Microsoft Translator Plugin for Azure OpenAI & Tools for PromptFlow

Seattle—May 23, 2023—Today, Microsoft Azure Cognitive Services (ACS) Translator announces a) ‘Microsoft Translator’ plugin for Azure OpenAI, and b) Translator and Language Detect tools for PromptFlow in private preview.

‘Microsoft Translator’ plugin for Azure OpenAI

On customer prompts requiring translation, OpenAI GPT models do good general domain translation among languages which have good web presence. However, it doesn’t provide a simple way to get translation customized for your organization and domain terminologies. Also, GPT models have limited language coverage and translation accuracy drops for low resource languages.

Microsoft Translator Plugin for Azure OpenAI allows users to translate a source text between any of the 125+ languages and dialects supported by Translator service. The plugin will also get you customized translation by using custom machine translation model which you have built and published using Azure AI Custom Translator portal. Using Microsoft Translator plugin for AOAI you could have uniform translation experience across your organization, when you are translating with any of your application built with AOAI API, ACS Translator API, or any of Microsoft products.

You could configure the Microsoft Translator plugin with your own Azure Translator resource and credentials. Whenever the user sends a prompt request to AOAI, the GPT model will decide and invoke the Microsoft Translator plug-in as required.

In the example below, GPT understands that the prompt asked to complete the task in ‘Tamil’ language. Hence it invokes Microsoft Translator plugin and gets the response translated.

Translator and Language Detect tools for PromptFlow

Translator and Language Detect tools for PromptFlow helps you to get better responses for various prompts submitted to GPT in non-English languages. You need to configure your Azure Translator resource in PromptFlow to enable this ability. A typical App workflow is as follows:

• Preprocess the prompt before sending it to GPT.
  
• Preprocessing could include detecting the language of prompt text and translating it into English if required. Optionally determine what content of the prompt text requires translation.
  
• Submit the pre-processed prompt to GPT.
  
• Post process the response from GPT before returning final response.
  
• Post processing could include translation of GPT response into the original language of the prompt.
  

In the example below, the language of the prompt is detected. On detecting the language as Tamil, the prompt is translated into English before making a GPT call. The response from GPT is translated back from English to Tamil.

Continue reading...