Skip to main content
Sentence analysis is the app’s core feature. After you enter some Japanese, the app splits it into tokens (words). It labels each word with its furigana (kana reading), romaji, and part of speech, then generates a full translation below.

​
Analyze Japanese text

1

Enter the text

Type or paste Japanese into the input box. You can also click Sentence of the day or an item under Recent on the home page to fill in the text directly.
2

Click Analyze

Click Analyze. With streaming output on, tokens appear as they are generated.
3

Read the results

The results area shows readings word by word. When the analysis finishes, the translation area automatically translates the whole text.
While the analysis is running, click Stop analysis to cancel the current request. When it finishes, click Edit text to change the sentence, or click New text to clear it and start over.

​
What the results include

Each token includes three kinds of information: The legend below the results lists the part-of-speech groups that appear in the sentence: Noun, Verb, Adjective, Na-adjective, Adverb, Adnominal, Conjunction, Interjection, Particle, Auxiliary verb, and Other. Punctuation has no underline.

​
Show or hide readings

Two toggles sit at the top of the results area:
  • Show furigana: Toggles the kana above kanji.
  • Show romaji: Toggles the romaji below words.
With the readings hidden, you can test your reading on the original text, then turn them back on to check.

​
Full translation

When the analysis finishes, the app automatically translates the whole text into your current interface language. The translation area offers these actions:
  • Copy: Copy the translation.
  • Translate again: Generate a new translation of the same text.
  • Collapse / Expand: Fold or unfold the translation area.
The translation language follows the interface language. When you switch to English, translations, word definitions, and AI assistant replies all switch to English. For details, see Settings.

​
Long texts are analyzed in chunks

Long passages are split automatically into chunks of about 280 characters each, up to a maximum of 420 characters. The app breaks at sentence endings such as periods, question marks, and exclamation marks, and avoids splitting paired brackets and quotation marks.
  • With streaming output on, up to 3 chunks are requested at the same time. With it off, chunks are analyzed one after another. Either way, the results are joined in the original order.
  • If a chunk doesn’t return a complete result, the error message tells you which chunk failed, for example “Chunk 2/5 returned an incomplete analysis. Please try again.”
  • The app checks that each chunk’s tokens join back into the original text. If they don’t, it asks you to try again, so no characters are dropped or added.

​
Paste from web pages and Markdown

When you copy text from a web page or a Markdown document, the app cleans up the formatting automatically:
  • Removes HTML tags, scripts, styles, and hidden elements, while keeping paragraphs and line breaks.
  • Removes Markdown markup, while keeping image alt text.
  • Removes link URLs, keeping only the link text.
If you type text that contains an https:// link, the app keeps the link intact as a single symbol so the model doesn’t split it or add readings.

​
Streaming output

Streaming output is on by default, so analysis, translations, and definitions appear in real time. If your connection is unstable and results often get cut off, you can turn it off in Settings to get the complete result all at once.

​
When analysis fails

Error messages appear in your interface language. Raw error messages from the model provider, such as HTTP 429: upstream quota exceeded, stay in their original wording to help you troubleshoot.
For more issues, see Troubleshooting.