Advice and answers from the LightTag Team

Go to LightTag

Making Your First Annotation Job

How to set up LightTag to work for you

Written by Tal Perry. Updated over a week ago

If a picture is worth a thousand words, how many words is a video worth ? We made a video showing you how to get started. If you don't want to watch, we'll explain how to do it in less then a thousand words here.

Key Concepts

A good relationship is based on communication and a common language. We want this relationship to work, so we ask you to take a minute to learn our language, it only has 3 important words.


Your here to annotate data. If you put a bunch of data together it becomes a Dataset.

In LightTag, datasets are made of a bunch of examples, the things you want to annotate. An example can be short, like a tweet, or long, like a contract.

The punchline is that to annotate data you need a Dataset.


You have examples in a Dataset, and you want to annotate them. But what exactly do you want to annotate them with ?

The collection of concepts that you'll be annotating with is a Schema.

Now, different people want to annotate in different ways. Some people want to annotate words and phrases, like this:

Span annotation with tags in LightTag

You see those colors, with the label above the word ? Those are Tags. Tags are the concepts we use to annotate words and phrases.

But sometimes, you don't care about the individual words, you want to annotate the entire Example. Like this tweet that has positive sentiment:

A tweet with positive sentiment annotated in LightTag

In that case, Positive is a Class (from classification).

So to recap, your Schema contains the Tags and Classes you'll use to annotate Examples in a Dataset.


A job is the work you want to get done. It's basically a way to say "Hey LightTag! I want this Dataset annotated with concepts from this Schema".

Like Britney Spears, we were born to make you happy, so we'll make sure that Job happens. You do need to bring the people who will do the work, but when you tell LightTag about your job we'll take care of assigning it to your team.

Enough Theory, What Do I Need To Do ?

So it's pretty simple. You just need to make a Job in LightTag, which means you need to have some data to annotate and know what the schema that you want to use is. Then you need to put those into LightTag.

Click Manage on the App Bar

Choose Quick Start

Press the button after you read the text

Define Your Schema

Give it a name, add tags and classes then save it

Upload Data

You can't label data without, well, data. So either upload it or paste it.

If you're uploading a file, we support JSON arrays or CSVs and TSVs.

But I don't have a file in that format

Not a problem. You can paste as much text as you want. We'll split it into sentences and make a Dataset of sentences for you.

Configure your Dataset

There are a lot of columns in your CSV or fields in your JSON. We don't know which one you want to annotate, so you need to tell us.

We call that field the content field. You can choose one that is all text, don't worry we'll check it for you. Heaven forbid we make you work hard validating your own data. It's bad enough we don't label it for you.

Define the Job

You got this far, you are amazing!

All that's left is to give your job a name. Don't be lazy, give it a meaningful name that you'll remember, it's for your own good. Once you've named it you can submit it.

Now, you may have seen some advanced field in the UI, they are clearly labeled Advanced.

You are not advanced yet, so don't touch those. We'll get to them

Make the Job Top Priority

You only have one job, but we want you to establish good habits early on. When you make a job you need to prioritize it. Now, you could be like your boss and say everything is top priority, or you could think about it and prioritize your teams work. How do you feel about your boss's prioritization ?

For now, it doesn't matter, you only have one job. But it's never too early for good habits


You are ready to label data and will be taken to the labeling screen