CaptionHub saves users' work continuously, but there are occasions when it takes a snapshot. A snapshot is simply what a particular set of captions looked like at that point in time.


Events which trigger a snapshot include:

  1. Approval
  1. Handover
  1. Machine translation or transcription
  1. If the source text has changed

To access snapshots, choose the "List snapshots" option from the drop down menu on the Edit page. Here, users can view the differences between sequential snapshots. When you view a snapshot, two versions of the caption set are presented, as plain text. On the left, the state before the user handed it over; on the right, the state after they did. Additions are highlighted in green; removals highlighted in red. Grey text is unchanged between versions.


All users can manually create a snapshot if they wish, using the Create snapshot button. This acts as a manual 'Save' button, effectively.


Superusers and producers are also able to restore from a snapshot, if necessary.

We use Jaccard similarity to calculate the differences between snapshots. It works on a technical level by concatenating all of the captions into a block or text, then it chunks the characters into every consecutive three character sequence, then takes those sets from text one and txt two and sees the percentage of sets that are the same. So it’s effectively a character-level difference, not individual captions.
Did this answer your question?