README file from
GithubLanguage Leaf
Read native Markdown, look up words and sentences, and save vocabulary as flashcards in the same note. Review with the built-in FSRS engine.
Installation and usage
Download main.js, manifest.json, and styles.css from the latest release. Copy them to <vault configuration folder>/plugins/language-leaf/, reload Obsidian, and enable Language Leaf in Community plugins. Preserve your existing data.json. Requires desktop Obsidian 1.14.4 or later.
- Enable learning mode on a Markdown note. Hold Shift while selecting text, or configure Alt/Option. Three-finger selection can query when the modifier is released. Alternatively, enable automatic selection lookup or click the selection's lookup button.
- Edit the new-word form and save with Command+Enter on macOS or Ctrl+Enter on Windows/Linux. The shortcut applies inside the new-word panel.
- The selection lookup command works in both editing and reading views.
- New cards always use the built-in FSRS engine.
- Built-in FSRS review supports recall and context layouts, audio, daily new-card limits, undo, pause/resume, and source-note navigation. Space reveals the answer; arrow keys rate it after reveal.
- Switching a note to editing exits learning mode.
Sentence context and repeated captures
For a target sentence with fewer than 15 words, Language Leaf includes one sentence before and after it by default. Change the word threshold and the number of sentences on each side in the reading and lookup settings; set either value to 0 to disable expansion. Changes apply to future captures only.
The original paragraph and line breaks are retained. Sentences already in that paragraph count toward the context range. Expansion stops at headings, code blocks, and ordinary blockquotes; translation callouts are skipped when locating original text. If a boundary falls inside quoted dialogue, the complete dialogue is included, so the range can exceed the configured sentence count. If the selected text occurs several times in the same paragraph, the original paragraph is retained without guessing which occurrence was selected.
When complete captured paragraphs all have matching translation callouts, their existing translations are kept in order. Otherwise, use optional automatic translation or enter the translation manually.
Capturing the same word or phrase again in the same note places the current example first and preserves earlier examples, meaning, notes, tags, and review progress. Selecting the same example again moves it to the front without adding a duplicate; whitespace-only differences are ignored. A successful capture can still increase the capture count. Opening an existing card for editing does not add an example or increase that count.
Across different notes, captures remain separate cards with their own source examples and review progress; the shared expression identity and existing meaning can be reused.
In built-in review, the first example is the main context. Context layout shows it before revealing the meaning and translation; recall layout shows it after reveal. After reveal, other examples are available in collapsed supplementary example sections containing their original text and translation. All examples belong to one card and receive one rating; examples do not rotate randomly.
Review data and migration
Card content stays in Markdown. Built-in review state is stored as append-only events under LinguaLeafData/review/ by default. Synchronize the entire data folder and resolve sync conflicts before continuing review.
Existing notes, vocabulary, and settings are not automatically migrated from the old Language Learner plugin.
Privacy and attribution
Enabled dictionaries receive the selected word or sentence. Optional context translation sends captured original paragraphs, including expanded neighboring context, to Youdao when a complete matching translation is unavailable. Online pronunciation uses dictionary recordings or Youdao audio. System speech uses voices available on the device through Obsidian; voice availability and quality depend on the system. Notes and review data stay in your vault; the plugin provides no remote data sync.
This project is derived from Language Learner by the_tree, retains dictionary and Vue panel code, and preserves the original MIT license and attribution. It is independently maintained by Cynthia Ding. Independent maintenance does not imply that inherited code was written from scratch.
The built-in review engine uses ts-fsrs 5.4.1, licensed under MIT. See third-party license.
Development
Use Node.js 22.12 or later:
npm ci
npm run test:run -- --pool=forks --maxWorkers=1 --no-file-parallelism
npm run build
Release assets are built and attested in GitHub Actions. See release notes and repository maintenance. Licensed under MIT.