Hi all,

I'm pleased to announce the following updates that have recently been added to Language Weaver.

New Features

Terminology (Dictionaries, Brands, Feedback)

Terminology Support has been added to Language Weaver Pro model outputs, meaning that users can leverage all their existing terminology in Language Weaver when translating with Pro models.

Dictionaries on Pro

• Support for dictionary terms have been added to Pro models, and can be applied in exactly the same way as with other Language Weaver models.

  • In the Translate Portal select the Translate Settings icon, then select the dictionary you would like to apply from the Dictionaries dropdown

  • Via API, enter the dictionary ID in the call (full information on how to use dictionaries via API are available in the API documentation).

• Multiple dictionaries can be applied at the same time.

• Dictionaries can be added via Settings > Dictionaries

Brands on Pro

• Support for brands have been added to Language Weaver Pro model outputs, and can be applied in exactly the same way as with other Language Weaver models.

• Brands are especially useful when you want to preserve specific terms untranslated, like company names, product names or codes.

• Brands are independent of the source and target language, and will always be applied.

• Brand terms can be added via Settings > Brands

• Full information on how to use Brands via API is available here.

Feedback on Pro

Support for Feedback has been added to Language Weaver Pro model outputs, and can be applied in exactly the same way as with other Language Weaver models.

• Feedback can be applied in the Translate UI, or via the Feedback Editor.

• Feedback acts like a 100% Translation Memory match. When you suggest a better translation for an output, Language Weaver will save this as Feedback. When the same Input is detected in future for the same language combination, Language Weaver will apply the improved output, rather than processing a new translation.

• Feedback will be applied to translations for all users in the Language Weaver account.

  • It is possible to configure whether feedback should be approved by an Admin, or auto-approved, in the Account Settings.

• Full information on how to use Feedback via API is available here.

Hallucination Detection

• Language Weaver has introduced Hallucination Detection to guardrail outputs of the Pro models.

• This feature is in-place to protect users from hallucinations and overgenerations, and ensures that outputs are semantically attached to the source for translation.

• Any detected overgenerations will be removed from the Pro output and replaced with an alternative translation.
  

  • Language Weaver will fix detected hallucinations before displaying an output - there is no notification or expectation for user intervention.

• Hallucination Detection is a proprietary technology built by the Language Weaver AI Research team.

Self-Serve Trials

• It is now possible to self-activate 14-day trials of the Pro models through the Language Weaver portal.

• Only account admins are permitted to begin a trial, and there will be no auto-renewal - the trial will expire after 14 days. Other account admins will receive an email notifying them that a trial has started.

• Trials can be activated via the Translate UI or the Account Settings.

• When a trial is started, admins can assign the Pro models to specific Groups of users.

• The instructions for activating a trial are below:

  • On the Language Weaver portal, select Try Pro models from the top ribbon

  • Select Start Trial

  • Select Configure and Set Up Permissions

  • Click on Go to Groups and select the group of users you wish to assign the language pairs to.

    • If you have not created Groups, all users will be assigned to the Default Group.

  • On the group page, click Language Pairs, on the left-hand side menu.

  • Click the Assign button.

  • Select the language pairs to assign to the current group.

  • You can filter the language pairs based on Language Pair Type filter.

Enhancements

• The Language Pairs page has been updated, including a new side bar to enable easier navigation between model types.

• The Reporting Dashboard has been updated to enable easier reviewing of usage.