Give your text to an AI tool and ask it to translate using this site's dictionary, grammar and texts. Translate in either direction between Northern or Southern Tujia and any other language the model supports.
Translation takes place in your AI tool. This page explains how to use the skill, answers common questions and shows actual examples.
Minimum requirement: the AI must have a tool that can fetch API data from the internet. It needs these tools to read the skill, dictionary and grammar materials.
A coding agent—an AI agent that can write code—is recommended. With internet access, it can usually read files and run lookup scripts to follow the skill. Chat tools may also work, depending on their available tools: check that they can access the internet and read API responses.
These test results from November 15–16, 2026 may help you choose a tool:
Tool
Mode or version
Test result
ChatGPT
Chat
✓
ChatGPT
Work
✓
Dola
Work
✓
Kimi
Free (Instant)
✓
Gemini
Free
✗
DeepSeek
Free
✗
Paste your text; the prompt updates automatically
The prompt below updates as you type or change the Tujia variety. Copy the entire prompt into your AI chat window and send it.
Copy the entire prompt from the box below, paste it into your AI chat window and send it.
Paste into your AI tool and send
Copy the entire prompt above, open your AI tool’s website, paste it into the chat window, and send it.
Compare the translation with the original. Check that every sentence is covered, the people and actions are right, and quantities and meanings such as ‘not’, ‘cannot’ and ‘not yet’ are preserved. For longer texts, translate one section at a time.
If you only get a word list, parts of the story are missing, or long passages remain in Chinese, follow up with:
Check the original for omissions, look up the missing words and examples, and provide the complete translation.
For a word-by-word analysis, ask: ‘Add an interlinear gloss with three rows: tone-marked romanization, IPA and word meanings.’
FAQ
Why did it translate only part of the text?
The AI may stop at an unfamiliar word, fail to load the dictionary, or leave out details even when its answer looks complete. Use the follow-up prompt above to check each sentence. Point out the missing sentence or meaning so it knows what to complete.
A missing lookup can often be resolved by rephrasing: try ‘a person walking’ for ‘passerby’, ‘smell’ for ‘fragrance’, or ‘something to drink’ for ‘beverage’. Preserve the original meaning. A failed lookup does not establish that Tujia lacks an expression.
Why are there Chinese words in the translation?
Tujia has Chinese loanwords. A translation may also borrow Chinese where a suitable native word has not been found. Personal names, place names and cultural terms may retain their Chinese spelling.
The lookup may also be incomplete. Ask the AI to check synonyms, phrases and examples for a suitable Tujia expression, and list any temporary Chinese insertions after the translation.
What if it cannot open the dictionary?
Give the AI the error message and ask it to try another reading method. The full dictionary in the repository is compressed. A code tool can download and decompress it; the repository also provides readable shards.
Resolve the reading error by downloading and decompressing the .gz file, or use the repository's readable dictionary shards. Once you can read actual entries, continue translating.
Why is it still searching without giving a translation?
Ask for the words it has found and the sentences already translated, then have it finish the section. Tool errors should be resolved or handled with another lookup method. On first use, lightweight lookup can start immediately if an embedding model and index are not already available.
Show the words you have found and the sentences already translated, explain where you are stuck, then finish this section.
How do embedding search and lightweight lookup differ?
Embedding search finds candidate entries by meaning. Lightweight lookup rephrases concepts with synonyms or simpler words before searching the dictionary. Both require checking meanings and examples. They help retrieve evidence; the AI still translates the sentence.
An embedding program in the repository does not mean a model was used for this translation. Ask: ‘Which lookup method did you actually use? Show a result from this run.’
Can I translate languages other than Chinese?
Yes. Translate in either direction between Northern or Southern Tujia and any other language the model supports. Results depend on its ability in both languages. For poetry and Classical Chinese, ask it to explain the original meaning first so you can check its interpretation.
What if the romanization and IPA do not match?
Point out the word and ask the AI to check its romanization and IPA against the same source record. Sources may use different transcription systems. Ordinary translations use tone-marked romanization by default; request IPA or Ye-style spelling in your prompt if needed.
Why can a sentence be wrong when every word was found?
One spelling may have several meanings, and the first dictionary result may not fit the sentence. Ask the AI to reconsider the word in context, look for similar examples, and check who performs the action and who or what it affects. A dictionary match alone does not establish the intended sentence meaning.
Can I continue in a new conversation?
In the new conversation, provide the repository link, the Tujia variety and the original text. If tools were installed earlier, ask it to reuse the existing installation and dictionary. Include the previous translation and your corrections when revising.
Examples translate English, Classical Chinese, Japanese, Tibetan and Korean into Northern Tujia, and Tujia into other languages. Each includes the source text and an interlinear gloss PDF.
The ancient implement is paraphrased as a farming tool, and the hope is expressed as wanting to obtain something. The state name and “mock” are Chinese insertions.
The full-length story is retained. Unlisted terms, including Jizō and New Year’s Eve, are inserted in Chinese characters without invented pronunciations.