Deploy a user-managed glossary

Azure Translator in Foundry Tools containers allow you to use many of the Translator's features within your own environment, making them ideal for meeting specific security and data governance needs.

Sometimes, when operating a container with a complex ingestion process, you may encounter the need to update sentence or phrase files. Because the default sentence and phrase files are encrypted and loaded directly into memory at runtime, you need a rapid engineering solution to allow for dynamic updates. You can address this need by implementing our user-managed glossary feature:

• To deploy the phrase​fix solution, you need to create a phrase​fix glossary file to specify that a listed phrase is translated in a specified way.

• To deploy the sent​fix solution, you need to create a sent​fix glossary file to specify an exact target translation for a source sentence.

• The phrase​fix and sent​fix files are then included with your translation request and read directly into memory at runtime.

Managed glossary workflow

Important

UTF-16 LE is the only accepted file format for the managed-glossary folders. For more information about encoding your files, see Encoding

1. To get started manually creating the folder structure, you need to create and name your folder. The managed-glossary folder is encoded in UTF-16 LE BOM format and nests phrase​fix or sent​fix source and target language files. Let's name our folder customhotfix. Each folder can have phrase​fix and sent​fix files. You provide the source (src) and target (tgt) language codes with the following naming convention:

   Glossary file name format	Example file name
   {src}.{tgt}.{container-glossary}.{phrase​fix}.src.snt	en.es.container-glossary.phrasefix.src.snt
   {src}.{tgt}.{container-glossary}.{phrase​fix}.tgt.snt	en.es.container-glossary.phrasefix.tgt.snt
   {src}.{tgt}.{container-glossary}.{sent​fix}.src.snt	en.es.container-glossary.sentfix.src.snt
   {src}.{tgt}.{container-glossary}.{sent​fix}.tgt.snt	en.es.container-glossary.sentfix.tgt.snt

   Note

   • The phrase​fix solution is an exact find-and-replace operation. Any word or phrase listed is translated in the way specified.
   • The sent​fix solution is more precise and allows you to specify an exact target translation for a source sentence. For a sentence match to occur, the entire submitted sentence must match the sent​fix entry. If only a portion of the sentence matches, the entry doesn't match.
   • If you're hesitant about making sweeping find-and-replace changes, we recommend, at the outset, solely using the sent​fix solution.

2. Next, to dynamically reload glossary entry updates, create a version.json file within the customhotfix folder. The version.json file should contain the following parameters: VersionId. An integer value.

   Sample version.json file

      {
   
       "VersionId": 5
   
       }
   
   

   Tip

   Reload can be controlled by setting the following environmental variables when starting the container:

   • HotfixReloadInterval=. Default value is 5 minutes.
   • HotfixReloadEnabled=. Default value is true.

3. Use the docker run command

   Docker run command required options

   docker run --rm -it -p 5000:5000 \
   
   -e eula=accept \
   
   -e billing={ENDPOINT_URI} \
   
   -e apikey={API_KEY} \
   
   -e Languages={LANGUAGES_LIST} \
   
   -e HotfixDataFolder={path to glossary folder}
   
   {image}
   

   Example docker run command

   
   docker run -rm -it -p 5000:5000 \
   -v /mnt/d/models:/usr/local/models -v /mnt/d /customerhotfix:/usr/local/customhotfix \
   -e EULA=accept \
   -e billing={ENDPOINT_URI} \
   -e apikey={API_Key} \
   -e Languages=en,es \
   -e HotfixDataFolder=/usr/local/customhotfix\
   mcr.microsoft.com/azure-cognitive-services/translator/text-translation:latest
   
   

Learn more

Create a dynamic dictionary Use a custom dictionary