Developer sokcuri Releases 'hdTrans', an Open-Source Offline Japanese-Korean Rule-Based Translator Replacing ezTrans

On October 20, 2026, developer sokcuri (@sokcuri) officially announced hdTrans, an open-source, offline, rule-based Japanese-to-Korean machine translation engine designed to replace the legacy commercial translation software ezTrans, which has been widely utilized across gaming and subculture translation workflows for over two decades.

Image source: @sokcuri / X official hdTrans GUI screenshot

Released under the Mozilla Public License 2.0 (MPL-2.0) on GitHub, hdTrans is built as a lightweight standalone engine that operates entirely locally without requiring internet access. The package bundles the default dictionary (dicts.hdsnap) and the translation core together, allowing users to extract the archive and run it immediately without complex installation procedures or build steps. A web-based demo (trans.kuri.kr) is also available for testing translation quality prior to local setup.

Seamless ezTrans API Compatibility and Third-Party Tool Integrations

The central practical highlight of hdTrans is its drop-in compatibility with the legacy ezTrans API. By simply specifying the hdTrans engine path within existing text hooking and on-screen overlay translation tools, users can operate their translation stack without needing an original ezTrans installation.

• Supported Tools: Supports extensive third-party utilities including LunaTranslator, TwoControl, Anemone, XUnity.AutoTranslator, AralTrans, and VNR.
• Multiple Interfaces: Provides a native Windows GUI (hdTransGUI.exe), a cross-platform command-line tool (hdtrans), and an npm package (hdtrans) tailored for Node.js and TypeScript environments.
• Cross-Platform Availability: Offers native standalone binaries for Windows, Linux, and macOS alongside WebAssembly (WASM) builds, eliminating Windows-only platform lock-in.

Context-Aware Speech Styles and Document-Level Processing

hdTrans introduces granular translation options that control tone and preserve semantic boundaries according to context:

• Speech Style Settings (strict / soft): The default strict mode preserves formal sentence endings ('-다') and only shifts to casual informal endings when the source explicitly contains conversational particles (such as よ, ね, or ！). The soft mode detects spoken dialogue—including bracketed quotes (「」)—and renders it in natural conversational Korean, making it well-suited for visual novel dialogue and chat interactions.
• Document Translation Mode: Beyond line-by-line translation (--each), the document-level mode examines surrounding context across multiple lines. This ensures consistent proper noun translation across passages and prevents narrative text from being improperly converted into informal speech.

According to developer self-reported benchmarks across a public regression suite of 66 sentences (comprising 30 revised expressions, 30 preserved expressions, and 6 sample phrases), exact match against expected output rose from 36/66 in the 0.2.0 source build to 66/66 on the refined C++ engine. As noted in the project documentation, this benchmark is a curated regression suite targeting specific dictionary corrections rather than a statistical measure of general translation accuracy.

As of October 26, 2026, the latest v0.6.2 release has been published, addressing compound noun segmentation and predicate marker glosses. Binaries and packages are available for download directly from the official GitHub repository.

Sources

• GitHub: sokcuri/hdTrans
• sokcuri (@sokcuri): hdTrans Release Announcement on X
• sokcuri (@sokcuri): Official hdTrans Windows GUI screenshot