google/mozc
Mozc
Mozc is a Japanese input method engine that converts romanized or kana keystrokes into Japanese kanji, kana, and mixed text with predictive conversion. It functions as a cross-platform input method framework, compiling the same engine into platform-specific binaries for Android, Linux, macOS, and Windows.
The engine combines a dictionary-based segmentation system with a hybrid of rule-based and statistical language models to determine word boundaries and predict character sequences. A prefix-tree structure enables real-time, context-aware conversion suggestions as the user types, while a key-event rewriting pipeline transforms raw keystrokes into composable text segments before passing them to the conversion engine.
Mozc manages input through a deterministic state machine that switches between hiragana, katakana, half-width, and direct ASCII modes during composition. The system separates the input method frontend from the conversion engine backend using inter-process communication, keeping the engine platform-independent while allowing platform-specific user interfaces.
Features
- Japanese Text Converters - Converts typed romanized or kana syllables into Japanese kanji, kana, and mixed text using a predictive engine.
- Prediction Engines - Uses statistical language models to predict and convert typed syllables into Japanese characters and words.
- Hybrid Language Models - Combines hand-crafted grammatical rules with a statistical n-gram model for predicting kanji and kana output.
- Morpheme Segmentation Dictionaries - Splits typed romanized or kana input into morphemes using a precompiled dictionary for accurate conversion.
- Input Method Engine Compilers - Compiles the input method engine into platform-specific binaries for Android, Linux, macOS, and Windows.
- Predictive Conversion Tries - Builds a trie of word and phrase candidates from a language model for real-time Japanese text conversion.
- Input Method Text Engines - Converts romanized or kana keystrokes into Japanese kanji, kana, and mixed text with predictive conversion.
- Japanese Input Mode Toggles - Toggles between hiragana, katakana, half-width, and direct ASCII input modes during composition.
- Keystroke Rewriting Pipelines - Intercepts and transforms raw keystrokes into composable text segments for complex Japanese input sequences.
- Input Mode State Machines - Manages a deterministic state machine switching between hiragana, katakana, half-width, and ASCII modes.
- Cross-Platform Build Systems - Uses a unified build configuration to compile the same engine source into native binaries for multiple platforms.
- Keyboard Input Modes - Toggles between hiragana, katakana, half-width, and direct ASCII input modes during text composition.
- Client-Server Architectures - Separates the IME frontend from the conversion engine backend via inter-process communication.
- Cross-Platform Development - Builds and compiles a Japanese input method engine into platform-specific binaries for multiple operating systems.
- Cross-Platform Frameworks - Provides a build system that compiles the same input method engine into platform-specific binaries for multiple operating systems.
