Frequently asked questions for conversational language understanding
Use this article to quickly get the answers to common questions about conversational language understanding
See the quickstart to quickly create your first project, or the how-to article for more details.
Yes, using orchestration workflow. See the orchestration workflow documentation for more information.
Conversational language understanding is the next generation of LUIS.
For conversation projects, long training times are expected. Based on the number of examples you have your training times may vary from 5 minutes to 1 hour or more.
See the entity components article.
See the language support article.
Take a look at the recommended guidelines for information on improving accuracy.
When you train and deploy a conversation project in any language, you can immediately try querying it in multiple languages. You may get varied results for different languages. To improve the accuracy of any language, add utterances to your project in that language to introduce the trained model to more syntax of that language.
See the service limits article.
Unlike LUIS, you cannot label the same text as 2 different entities. Learned components across different entities are mutually exclusive, and only one learned span is predicted for each set of characters.
Yes, you can import any LUIS application JSON file from the latest version in the service.
No, the service only supports JSON format. You can go to LUIS, import the .LU
file and export it as a JSON file.
Yes, you can use orchestration workflow to orchestrate between different conversational language understanding and question answering projects. Start by creating orchestration workflow projects, then connect your conversational language understanding and custom question answering projects. To perform this action, make sure that your projects are under the same Language resource.
Add any out of scope utterances to the none intent.
You can control the none intent threshold from UI through the project settings, by changing the none intent threshold value. The values can be between 0.0 and 1.0. Also, you can change this threshold from the APIs by changing the confidenceThreshold in settings object. Learn more about none intent
Yes, only for predictions, and samples are available for Python and C#. There is currently no authoring support for the SDK.
Training mode | Description | Language availability | Pricing |
---|---|---|---|
Standard training | Faster training times for quicker model iteration. | Can only train projects in English. | Included in your pricing tier. |
Advanced training | Slower training times using fine-tuned neural network transformer models. | Can train multilingual projects. | May incur additional charges. |
See training modes for more information.
Yes, all the APIs are available.