Skip to main content

Bulk annotation mode

For projects with simple classifications (single-choice labels, yes/no judgements), Bulk mode lets annotators rapid-fire many tasks at once with a single click per task.

Useful for:

  • Single-choice text classification (sentiment, toxicity, spam)
  • Yes/no audio judgements ("is this clean speech?")
  • Image quick-sort ("is this a cat?")

Not useful for:

  • Tasks with regions / spans / bounding boxes (any <Labels>-style control)
  • Tasks with free-text inputs (TextArea)
  • Long-form transcription

Enabling bulk mode​

Project → Settings → Annotation → Enable bulk mode.

Toggle on. A Bulk tab appears in the project's navigation, alongside the standard one-task-at-a-time labelling flow.

Requires the project's labelling config to be a single <Choices> control plus its object tag — no region tags (<Labels>, <RectangleLabels>, …) and no free-text inputs. The toggle is greyed out, and the bulk view refuses to mount, for any other config.

What annotators see​

The bulk view shows a page of tasks as compact rows — text clamps to a few lines, images render as thumbnails, audio clips get an inline player.

Each row carries:

  • The data (text / image / audio clip)
  • The choice options as pill buttons
  • A status chip (blank / Labelled / Skipped)

Click a choice — the task is instantly labelled, the chip flips, and focus advances to the next row. When a page is exhausted, a Load next batch button pulls the next set, with a running "N remaining for you" count.

Keyboard: 1–9 apply the matching choice to the focused row, S skips it, and the arrow keys move focus.

Submit happens per task as soon as the annotator picks a choice — there's no "submit batch" step. Clicking a different choice on an already-labelled row re-saves it.

Configuring batch size​

By default, bulk mode shows 20 tasks per page. Annotators paginate as they finish.

Configure at Project → Settings → Annotation → Bulk page size: 1 – 100 tasks per page.

Larger pages mean less pagination overhead but more memory + more cognitive load. 20 is the sweet spot for most projects; 50+ for text classification on phones with weak connections (fewer roundtrips).

Skipping in bulk mode​

Skipping works the same as single-task mode — the task is marked skipped, the chip shows Skipped, and the next row focuses.

What bulk mode does NOT do​

  • It doesn't bypass agreement / review settings. Bulk-labelled tasks still flow through whatever review workflow is enabled on the project. Reviewers see them in their queue just like single-mode annotations
  • It doesn't bypass audio QC. Audio tasks labelled in bulk still run through the audio QC pipeline on submit
  • It doesn't bypass auto-suspend rules. Throughput from bulk mode counts the same as single mode — abusing bulk to inflate task counts past quality is caught by the agreement-based rules

When NOT to use bulk mode​

If your project's task variance is high — some tasks are clear, others need a long look — bulk mode encourages annotators to move too fast through the ambiguous ones. Single-task mode forces a pause before each submit, which helps quality.

Bulk mode pairs well with:

  • Single-choice classification with low variance
  • Tasks where ground truth is fast to assess (under ~5 s/task)
  • High-volume corpora where the labour cost of per-task UI overhead matters

What's next​