; {"amb":{"t","s"}} is an ambiguous span whose output text t (the default candidate) came from the source with index s; the final {"end"} record carries stream totals. Concatenating every lit and amb.t in order reproduces the plain conversion output exactly. Streaming is bounded-memory regardless of input size or line length, and positions are implicit in record order, so consumers can rebuild offsets in their own string-index units. Single-value conversions (e.g. 头发 → 頭髮) are not flagged.

The record stream is a machine-readable contract: every emitted line is valid JSON, and the final {"end"} record only appears when the whole input was processed. Input containing invalid UTF-8 aborts the stream with an error (unlike plain conversion, which is byte-transparent), and a missing {"end"} record means the stream is incomplete.

These modes are useful for diagnosing conversion issues:

1. Use --segmentation to verify that the input is segmented as expected.
2. Use --inspect to see which conversion stage produces an unexpected result.
3. Use --ambiguities to locate one-to-many conversions and resolve each deduplicated def source to its candidates.

Rules:

• --segmentation, --inspect, and --ambiguities are mutually exclusive.

Official / Recommended Ports

The following ports are maintained within the OpenCC ecosystem and are generally up to date with current configuration and dictionary data.

• Data package: opencc-data
• Pure JavaScript: opencc-js
  • See notes about different OpenCC NPM packages below.
• WebAssembly: opencc-wasm (website)
• Pure Python: opencc-py (pre-release)

Other Ports (Unofficial)

These ports are community-maintained and may not always track upstream updates.

• Swift (iOS): SwiftyOpenCC
• iOSOpenCC (pod): iOSOpenCC
• Java: opencc4j
• Android: android-opencc
• PHP: opencc4php
• WebAssembly: wasm-opencc
• Browser Extension: opencc-extension
• Go (Pure): OpenCC for Go
• Dart (native-assets): opencc-dart

Configurations 配置文件

預設配置文件

• s2t.json Simplified Chinese to Traditional Chinese (OpenCC Standard) / 簡體 到 OpenCC 標準繁體
• t2s.json Traditional Chinese (OpenCC Standard) to Simplified Chinese / OpenCC 標準繁體 到 簡體
• s2tw.json Simplified Chinese to Traditional Chinese (Taiwan Standard) / 簡體 到 台灣正體
• tw2s.json Traditional Chinese (Taiwan Standard) to Simplified Chinese / 台灣正體 到 簡體
• s2hk.json Simplified Chinese to Traditional Chinese (Hong Kong variant) / 簡體 到 香港繁體
• hk2s.json Traditional Chinese (Hong Kong variant) to Simplified Chinese / 香港繁體 到 簡體
• s2twp.json Simplified Chinese to Traditional Chinese (Taiwan Standard, with Taiwan Phrases) / 簡體 到 台灣正體（含台灣常用詞彙）
• tw2sp.json Traditional Chinese (Taiwan Standard) to Simplified Chinese (Mainland China Phrases) / 台灣正體 到 簡體（含中國大陸常用詞彙）
• t2tw.json Traditional Chinese (OpenCC Standard) to Traditional Chinese (Taiwan Standard) / OpenCC 標準繁體 到 台灣正體
• tw2t.json Traditional Chinese (Taiwan Standard) to Traditional Chinese (OpenCC Standard) / 台灣正體 到 OpenCC 標準繁體
• t2hk.json Traditional Chinese (OpenCC Standard) to Traditional Chinese (Hong Kong variant) / OpenCC 標準繁體 到 香港繁體
• hk2t.json Traditional Chinese (Hong Kong variant) to Traditional Chinese (OpenCC Standard) / 香港繁體 到 OpenCC 標準繁體

下列配置文件仍在開發中，歡迎貢獻新詞組：

• s2hkp.json Simplified Chinese to Traditional Chinese (Hong Kong variant, with Hong Kong Phrases) / 簡體 到 香港繁體（香港常用詞彙）
• hk2sp.json Traditional Chinese (Hong Kong variant) to Simplified Chinese (Mainland China Phrases) / 香港繁體 到 簡體（含中國大陸常用詞彙）

下列配置文件僅供探索性研究，不建議用於生產環境：

• t2jp.json Old Japanese Kanji (Kyūjitai) to New Japanese Kanji (Shinjitai) / 日文舊字體 到 日文新字體
• jp2t.json New Japanese Kanji (Shinjitai) to Old Japanese Kanji (Kyūjitai) / 日文新字體 到 日文舊字體，並將少量日文詞組轉換爲對應中文
• t2seal.json Traditional Chinese Characters to Small Seal Script (Unicode 18.0 Seal block) / 繁體漢字 到 小篆（Unicode 18.0 篆書區塊 U+3D000..U+3FC3F），需搭配支援該區塊的字型顯示
• s2seal.json Simplified Chinese to Small Seal Script (Unicode 18.0 Seal block) / 簡體 到 小篆，相當於 s2t 後接 t2seal。反方向（小篆到簡體）請依序執行 seal2t 與 t2s：分詞只在輸入端進行一次，小篆輸入無法為 t2s 的詞組轉換分詞
• seal2t.json Small Seal Script (Unicode 18.0 Seal block) to Traditional Chinese Characters / 小篆（Unicode 18.0 篆書區塊） 到 繁體漢字，依 Unicode SealSources.txt 的 kSEAL_MCJK 屬性對應，並以 SealVariants.txt 銜接現代標準字與《說文》隸定字（如 年／秊）；《說文》未收之字（如 你、們）保持不變

指定配置文件

通过环境变量OPENCC_DATA_DIR加载指定路径下的配置文件

OPENCC_DATA_DIR=/path/to/your/config/dir opencc --help

內聯字典（inline dictionary）

配置檔中的字典可使用 type: "inline"，直接在 JSON 裡定義小型自訂詞彙， 不必修改外部字典檔。例如在 group.dicts 最前面加入覆寫規則：

{
  "conversion_chain": [
    {
      "dict": {
        "type": "group",
        "dicts": [
          {
            "type": "inline",
            "entries": {
              "麦旋风": "冰炫風",
              "服务器": "伺服器"
            }
          },
          {
            "type": "ocd2",
            "file": "STPhrases.ocd2"
          },
          {
            "type": "ocd2",
            "file": "STCharacters.ocd2"
          }
        ]
      }
    }
  ]
}

規則與限制：

• entries 必須是 JSON 物件。
• entries 的 key/value 必須是非空字串。
• 重複 key 不受支援；如包含，載入會直接失敗（拋出錯誤）。
• key/value 會按解析結果原樣使用，不做 trim、大小寫折疊或 Unicode normalization。
• 內聯字典與普通字典行為一致，優先級由 group.dicts 的順序決定。
• 內聯字典輸出仍會繼續經過後續 conversion_chain 步驟，不提供鎖定最終輸出。

備註：OpenCC 1.3.2+ 解析器支援有限 JSONC 語法（//、/* */ 註解與尾逗號）。 若需跨實作相容，建議使用嚴格 JSON，不依賴 JSONC 擴充。

更多完整示例可見 examples/config/。該目錄僅供學習與自訂參考，不屬於官方內建 配置列表。

Experimental Plugins 試驗性插件

OpenCC 現已支援外部 C++ 分詞插件。當前第一個插件為 opencc-jieba， 可通過 s2t_jieba.json、s2tw_jieba.json、s2hk_jieba.json、 s2twp_jieba.json、tw2sp_jieba.json 等插件配置啓用。

OpenCC now supports external C++ segmentation plugins. The first plugin is opencc-jieba, which can be enabled through plugin-backed configs such as s2t_jieba.json, s2tw_jieba.json, s2hk_jieba.json, s2twp_jieba.json, and tw2sp_jieba.json.

注意：

• 該插件機制目前仍為試驗性功能。
• jieba 插件是可選組件，Python 套件和 Node.js 套件都不要求它。自 1.4.1 起，macOS 上的頂層 CMake 構建（含 Homebrew）預設啓用該插件（BUILD_OPENCC_JIEBA_PLUGIN=ON）；其他平台與以子項目方式引入的構建預設仍為關閉。
• opencc-jieba 額外依賴 cppjieba 及其配套詞典資源，這些依賴僅在構建或分發該插件時需要。
• 在下一次正式發布版本之前，插件 ABI 仍可能發生變化，不應視為穩定介面。
• 我們預計從下一次正式發布版本開始，將插件 ABI 視為穩定介面。
• Windows 下插件必須與宿主 OpenCC 二進位使用 ABI 相容的工具鏈／執行時構建；MSVC 與 MinGW 產物不支援混用。

Notes:

• The plugin mechanism is currently experimental.
• The jieba plugin is optional and is not required by the Python package or the Node.js package. Since 1.4.1, top-level CMake builds on macOS (including Homebrew) enable the plugin by default (BUILD_OPENCC_JIEBA_PLUGIN=ON); other platforms and subproject builds keep it disabled by default.
• opencc-jieba additionally depends on cppjieba and its dictionary resources. These dependencies are only needed when building or distributing the plugin itself.
• The plugin ABI may still change before the next formal OpenCC release and should not yet be treated as stable.
• We expect to treat the plugin ABI as stable starting with the next formal OpenCC release.
• On Windows, plugins must be built with an ABI-compatible toolchain/runtime as the host OpenCC binary. Mixing MSVC-built hosts with MinGW-built plugins, or the reverse, is unsupported.

Build 編譯

Build with CMake

Linux & macOS

g++ 4.6+ or clang 3.2+ is required.

make

Windows Visual Studio:

build.cmd

Build with Bazel

bazel build //:opencc

Test 測試

Linux & macOS

make test

Windows Visual Studio:

test.cmd

Test with Bazel

bazel test --test_output=all //src/... //data/... //python/... //test/...

Benchmark 基準測試

make benchmark

詳情見 doc/benchmark.md 檔案。

Projects using OpenCC 使用 OpenCC 的項目

Please update if your project is using OpenCC.

• ibus-pinyin
• fcitx
• rimeime
• libgooglepinyin
• ibus-libpinyin
• alfred-chinese-converter
• GoldenDict
• China Biographical Database Project (CBDB)
• OpenCC-Traditional Chinese to Traditional Chinese (The Chinese Government Standard)
  • 將混雜不同標準的繁體字形轉換為中國《通用規範漢字表》(2013) 中的繁體字形
  • 展示說明頁面
  • 轉換方案設計說明

License 許可協議

Apache License 2.0

Third Party Libraries 第三方庫

• darts-clone BSD License
• marisa-trie BSD License
• tclap MIT License
• rapidjson MIT License
• Google Test BSD License
• cppjieba MIT License
  • Optional dependency used by the experimental opencc-jieba plugin.
  • 試驗性 opencc-jieba 插件使用的可選依賴。

Code Signing 程式碼簽章

自 1.4.2 起，Windows 發佈的二進位檔（可攜式 CLI 壓縮包中的執行檔，以及 npm win32-x64 套件中的 opencc.node 與 opencc-jieba.dll）皆經過 Authenticode 簽章。

Since 1.4.2, the Windows binaries — the executables in the portable CLI zip and the opencc.node / opencc-jieba.dll shipped in the win32-x64 npm packages — are Authenticode-signed.

This program uses free code signing provided by SignPath.io, and a free code signing certificate by the SignPath Foundation.

本項目使用 SignPath.io 免費提供的程式碼簽章服務， 憑證由 SignPath Foundation 免費提供。

• Code signing policy 簽章政策：只有合併進 master 的提交會被簽章，簽章請求 由 維護者在發佈流程中透過 GitHub Actions 發起 （見 release-winget.yml 與 release-npm-binaries.yml）。
• Privacy policy 隱私政策：OpenCC 為離線的文字轉換程式庫與命令列工具，除非 使用者、安裝者或操作者明確要求，否則不會將任何資訊傳輸至其他網路系統。 This program will not transfer any information to other networked systems unless specifically requested by the user or the person installing or operating it.

Change History 版本歷史

• NEWS
  • 另見 https://opencc.byvoid.com/news/

Links 相關連結

• Publications Using OpenCC - 近年來使用了 OpenCC 的研究論文選錄
• 現代漢語常用繁簡轉換匹配辨析表
• 關於 opencc, opencc-js 与 opencc-wasm 三个 NPM packages 區別的說明 https://github.com/nk2028/opencc-js/blob/HEAD/README-zh-TW.md#%E8%88%87-opencc-npm-package-%E7%9A%84%E5%8D%80%E5%88%A5
• 異寫字表 - Unicode 中雖碼位不同，但實屬同一漢字。不收錄已出現在「CJK 相容表意文字」及其補充區段中的字元。

Contributors 貢獻者

• BYVoid
• 佛振
• Peng Huang
• LI Daobing
• Kefu Chai
• Kan-Ru Chen
• Ma Xiaojun
• Jiang Jiang
• Ruey-Cheng Chen
• Paul Meng
• Lawrence Lau
• 瑾昀
• 內木一郎
• Marguerite Su
• Brian White
• Qijiang Fan
• LEOYoon-Tsaw
• Steven Yao
• Pellaeon Lin
• stony
• steelywing
• 吕旭东
• Weng Xuetian
• Ma Tao
• Heinz Wiesinger
• J.W
• Amo Wu
• Mark Tsai
• Zhe Wang
• sgqy
• Qichuan (Sean) ZHANG
• Flandre Scarlet
• 宋辰文
• iwater
• Xpol Wan
• Weihang Lo
• Cychih
• kyleskimo
• Ryuan Choi
• Prcuvu
• Tony Able
• Xiao Liang
• Frank Lin

Please feel free to update this list if you have contributed OpenCC.