Machine Translation
    1. Phrase
    2. Phrase TMS
    3. Machine Translation

    Phrase Agentic Content System (PACS)

    Content is machine translated from English by Phrase Language AI.

    Available for

    • Team, Professional, Business, Enterprise and Software UI/UX plans

    Get in touch with Sales for licensing questions.

    A framework of GenAI powered agents with specialized goals such as context-aware translations with built-in review and refinement. Consumes AI Units (AIUs).

    AI Translation Agent

    The AI Translation Agent is the translation component of the Phrase Agentic Content System (PACS). It is designed to streamline and improve translation workflows.

    Unlike traditional setups that use multiple disconnected AI components, AI Translation Agent integrates multiple capabilities into a single component, managing the process end-to-end and minimizing the need for manual review of intermediate steps.

    The AI translation agent uses GenAI and resources like TMs and TBs. When content is automatically translated, it goes through a full agentic pipeline: translation, error correction, terminology handling, and fluency refinement without any additional configuration.

    AI Translation Agent leverages multiple OpenAI large language models and performs the following tasks:

    • Context-aware translation across multiple segments

    • Dynamic real-time customization through retrieval-augmented generation (RAG), using translation memories attached to projects

    • Refinement for consistency and fluency

    • Fixing of linguistic errors

    • Terminology correction based on term bases attached to projects

      When a term base is assigned to the project, AI Translation Agent respects preferred and forbidden terms defined in the term base to guide the generated output. Other term base information is not currently used.

    • Correction of tag errors

    Usage costs

    Customers do not need their own OpenAI credentials to use AI Translation Agent. Phrase handles token consumption and normalization making costs predictable as they are based on the number of input words.

    AI Translation Agent usage is priced in AI Units (AIUs).

    Example of a pre-translation configuration where 101s are locked, and TM threshold is 90.A job has with a total of 100 words across 10 segments.

    • 1 segment (10 words) come from TM as 101% match, and is locked.

      NO AIU or MTU consumed - as it coming from TM

    • 1 segment (10 words) come from TM as 100% match, and is unlocked.

      NO AIU or MTU consumed - as it coming from TM

    • 1 segment (10 words) is a 90% fuzzy match from TM.

      NO AIU or MTU consumed - as it coming from TM

    • 1 segment (10 words) is a 99% match from Non-Translatable.

      NO AIU or MTU consumed - as it Non-Translatable

    • 6 segments (60 words) are new, empty targets, with no suitable TM or NT available.

      AIU is consumed.

    Using AI Translation Agent

    AI Translation Agent can be selected directly from Phrase TMS projects, Phrase Portal and API to translate and improve content.

    • TMS projects

      Select AI Translation Agent in the Machine translation settings when creating or editing a project.

      The system will use it to pre-translate content in the CAT editor with the specified tone and style, while leveraging any translation memories (TMs) and term bases (TBs) attached to the project.

      Note

      The AI Translation Agent is available for selection only if multiple MT profiles are enabled in the organization.

    • Phrase Portal

      Select AI Translation Agent in the Languages and resources settings when creating or editing a portal.

      The system will use it to generate translations in the translation interface of the portal, while leveraging any translation memories (TMs) assigned to the portal itself.

    AI Translation Agent Insights

    Available for

    • Team, Professional, Business, Enterprise and Software UI/UX plans

    Get in touch with Sales for licensing questions.

    An insight alert is displayed when the AI Translation Agent is estimated to outperform the MT engines currently configured for a specific language pair. The alert highlights potential improvements in translation quality and post-editing efficiency based on recent activity data.

    The alert appears during project creation after selecting source and target languages, in the Machine translation settings when MT engines are configured for the selected language pair. It is also displayed in the Project details section if the selected project template includes an MT profile with engines already configured.

    The insight alert is visible to:

    • TMS Administrators

    • TMS Project Managers who have access to view all organization data and their own data. PMs with no analytics viewing rights are excluded.

    Selecting View comparison opens the Insight side panel, which explains how the recommendation is calculated and presents detailed results per language pair.

    Comparison logic

    The AI Translation Agent (%) indicates the estimated improvement in post-editing performance when replacing the current MT engine setup with the AI Translation Agent. For each language pair:

    • Post-edited segments translated with the current MT configuration are analyzed.

    • The performance of the current MT engine setup is compared with how the AI Translation Agent typically performs for the same language pair.

    • The comparison includes native MT engines, ecosystem MT engines and BYO engines configured within the MT profile.

    Note

    Results are calculated and displayed per language pair. Language locales (e.g., en-US vs. en-GB) are not considered.

    The provided estimates are indicative only and based on historical data and typical AI Translation Agent performance. Actual results may vary.

    If multiple MT engines are configured for a language pair:

    • A weighted average is calculated across all engines.

    • As the AI Translation Agent replaces all engines configured for that language pair, the weighting reflects the actual distribution of editing activity across those engines.

    AI Translation Agent in the CAT web editor

    • AI Translation Agent is available only for batch pre-translation and does not generate real-time suggestions in the CAT web editor. Any suggestions shown in the CAT pane are taken from the existing target content.

    • AI Translation Agent suggestions are displayed in the same way as standard MT suggestions: the score and origin appear in the relevant column, metadata is shown in the CAT pane.

    • AI Translation Agent is listed as an MT provider in the Job resources panel.

    Language Support

    The massively multilingual nature of the Large Language Models (LLM) that power AI Translation Agent makes it capable of addressing many use cases and language needs.

    Phrase has identified a subset of language pairs that meet the company’s high-quality standards in both translation directions. While this subset is officially supported, AI Translation Agent is not limited to these language pairs and may be used more broadly.

    Recommended language pairs (bidirectional):

    • English - Czech

    • English - Spanish

    • English - French

    • English - Russian

    • English - German

    • English - Italian

    • English - Dutch

    • English - Chinese Simplified

    • English - Swedish

    • English - Japanese

    • English - Portuguese

    • English - Korean

    • English - Polish

    • English - Danish

    • English - Greek

    • English - Norwegian (Bokmål)

    • English - Indonesian

    • English - Hungarian

    • English - Arabic

    • English - Turkish

    • English - Thai

    Was this article helpful?
    ★ ★ ★ ★ ★

    Sorry about that! In what way was it not helpful?

    The article didn’t address my problem.
    I couldn’t understand the article.
    The feature doesn’t do what I need.
    Other reason.

    Note that feedback is provided anonymously so we aren't able to reply to questions.
    If you'd like to ask a question, submit a request to our Support team.
    Thank you for your feedback.