Pangur docsTranscribe a manuscript
A transcription is the text of a facsimile's leaves, kept line by line against the image. This page walks through a first transcription, with or without a model's draft, and then describes the transcription desk's controls, versions and reading rules.
A transcription in Pangur is the text of a facsimile's leaves, kept line by line and tied to the image, so that each line of text knows which line on the leaf it transcribes. Every line has two readings. The Diplomatic reading keeps the text as written, with its abbreviations and spellings, and the Expanded reading gives the same line with the abbreviations expanded. These are two readings of one transcription rather than two transcriptions.
Once saved, a transcription becomes the item's text, which is what Ask, Analysis and the reading room read, unless the item also has an imported edition, which takes precedence. How an item's text is chosen is explained in How Pangur reads your sources.
This page walks through a first transcription and then describes the desk's controls. The models involved, what they cost and their daily limits are covered in Reading pages with a model.
Before you start
You need an item with a facsimile whose images are ready. The facsimile can be a repository's digitisation (Link a digitised facsimile), your own photographs (Photograph a manuscript), or a PDF or image attached to the item, which Pangur renders into leaves. The item's Transcribe tab lists each facsimile that is ready. While images are still being prepared it says so, and if the item has no images at all it suggests importing a file, linking a IIIF facsimile or uploading photographs.
The model-assisted steps also depend on two server features. Line detection must be set up by the server's administrator, and so must reading with a model. If either is missing, its button is disabled and says why. Model reading needs line detection as well, so without line detection the Transcribe with AI button is disabled too.
Opening the transcription desk
- Open the item and choose Transcribe.
- Under the facsimile you want to work on, choose Open editor. The desk opens in a new browser tab.
From the catalogue, a manuscript's Transcribe button opens the same desk, and if the manuscript has no library item yet, Pangur makes one first.
The Transcribe tab itself only summarises the work. It shows the Current draft with the time it was last edited, and each saved version with its label, whether it was made by a model ("AI"), by hand ("manual") or both ("mixed"), and an Open button.
A first transcription, step by step
1. Detect the lines
Go to the leaf you want with the leaf navigator and choose Detect lines. Pangur's server finds the baselines and line regions on the image with Kraken 7, and when it finishes each detected line has an empty line of text beside it. Line detection uses no language model and has no daily limit.
2. Read the leaf with a model, or supply the text
There are now three ways to fill the lines. Transcribe with AI sends the leaf to a vision model, which reads every line twice: a first pass reads each line, and a second checks each line against its image. When it finishes, the desk tells you the draft is ready to review. If the lines have not been detected yet, the model's reading detects them first. Use a supplied transcription takes a transcription you already have, one manuscript line per line of text. You paste it, choose Align to this leaf, and Pangur places the lines on the detected lines in reading order, keeping any extra lines without linking them to the image. The leaf's lines must be detected before you can align. Or you can simply type each line yourself in the editor.
Before the first model reading of a catalogued manuscript, Pangur may ask About this hand, which is the Language and Period of the writing (English, Latin or Scots, say, and the 18th century or 1710). These help the model expect the letterforms and abbreviations of the hand. Both are optional. If you fill in either, the button reads Save and read, and Pangur saves your answers on the manuscript's description before starting the reading. Skip starts the reading without them. Pangur asks once per manuscript in each browser session, and only when the catalogue description gives neither.
3. Correct the draft
Choose Edit transcription and correct the lines against the image. Clicking a line highlights its region on the image, and selecting text highlights the region it comes from. Switch between Diplomatic and Expanded to correct each reading. Switching shows the other reading and does not read the leaf again.
The editor protects the line structure, so you cannot type a new line break or edit text of a type that is hidden. To change lines, use Line actions in the markup menu: Split at caret, Join selected lines or Column.
If you later choose Read again with AI after correcting some lines, Pangur keeps your corrected lines as they are, and shows the model up to three of them as examples of the hand.
4. Save a version
Your edits to the Current draft save themselves a little over a second after you stop typing, and the status beside the desk moves from unsaved to saved. To keep a fixed copy of the transcription as it stands, choose Save version…, give it a label, and save. A version cannot be changed afterwards (see Versions and the current draft).
When you save, Pangur re-indexes the main text of the transcription, so Ask and search find it straight away. Page furniture such as running titles and catchwords is left out of that index.
The desk's layout
The desk shows the leaf's image beside its transcription. Choose a layout in Editor layout: Split horizontal (the default), Split vertical, Image only or Text only. Dragging the divider between the panes changes the split, and your browser remembers the layout. On a phone the panes are stacked.
The bar above the panes has the leaf navigator, a Facsimile list when the item has more than one ready facsimile, the Export menu, and the layout switch. You cannot change facsimile while a save is in progress.
When a leaf has several photographs, you can view another one from the image pane. Viewing it is temporary, and the model always reads the leaf's default photograph.
Reading options and find
Reading options has Reading scope, which is This leaf or Whole manuscript (This page or Whole book for printed books), and Flowing text, which joins lines into paragraphs for reading. Visible text types chooses which kinds of text are shown, although the main text and page references are always shown. Find in the manuscript (or Find in the book) searches every leaf and turns to each match.
Marking up the text
Right-click in the transcription, or press Shift+F10, to open the markup menu. Text type sets what a line or selection is: Main text, Numbering, Quire marks, Running title, Margin text, or a type you define. Scholarly markup marks a selection as an Addition, a Deletion, Uncertain, Illegible or Supplied, and Remove marks takes the markup off. Line actions splits and joins lines or sets a column. Copy with reference copies the selection with the item's title, repository, shelfmark, leaf label and a link. When the model offered alternative readings of a line (up to two), they appear in this menu to choose from.
Cutting is disabled in the editor, so that text is never moved without its reference. Use Copy with reference and then delete the selection if you need to.
Once you set a text type or column on a line, later model readings leave it as you set it.
Text types and guidance for the model
Text types & AI guidance lets you add text types (names up to 80 characters) and tell the model what to look for in each (up to 2,000 characters). Changes guide future readings, and lines you have already classified keep their types.
Reading rules
Reading rules are standing instructions for every future model reading of this manuscript, such as how the scribe writes a particular abbreviation, or that a word should be expanded in a particular way. They can be up to 4,000 characters long. Reading rules belong to the manuscript and are kept outside versions, so saving or restoring a version does not change them.
Versions and the current draft
The Current draft is the transcription you are editing. It saves itself, and there is only one for each facsimile. A version is a named, fixed copy of the draft's pages, its source and its text types at the moment you saved it. A version does not include your reading rules or your choice between the diplomatic and expanded readings.
To look at a version, choose it in Version. The desk shows it read-only. To make that version the current draft again, choose Restore this version.
Before it restores a version, Pangur saves the current draft as a version named "Before restoring version" followed by the number of the version you restored. To undo the restore, restore that version.
When an assistant saves a reading over the current draft through MCP, Pangur first saves a version of the leaf named "Before agent reading of leaf" followed by the leaf's number, so that its reading can be undone.
When saving fails
If someone else, or another of your own browser tabs, saved the same transcription since you opened it, Pangur refuses the save with "This transcription changed elsewhere. Reload before saving." After any failed save, autosave stops and the desk tells you that your edits are still there. Choose Retry save once the problem is resolved. The desk keeps 100 steps of undo for each leaf (⌘Z or Ctrl+Z, with redo on ⇧⌘Z, Ctrl+Shift+Z or Ctrl+Y), and the history is cleared when you move to another leaf.
Numbering found on the leaf
When the model reads a folio or page number written on the leaf and it differs from the label in the catalogue, the desk points this out, quoting both. Choose Use this to give the leaf the label found on it, or Dismiss to keep the catalogue's. Pangur does not change the catalogue's numbering without your choice. Labels are explained in Numbering systems.
Who can transcribe
In an organisation's project, a member with the viewer role can open the desk and export a transcription, but cannot detect lines, read with a model, edit, or save or restore versions. The desk does not hide these buttons from viewers; Pangur's server refuses the change instead (Project members and roles).
The transcription desk needs a connection, and has no offline mode on the web or in Pangur for Mac. The iPhone and iPad app has a transcription view, but some work, such as reviewing the alignment of markup on a line, is done on the web, and the app says so when it applies.
Exporting
Export in the desk's bar downloads the transcription as TEI, PAGE XML, ALTO or plain text, with or without page images. The formats are listed in TEI texts and transcription exports.
Link a digitised facsimile
You can connect a manuscript or book in your library to a library's or archive's online digitisation, so that its leaves open in Pangur's reader and transcription desk. This page explains what you need, how the link is kept, and what happens when the images do not load.
Reading pages with a model
Transcribing a page image is two jobs, finding the lines and reading them, and Pangur does them with different software. This page explains which parts use a vision model, what the daily limits are and whose they are, and how a connected assistant can do the reading itself.