I have newly trained a Google AutoML model (English to Traditional Chinese (zh-TW)) with legacy format, which uses AutoML API instead of the new Cloud Translation API. 

After some tests on the latest Trados 2022 I found Google Cloud Translation Provider returns translation from un-trained Google Translate instead of translation from trained models.

I also tested the same model using the same JSON key on MT Enhanced Trados Plugin, and the returned translation is from trained model as expected.

Below are the steps to reproducing the issue I encountered:

1. Prepare an GCP AutoML model with the following language pair: English => Traditional Chinese (zh-TW)
2. Install both MT Enhanced Trados Plugin and Google Cloud Translation Provider on Trados Studio 2022 (version: 17.0.6.14902)
3. Create a translation project in Trados 2022 with language pair English => Traditional Chinese (zh-TW)
4. Using the same .json file from Step 1 in both MT Enhanced Trados Plugin and Google Cloud Translation Provider
5. Open the translation project in Step 3 and review the NMT translation from both MT Enhanced Trados Plugin and Google Cloud Translation Provider
6. The translation from the two plugins will be different.

One thing worth mentioning is that when I am first loading the JSON key file into Google Cloud Translation Provider, the following two error messages occurred. 

While MT Enhanced Trados Plugin can load the JSON key file without any problem.

1. Reading failed. The selected json file has less fields than it is supposed to have. Please download the file again and use an unedited file.

2. Authentication failed. Could not authenticate the user with the ProjectID provided in the JSON file.

 I found a workaround is to delete the last line "universe_domain": "googleapis.com" in the JSON file, after deleting the line,  Google Cloud Translation Provider can load the JSON file without error. 

Regards,

William