BEGIN
Upload a manuscript
Select one or several page images from the home page. Files are ordered naturally by filename. A PDF is rendered page by page, and every rendered page becomes an ordinary manuscript page that can be segmented and edited.
.serh backupNew projects are stored in this browser on this device. ŞERḤ accepts each stored page image up to 100 MB; the total space available for all manuscripts depends on the browser, device, and remaining disk space. Private or incognito browsing may provide only temporary storage, so do not rely on it for long-term work.
DEFINE THE PAGE
Segment the page into lines
Segmentation tells ŞERḤ which pixels belong to each text line and establishes reading order. On a two-leaf photograph, first draw optional reading panes in the order they should be read—normally the right leaf and then the left. Next, create or generate line polygons and review them before approval.
Use four corners to isolate leaves, columns, or other reading areas.
Draw manually, import YOLO pre-segmentation, or run a model.
Move, reshape, delete, split, merge, and classify regions.
Enclose the letters, ascenders, descenders, and the majority of the dots and diacritical marks belonging to the line. Do not worry if a few marks escape the boundary, and it is fine for neighboring line polygons to overlap slightly.
Click Six-point line to draw a new segment in the six-point order shown above. Click Select & adjust when you want to move an entire polygon or drag any of its six points independently to refine its shape.
OPTIONAL ASSISTANCE
Choose a segmentation method
ŞERḤ line model
An exploratory Ottoman line-segmentation model trained on a small manually annotated collection composed primarily of Maliyeden Müdevver Defterleri and Mühimme Defterleri.
- Runs entirely in the browser
- No API key or credits
- First use downloads roughly 25 MB
- Results require scholarly review
Vision AI models
OpenAI and OpenRouter models can propose editable line polygons. Add the relevant API key in AI settings and choose a model in the segmentation panel.
- Sends only the requested page or pane image
- Uses your API account and credits
- Useful for unusual layouts and pane-by-pane analysis
- Always review before approval
Local subscription runner
The guided “Segment with my subscription” workflow prepares one bounded task for Codex, Gemini CLI, or Claude Code. Paste the supplied command into the selected local coding assistant and import its result.
- No separate API-credit workflow
- Guided first-run setup
- One page in, structured polygons out
- Imported polygons use the normal editor
Manually drawn, imported, ŞERḤ-model, and hosted-AI results all become ordinary editable six-point polygons. There is no separate annotation system.
Once every line is present, correctly ordered, and acceptably bounded, approve the page to save its reading panes and polygons and make it available in the transcription editor.
After approval is complete, click this button to leave the segmentation workspace and begin entering the diplomatic transcription line by line.
READ LINE BY LINE
Transcribe the manuscript
After approving segmentation, open the line editor. Select a numbered polygon on the image or use the line buttons to move through the page. Enter a diplomatic reading first; you can type Latin transliteration, Ottoman Arabic script, XXX, or an uncertain letter skeleton.
The editor keeps the page image and active line visible while you work. Line and page review tools use the active page—and the immediately preceding page when available—as context, but return changes only for the requested lines.
EDITORIAL ASSISTANCE
Normalize, check, and translate
Request suggestions for one line or the whole page. Suggestions remain pending until you accept or ignore them. Accepted suggestions retain reconstructed Ottoman spellings for dictionary lookup.
Generate literal and readable translations for a line or an entire page. Keep a separate User tab for your own preferred wording and notes.
Send all active lines together as one bounded review task to Codex, Gemini CLI, or Claude Code, then import normalization suggestions, Ottoman spellings, and translations.
LEXICAL CHECKING
Dictionary lookup
Words in the active transcription become clickable lookup chips. ŞERḤ uses local lemmatization and conservative suffix analysis for Osmanlıca Sözlükler. After AI normalization has saved an Ottoman-script headword, you can also search LexiQamus with an institutional subscription or conduct more refined searches of Osmanlıca Sözlükler.
ḫazīneخزینهmezbūrمزبورaḳçeاقچه
AI-reconstructed spellings support lookup; they are not claims that every letter is visibly certain in the manuscript.PRESERVE THE PROJECT
Save, export, and continue later
Transcriptions, normalized text, translations, notes, and accepted editorial work save after a short pause.
Use Approve page & continue to save reading panes, line order, and six-point polygons.
Export a .serh file containing the page images and complete editorial state. Selected manuscripts can be downloaded together in one ZIP.
Export segmentation and transcription material when you want to reuse corrected pages for HTR or line-segmentation training.
Export a new .serh backup after roughly every half hour of active work—or immediately after a substantial segmentation or transcription session. Device storage belongs to one browser profile and can be cleared by browser settings, private-mode closure, or device loss. The .serh export is the portable copy of your work.