Translation

💡TL;DR

Notemd translates text between 21+ languages using LLM-powered translation. Supports single-selection translation, full-note translation, and batch folder translation. Each translation task can use a dedicated provider and model via per-task settings. Output language is independently configurable from the UI language. Results are appended or written to a new file depending on your preference.

This is part of the Obsidian AI Knowledge Management Guide.

Overview

Translation in Notemd is not a dictionary lookup -- it is LLM-powered, context-aware translation. The model sees the full paragraph or note, preserving tone, domain terminology, and sentence structure. This produces higher-quality results than phrase-by-phrase services, especially for technical, academic, and creative writing.

The feature supports three scopes: selection, active note, and entire folder. Combined with per-task model selection, you can use a fast model (Gemini Flash) for casual translation and a powerful model (Claude Sonnet) for nuance-sensitive content -- without changing your global provider.

How It Works

The Translate Command

1. Source detection -- The LLM infers the source language from the content. You do not need to specify it manually.
2. Prompt construction -- Notemd builds a prompt that includes the target language, optional domain hint, and the content to translate.
3. LLM translation -- The configured translateProvider / translateModel processes the request. The model preserves markdown formatting, wiki-links, and code blocks.
4. Output -- The translated text is either appended below the original or written to a new file in the vault.

Language Pairs

Notemd supports any language pair that the underlying LLM supports. Common pairs include:

Source	Target	Typical Quality
English	Chinese (Simplified)	Excellent
Chinese	English	Excellent
English	Japanese	Very good
English	German / French / Spanish	Very good
Any supported	Any supported	Model-dependent

The translateLanguage setting controls the output language. The source language is auto-detected.

Per-Task Model Selection

Translation quality varies significantly by model. Notemd lets you assign a dedicated model just for translation:

Model	Speed	Quality	Cost	Best For
gemini-2.0-flash-exp	Fast	Good	Low	Casual, high-volume
gpt-4o-mini	Fast	Good	Low	Quick lookups
deepseek-chat	Medium	Good	Very low	Budget multilingual
claude-3-5-sonnet	Medium	Excellent	Medium	Technical / academic
gpt-4o	Medium	Excellent	Medium	Nuance-sensitive prose

Batch Folder Translation

Right-click a folder and select "Notemd: Translate folder" to translate every note in that folder. Each file is processed independently. The concurrency setting controls how many files translate in parallel.

Configuration

Setting	Default	Effect
translateProvider / translateModel	DeepSeek	Dedicated provider for translation tasks
translateLanguage	'en'	Target output language
translationAppendToNote	true	Append translated text below the original. If false, creates a new file.
batchConcurrency	3	Number of files processed in parallel during batch translation

Example

You are reading a Chinese research note and want an English version:

1. Open the note
2. Right-click --> "Notemd: Translate current file"
3. Notemd detects Chinese, translates to your configured target language (English), and appends:

## Translation (English)

The experimental results show that the proposed method achieves
a 12% improvement in F1 score compared to the baseline, primarily
due to the enhanced feature extraction module described in Section 3.

The original Chinese text is untouched above the translation. The ## Translation heading keeps both versions in the same file for easy reference.

Tips

• Use Gemini Flash for volume -- it is the fastest and cheapest option for batch translation of large folders.
• Preserve wiki-links -- Notemd's prompt instructs the LLM to keep [[wiki-links]] intact in the translation. Verify after translation, as some models occasionally unwrap them.
• Set output language explicitly -- auto-detection works for source, but always configure translateLanguage to avoid ambiguity about the target.
• Batch-translate concept notes -- if your concept folder is in one language and you need it in another, folder-level translation handles it in one step.

---

Next Steps

• Research -- Search and summarize in any language, then translate results
• Workflows -- Chain translation with wiki-linking or concept extraction
• Batch Processing -- Concurrency and overwrite behavior for folder operations
• LLM Providers -- Choose the best model for your language pair