Lilt’s Contextual AI Engine is run on large language models (5x more parameters than the models they are replacing). These models provide predictive translation suggestions to the translator, while also learning in real time from the translator’s decisions and feedback. The result is a continuous loop in which the human always has the final say on the translation, but receives increasingly better suggestions from the predictive machine translations. Over time, the engine learns each organization’s unique voice and frequently used words and phrases and allows for translations at greater velocity to reach scale within a given budget.

Each Project in Lilt is associated with one Model + Translation Memory pair, and each language pair can have multiple Model + Translation Memory pairs to optimize model training for specific domains or content types.

To learn more about Translation Memories, read more here.

Managing your Contextual AI Models

Managing your Contextual AI Models in Lilt can provide transparency and control over the AI translation systems used for localization. To get started, navigate to the Contextual AI tab.

Inside the Contextual AI tab you’ll see a list of models. Models are grouped into sections based on language pair.

Each Contextual AI Model is displayed as a card with the following information:

  • Lilt Model name

  • language pair

  • associated Lilt Memory name

  • total segments within the Lilt Memory

  • date the Lilt Model was last updated

Clicking “Show memory” will take you to the associated memory for more information about the resources the model has been trained on.

Create a New Contextual AI Model

Lilt Contextual AI Models can be created at any time from the Models page of the Contextual AI tab.

To create a new model, click the + New Model button in the upper-right.

This will open a modal with the following fields to fill out in order to create your new Contextual AI Model:

  • Model name: The name of the Lilt Contextual AI Model.

  • Source: The source language of the model.

  • Target: The target language of the model.

  • Memory: Selection of an existing memory.

    • NOTE: This selection will automatically train the new model with the resources in that translation memory. A new, empty translation memory will then be created to be paired with the new model.

After filling in the fields, click the Create button to create the Lilt Contextual AI Model. Once created, you can use it for translation workflows. This is done by selecting the associated (new) memory during job creation.