Does LightTag work with RTL languages such as Arabic and Hebrew ?


Does LightTag work with CJK languages ?


Does LightTag support document classification e.g. labeling an entire document instead of individual words ?

Yes! You can define a task to classify individual examples. You can specify if you'd like exactly one class applied to a document or multiple classes.
For example, you can specify a task to label product reviews as Praise, Insult or Unknown.
Alternatively, you could specify a task to label recipes as Dairy, Vegetarian, Kosher and Spicy.

Can we annotate subwords and multi word expressions ?

Yes! Sometimes a picture is worth a thousand words.
Subwords, multi word expressionsLightTag makes no assumptions about how your data is tokenized, you can easily label subwords, or multiple phrases.

Does LightTag support nested annotations ?

We've decided not to support this feature directly. You can achieve the same thing using our relationship annotation features, and we think that is the right way to annotate.
LightTag helps teams prepare their data for machine learning, and the bulk of machine learning algorithms can't output "nested annotations". Our job is to make it easy for you to take your annotations and plug them in a model, so it doesn't make sense to support something that doesn't align with that.
With that in mind, we do support annotating relationships

Do you support relationships, coreference structures phrase structure grammars or dependency grammars ?

Yes! LightTag supports relationship annotation via a drag and drop interface. For the non-linguists, you can drag and drop annotations and define the way they relate to each other.
For the Linguists out there, LightTag supports any relationship annotation you can express as a tree (but not a DAG) including phrase structure grammars (with non-terminals that don't appear in the text) as well as dependency style annotations.Annotating relationships

We want to ensure data quality by having multiple people label the same document. Can LightTag help us do that ?

You can tell LightTag you want n={1,2,3..100} annotators to work on each example. Every time a labeler logs in, LightTag will assign them the next document to label based on what you've defined.
Your labelers never have to search for a particular file, and you don't have to manage who needs to do what work. LightTag does it for you automatically.

We need to divide our labelers into different teams and ensure that team A only works on data X. Can LightTag help ?

You can define teams in LightTag. When you you define a task, you can specify which teams are allowed to work on it.
For example, if you had a team of Chinese lawyers and another team of Irish doctors, you could divide them into two teams and LightTag would ensure that only Chinese lawyers work on chinese legal texts, while Irish doctors would only work on Gaelic radioligy reports.

Data & Privacy

Does it come on premise ?

Yes it does

What is your data policy?

In a nutshell, your data is yours , we don't touch it, use it, sell it or see it.
The details are in our privacy policy.

How is Our Data Stored ?

Users of our community version have their data stored on their local machines, never hitting our servers.
Users of our SaaS offering have their data stored in a database on AWS.
Every SaaS customer's data is stored on a seperate and private database
Our On-premise customers naturally host their data on premise.

Does LightTag backup our data ?

Of course! LightTag takes autoamtic daily snapshots of your data and annotations. These snapshots are encrypted and stored for seven days. Longer retention periods are available.


I'm a student or academic, can I have a free License ?

LightTag is commited to helping academia flourish and move NLP and it's applications forward. Our tools are 100% free for academic use.
Reach out or read about our discounts

We're a poor startup, can we have LightTag free until we grow ?

Probably not free, but we're happy to work with you and see you grow. Talk to us

Can we have a free trial ?

Alongside our fully featured demo LightTag offers a two week free trial to everyone. You can cancel anytime.

LightTag's SaaS pricing is for monthly active user. What is the definition of an active user ?

LightTag's pricing is a pay-for-what-you use model, designed to let you scale your annotation efforts up or down as needed. An active user is a user account that has submitted annotations during the calendar month.
For example, Alice and Bob start using LightTag in January and both submit work. Then their were two monthly active users.
In February, Alice continued working but Bob went skiing and submitted no annotations. In February their was one active user.
In March, the team decided to double down on labeling and brought in Charles and Danny who both submitted work along with Alice and Bob. In march their were four active users.

Automation & AI

Can LightTag learn from our annotators and provide suggestions automatically ?

LightTag's suggestions in actionLightTag's payed offerings include machine learning models that will learn from your annotators as they work and provide suggestions.
We provide two models, the first is based on multi-armed-bandit algorithms and provides high precision suggestions fast.
We additionally offer a deep learning model which generalizes to unseen examples and leverages context. This feature is in beta contact us if you'd like to join the beta.

We have a pre-existing model/dictionary/regular expressions that we want to use as suggestions to our annotators. Can we do that with LightTag ?

Yes. You can upload suggestions to LightTag via the API. Docs.
When you define a task, you can specify which source of suggestions LightTag should use. LightTag will present these suggestions to your annotators and record which suggestions they accept or reject.

Can we use the model's that LightTag learned instead of building our own ?

We can expose the models that LightTag learns via an API which you can apply on your unlabeled text. Contact Us for details.

Does LightTag support Active Learning ?

It depends on how you define Active Learning. LightTag certainly learns from your annotators as they work reducing their workload and creating a useable model. However we don't use "by the book" Active Learning. We've written bout why Active Learning for NLP isn't the best idea.

How many annotations per hour (or other time unit) are typical?

That depends on many factors, notably how hard the task is and how good your annotators are.

Does LightTag provide analytics about the performance of individual annotators ?