Key term renderings will now be used for training machine translation models by default if the target project has renderings added. For training the model, the terms will be matched with the renderings from the source project. If the source project does not have terms added, terms will be matched with renderings from the glosses for the same language as the source.

If the source project does not have key terms and the source project language does not have glosses, the team can still use the additional training data option to upload a CSV file with key term pairs for the source and project languages.