Beyond Words

不止单词 · 织词成文 — an Obsidian plugin.

Turn a vocabulary list into a graded English article via LLM — every target word appears naturally in context and gets highlighted, with per-paragraph translations and a phrase bank. Learn words in context, not in isolation.

中文说明

Features

• Word sources: paste words directly, pick a .txt word list (one word per line), or a PDF vocabulary file from your vault (text is extracted with Obsidian's bundled PDF.js).
• Graded generation: pick your IELTS listening band (4.0–9.0) — the article stays comprehensible at your level with a few stretch items (Krashen's i+1).
• Style & length control: narrative / expository / journal, 100–800 words.
• Translations: optional per-paragraph translation (简体中文 / English / 日本語) shown as a subtle quote block under each paragraph.
• Phrase bank: the LLM extracts 5–10 useful collocations from the article into a table.
• Local verification: articles are checked for missing target words and retried automatically (up to 3 attempts).
• Highlighting: target words (including inflected forms) are highlighted with ==mark== in the generated note — done locally, no reliance on LLM markup.
• Mock mode: no API key? A built-in mock generator lets you try the whole pipeline for free.

Usage

1. Click the ribbon icon (book) or run "生成单词文章" from the command palette.
2. Paste words, or select a word list file from your vault.
3. Choose count, difficulty, length, style, translation language.
4. Click 生成文章. The note is saved to the BeyondWords/ folder and opened automatically.

API setup

Any OpenAI-compatible endpoint works. Presets included:

Provider	Base URL	Model
DeepSeek (default)	https://api.deepseek.com/v1	deepseek-chat
Kimi (high-speed)	https://api.kimi.com/coding/v1	kimi-for-coding-highspeed
OpenAI	https://api.openai.com/v1	gpt-4o-mini

Your API key is stored only in this vault's plugin data.json. Without a key, the plugin falls back to the mock generator.

Development

npm install
npm run dev    # watch mode, rebuilds main.js on change
npm run build  # type-check + production bundle

main.js is the bundled artifact — always regenerate it from src/ with npm run build, never edit it by hand.

Known limitations

• Stem matching does not recognize internal mutations (e.g. run → ran); the LLM-reported form field covers this in real generation.
• Scanned (image-only) PDFs yield no text — use OCR first.
• Word hover definitions, spaced-repetition write-back, and TTS are not implemented yet.

---

中文说明

把背诵单词列表交给 LLM，生成一篇自然用到全部目标词的分级英语文章，目标词自动高亮，逐段附译文，文末附高频短语表和词汇对照表。在语境中学用法，而不是孤立背单词。

• 单词来源：直接粘贴 / 库内 txt 词表 / 库内 PDF 词表（自动提取每行首个英文词）
• 难度：雅思听力 4.0–9.0（可理解输入 i+1）
• 文体：记叙文 / 说明文 / 日记；长度 100–800 词
• 译文：每段英文下给出整段译文（中/英/日可选）
• 本地校验缺词自动重试（≤3 次）；无 API key 时自动走 Mock 生成器

API Key 仅保存在本库插件目录的 data.json 中，不会上传。