Appearance
Running a Transkribus project with a team comes down to three things: share collections with the right roles, assign work clearly so people don't collide, and enforce consistency and review. Transkribus supports multiple users per collection with tiered permissions, version history per page, and shared models — but it does not lock pages, so your coordination layer (assignments and a status tracker) is what keeps a distributed team productive. Here is how to set it up.
How do you structure collections and roles?
Organise around collections as project units and grant roles per collection:
| Role | Can do | Typical use |
|---|---|---|
| Owner | Manage members, settings, all editing | Project lead |
| Editor | Transcribe, run jobs, train models | Core staff |
| Transcriber | Transcribe / correct text | Volunteers, students |
| Reader | View only | Stakeholders, reviewers |
Invite people by their account email and assign the least privilege that lets them do their job. Keep one collection per coherent body of material so permissions and models stay scoped sensibly.
How do you stop people editing the same page?
Transkribus does not hard-lock pages, so two people can open the same one. Prevent collisions with an explicit assignment scheme tracked outside the app:
text
Assignment tracker (shared sheet)
doc pages assignee status
Register_1 1-40 Amara in-progress
Register_1 41-80 Ben done
Register_2 1-60 Chen needs-reviewGive each transcriber non-overlapping page ranges and a status column. This single spreadsheet eliminates the most common team failure: duplicated work and silent gaps.
How do you keep transcription consistent across people?
Every transcriber brings their own habits, so a written transcription guideline is non-negotiable for teams. Cover the decisions that otherwise diverge:
- Abbreviations: expand or retain, and how to mark expansions.
- Letterforms: handling of
u/v,i/j, long-s, capitalisation. - Special characters and damaged/illegible text (
gaptagging). - Which structural and text tags to apply, and when.
Train everyone on it, and put it where they work. Consistency that one person manages by instinct must become an explicit rule for a group.
How does quality control work with a team?
Use a two-pass model and make status explicit:
text
Workflow states
untouched -> transcribed (pass 1) -> reviewed (pass 2) -> done
\-> needs-reviewA transcriber does pass one; a separate reviewer checks against the image and corrects, then sets done. For volunteer-heavy projects, also sample a percentage of finished pages to catch systematic errors (a misread letterform repeated everywhere) that page-by-page review can miss. Page version history gives you the audit trail to see who changed what and revert mistakes.
How do you share models across the team?
A custom model trained by one person can be shared within the collection so everyone recognises with the same engine — essential for consistent output. The healthy loop is: the team transcribes ground truth, an editor trains a model from it, the model is applied to the rest, and corrections feed the next training round. Centralising model training under one or two editors keeps versions clean and avoids a proliferation of half-trained models.
What coordination keeps a project on track?
- A kickoff that walks everyone through roles, assignments and the guideline.
- A shared tracker for page assignments and status.
- A regular check-in to surface blockers and recurring errors.
- A review queue so nothing reaches "done" without a second pair of eyes.
The platform handles storage, recognition and versions; your job is the human process around it. Get assignment and review right and a distributed team scales smoothly to tens of thousands of pages.
Key Takeaways
- Treat collections as project units and grant least-privilege roles.
- Transkribus does not lock pages — assign non-overlapping ranges in a tracker.
- A written transcription guideline is essential once more than one person types.
- Use a two-pass transcribe-then-review workflow with explicit status states.
- Sample finished pages to catch systematic, repeated errors.
- Share models within the collection so output stays consistent.
- Page version history is your audit trail; use it to review and revert.
Frequently Asked Questions
How do I share a Transkribus collection with my team?
Share at the collection level by inviting users via their account email and assigning a role. Everyone you add can then open the documents in that collection, with their permitted actions governed by the role you grant.
What roles can collaborators have?
Transkribus offers tiered roles per collection, typically owner, editor and a more limited transcriber or reader role. Owners manage members and settings, editors transcribe and run jobs, and limited roles can transcribe or only view without changing structure.
How do we avoid two people editing the same page?
Assign page ranges or documents to named transcribers and track them in a shared spreadsheet, since Transkribus does not hard-lock pages. A clear assignment scheme plus a status field prevents duplicated effort and gaps.
How do we keep transcription consistent across many people?
Write a short transcription guideline covering abbreviations, capitalisation, special characters and tagging, and train everyone on it. Consistency rules matter even more with a team because each person introduces their own habits.
Can we see who changed what on a page?
Yes. Transkribus keeps a version history per page, so you can see earlier states and who saved them, and revert if needed. Use versions as your audit trail during review.
How should quality control work with volunteers?
Use a two-pass model: a transcriber does the first pass, then a reviewer checks and corrects it, marking pages with a status like done or needs-review. Sampling a percentage of pages also catches systematic errors early.