Machine translation (MT) is an automated process where text segments are translated by an MT engine. In localization workflows, MT is typically used to pre-translate content that is not matched by translation memory (TM), helping reduce manual translation effort and accelerate translation workflows.
MT is commonly used in combination with human review, post-editing or automated AI-assisted workflows to improve translation efficiency and consistency.
In the Phrase Platform, MT is provided primarily through Phrase Language AI.
MT in the Phrase Platform is orchestrated through Phrase Language AI. Phrase Language AI acts as the central MT layer that connects Phrase products with different MT engines.
When translation is requested in a Phrase product, the request is sent to Phrase Language AI. Based on the configuration and MT profiles, Phrase Language AI either selects the most appropriate MT engine using Auto-select (when multiple engines are available) or uses the predefined engine in the profile.
MT can be used in several Phrase products:
-
Phrase TMS, for enterprise localization workflows and machine translation post-editing
-
Phrase Strings, for software and UI localization
-
Phrase Portal, a self-service tool that allows non-localization teams to translate text or files using the company’s configured MT
-
Phrase Studio, for audio transcription, dubbing, and video subtitle localization
Phrase Language AI can work with:
-
Native MT engines
-
Ecosystem MT engines integrated through API keys
-
Custom MT engines trained in Phrase Custom AI and deployed to Phrase Language AI
-
Custom MT engines or internal AI models connected through the Bring Your Own Engine (BYO Engine) feature
MT profiles and engine selection logic are managed within Phrase Language AI.
When content is translated in the Phrase Platform, multiple language resources can contribute to generating translation suggestions. These resources often interact in sequence to automate translation while maintaining consistency and terminology control.
In many localization workflows, any attached translation memory (TM) is checked first to determine whether the content has already been translated. If a suitable match is found, the existing translation can be reused or suggested.
If no appropriate TM match is available, the system may generate a translation using MT. MT requests are processed through Phrase Language AI, which orchestrates the use of configured MT engines.
Terminology resources can also influence the MT process:
-
MT glossaries can guide MT engines toward preferred terminology when supported by the selected engine.
-
Term bases (TBs) provide terminology guidance for translators during editing, review, or AI-assisted translation workflows.
Depending on the Phrase product and workflow configuration, the exact interaction between these resources may vary.
The diagram below illustrates a conceptual example of how TM, MT, and terminology resources may interact.