Language Studio and Azure Machine Learning Unification

This post has been republished via RSS; it originally appeared at: New blog articles in Microsoft Community Hub.

In May 2022, custom named entity recognition and custom text classification were announced as generally available features under Language Cognitive Services. The features allowed you to create your own custom models to classify and extract domain specific information from your documents. A key component of building out a quality model lies in the labels and data you provide.

 

We have listened to your feedback, and today we are delighted to unveil our exciting unified story with Azure Machine Learning Studio’s for Text Named Entity Recognition and Text Classification’s labeling projects. You can now leverage the powerful labeling experience in the Azure Machine Learning Studio, combined with the simplicity of the authoring features within the Language Studio. These services offer different Studios as their labeling experience with different labeling formats, making it impossible for users to migrate from one service to the other. The goal of this integration is to allow for easier collaboration and switch between the labeling activities in these separate studios, resulting in a better and more flexible experience where users can easily jump back and forth between the two studios to label and continue their customization path in any of the studios depending on their use cases.

 

This integration will be very beneficial for users for multiple reasons:

  • Language Studio users can now leverage the Azure Machine Learning labeling and outsource any labeling tasks to external labeling vendors through the Azure. 
  • Language Studio users can now benefit from the multi-labeling feature in Azure Machine Learning Studio
  • Language Studio users can choose to continue to label in Azure Machine Learning while continuing the authoring flow (training, evaluation, and deployment) on the Language Studio.
  • Users can seamlessly switch their labeling experience from the Language Studio to Azure Machine Learning Studio and vice versa, with just a click of a button, keeping the class/entities along with the labels intact.
  • Similarly, Azure Machine Learning users can also choose to continue labeling in Language Studio while continuing the authoring flow (training, evaluation, and deployment) on Azure Machine Learning Studio.

 

The integration always starts from the Language Studio side, where users can choose to connect with Azure Machine Learning and create a new labeling project in Azure Machine Learning using the same dataset the Language project was created with. Projects on both services will be connected and users can choose to switch and label from either studio, maintaining the labels created.

Let's walk through the steps of connecting your project. Remember, there are some pre-requisites and limitations that you should be aware of before starting your connection. Learn more about it in our documentation.

 

First, you will find in Language Studio a new “Azure Machine Learning” tab is added in the Activity pane on the right.

sarakandil_0-1684425769292.png

 

 

Once you click on the “Connect to Azure Machine Learning” button, a wizard appears that explains the integration’s benefits, limitations and links to the external documentation. To simplify the experience and the connection , we detect some settings and show them to you, asking you to select their workspace and enter a name for the new Azure Machine Learning project that needs to be created.

 

sarakandil_3-1684425915989.png

 

sarakandil_4-1684425915993.png

 

 

One of the biggest advantages that Language Studio users may find beneficial is the ability to outsource the labeling to a vendor company.

 

sarakandil_10-1684426347713.png

 

 

After the user finalizes the connection, a new Azure Machine Learning labeling project is now created and is connected to this Language Studio project. Users cannot label in the two experiences simultaneously, and with that, we introduced two new concepts: “Labeling activity” and “switching”. If the “labeling activity” in one experience is enabled, the “labeling activity” in the other experience is by default disabled. To enable the labeling activity in an experience, you will need to “switch” the labeling activity to it. This will import the labels from Language Studio to Azure Machine learning or vice versa. For each state, we have guides in both studios to explain and guide the users on their current setup. Below are the possible states in the Language Studio.

 

sarakandil_7-1684426324012.png

 

sarakandil_8-1684426324017.png

 

sarakandil_9-1684426324019.png

 

 

To further simplify the experience, we made sure that we do not block users who are labeling in Azure Machine Learning from other functions available in Language Studio. Language Studio users can start a training job with the latest labels they have completed in Azure Machine Learning. They do not need to enable the labeling activity in Language Studio to continue creating or deploying the model.

 

sarakandil_6-1684426294085.png

 

In the Azure Machine Learning Studio, any connected project will show this “Language Studio” tab where users can again “switch” their labeling activity back to the Language Studio.

 

sarakandil_5-1684426271347.png

 

 

We’re excited to begin our journey collaboration with Azure Machine Learning! Try it out, and let us know what you think!

 

Leave a Reply

Your email address will not be published. Required fields are marked *

*

This site uses Akismet to reduce spam. Learn how your comment data is processed.