Creating adapted language pairs

Create adapted language pairs using bilingual data to produce translations tailored to a specific domain.

Before you begin

Ensure that you meet the following requirements before you start adapting a language pair:
  • You have a model name for the new language pair that you will create.
  • Standard deployment:
    • You have added a training engine to the same host as the generic language pair that you will use as a baseline to create the new adapted language pair. You can create an adapted language pair only if you already have a corresponding generic language pair on the host.
    • You have started the training engine.
  • Kubernetes deployment:
    • You have added a training engine to the Kubernetes cluster by changing the default number of replicas for the training engines to 1. See Changing the number of training engines.
    • The generic language pair that you will use as a baseline to create the new adapted language pair is already deployed in the language pair storage. You can create an adapted language pair only if you already have a corresponding generic language pair in the language pair storage. See Installing new language pairs.
    • You have deployed the placeholder translation engine. The engine can be deployed after training the adapted model but must be available before deploying it. See Understanding placeholder language pair engines.
  • You have cleaned the data and prepared the *.tmx file or files that will be used for the training.
  • (Optional) You have cleaned the data and prepared the *.tmx file that will be used for the testing.

Procedure

  1. Log in to Language Weaver Edge using a valid account.
  2. Select the Adaptation tab at the top.
  3. Select Manual Adaptation on the left pane.
  4. Select Create Adapted LP.
    The New Adapted Language Pair dialog is displayed.
  5. In the Model Name box, enter a name for your adapted language pair.
  6. Under Language Pair, select a source language and a target language.
  7. Select the version of the generic language pair that will be used as a baseline for your new adapted language pair.
  8. Under Training Host, select the host where the adapted language pair will be trained.
  9. Under Input Data, select Upload Training Data to upload the *.tmx file or a *.zip file with the content that will be used for the training of the adapted language pair.
  10. (Optional) Select Upload Test Data to upload the *.tmx file with the content that will be used for the testing of the adapted language pair.
  11. Select an option for Adaptation Mode:
    • Generic: Retains more of the generic nature of the baseline language pair. Recommended for use with multiple domains, especially the one covered by training data. Requires a longer adaptation time.
    • Balanced: Offers a good balance between generic and domain-specific content.
    • Domain Specific: Retains more of the domain-specific training set. Recommended for use with a single domain that matches the provided training set.
  12. (Optional) In the Comments (optional) box, add any other relevant information about the adapted language pair.
  13. Select Analyze.
  14. Once the analysis has finished, select Create to start the adaptation process.

What to do next

Once the adaptation process has finished, you can deploy the adapted language pair.