Multiple Label Sets

We are aware that one project does not always have one type of labeling. For example, a project can require both POS-tagging and NER labeling. Datasaur can support multiple task types too.

Multiple label sets allows you to apply more than one type of labeling to a single project. Before we start, you can see the detailed information about label sets here. Note: multiple label sets can currently only be applied for multiple token-based label sets. More types coming soon!

Create Multiple Label Sets

Creating multiple label sets is similar to creating a label set. You can create label sets manually or upload a label set file via a Custom Project and the Labels Extension.

Multiple Label Sets via Custom Project

On Step 3, Datasaur will automatically create Untitled Label Set 1. You are allowed to rename, add, remove, and reorder label sets.

Adding a new label set can be done by clicking the + icon in the Label Set section. Datasaur allows up to 5 label sets maximum for a single project.

Label set in Custom Project

Multiple Label Sets via Labels Extension

After the project is successfully created, you still will be able to change, add, and remove label sets.

Adding the label sets can be done by clicking the dropdown, then choosing the desired label set. You can create the label set manually or upload the label set file there.

If you would like to edit a label set, click the dropdown and select the desired label set.

After you select a label set, you can click the triple-dots. You can replace the selected label set by uploading a new one or choosing one of our built-in label sets.

Note: deleting the selected label set will remove any labels from that set in the current project.

Active Label Set

Active Label Set refers to the currently applicable label set. It can be set or changed in the dropdown above the project.

For example, let's say you have a POS label set and a NER label set. After labeling the POS, you can switch the apply NER labels by selecting the appropriate label set from the label set dropdown.

Choose Default Ner

Now the NER label set is active in your project. POS labels will turn to gray because they are not active. You can now apply NER labels to the document.

Default NER label set is active

Multiple Label Sets in a Token-based Project with Arrows

Sometimes we require a project that combines token-based label sets with relational arrows. In some projects, those arrows may also require an additional label to specify the relation.

Creating a token-based project with arrows can be done by checking the setting for Allow arrows to be drawn between labels on Step 3. You can add both token label sets and arrow label sets in the Label Set section.

We're going to create a dependency parsing project

Note: if you don't want to label the arrows, an arrow set is not required.

After the project successfully created, your labeling interface will look like the screenshot below.

There are two label set dropdown boxes!

The second label set with an arrow icon indicates the label set to be applied to the arrow labels. Click the dropdown and select the appropriate label set.

Choosing a label set for the arrows

You will be now be able to draw arrows and apply the corresponding labels.

Arrow labeling

Export

Projects with multiple label sets can be exported to .json_advanced and .tsv_non_iob.