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

    MTU Consumption and Weights

    Content is machine translated from English by Phrase Language AI.

    Monitoring consumption

    Monitor MT usage in the MT usage tab of Phrase Language AI.

    The tab presents:

    • MT units and translated characters consumed by project, workflow step, locale pair, MT engine, and MT profile.

    • Comparison of translated content vs. MTU consumption.

    • The TMS MT origin filter shows the split of MT units across Phrase TMS:

      • TMS pre-translation

      • TMS analysis

      • TMS CAT editor

      MT units consumed prior to the 1st of December, 2024 are not split by origin and are marked as Uncategorized.

    MTU Calculation

    Based on this formula, Phrase calculates Machine Translation Units (MTUs) based on the number of input characters sent to the Phrase Language AI service:

    MTUs = Number of input characters × (MT Service Weight + PLAI Service Weight)

    Number of input characters is calculated based on the number of Unicode code units within a string (unlike how calculations are made for analysis).

    Note

    The number of input characters does not directly correspond to the size of an uploaded document or file.

    Examples:

    • A file with very good TM matches would have a far smaller number of input characters for MTU than a file with little to no TM matches.

    • If a single file is processed by MT multiple times (once in pre-translation, once in analysis and than again in analysis), this would triple the number of input characters.

    MT Service Weight is the weight assigned to the machine translation engine being used.

    • Fully Managed Engines

      Each has a specific weight.

    • Manually Managed Engines

      Where customers bring their own credentials have a weight of 0; only the PLAI service fee is paid.

    Phrase Language AI (PLAI) Service Weight

    • A fixed fee of 0.5 MTUs per input character for accessing Phrase Language AI services is calculated regardless how many benefits are used.

    • Benefits of PLAI:

      • Centralized MT profile configuration with custom translation settings and integration across the Phrase Suite.

      • Access to Phrase Quality Performance Score (QPS) for translation quality assessments.

      • Use of custom translation portals for organizational MT needs.

      • Access to Next MT and Next GenMT (LLM-based) engines.

      • Support for custom-trained engines via Custom AI.

    Calculation Examples

    • Manually Managed Engine (e.g., custom credentials):

      10 characters × (0 + 0.5) = 5 MTUs

    • Fully Managed Engine (e.g., Phrase NextMT with weight = 0.5):

      10 characters × (0.5 + 0.5) = 10 MTUs

    Consumption management

    • Pre-translation

      A TM threshold sets when pre-translation uses MT and as MT has lower editing times, higher TM thresholds reduce human post-editing efforts. Pre-translation only uses MT if no TM match is found at or above the set threshold. Using the recommended threshold is likely to reduce total translation costs.

      • In the global Pre-translation settings or for a specific project or project template, ensure Apply MT only when there's no TM match at or above the threshold is enabled.

    • Default analysis

      The standard analysis is run on uploaded source segments before translation to assess translation effort and serve as the baseline for generating quotes for clients.

      MT is only consumed if a quality score (QPS) is required and if there are no 100/101% TM or non-translatables (NT) matches.

      If not required:

      • Disable the Include machine translation suggestions (QPS) option.

      • Ensure 100/101% TM or non-translatables (NT) matches are prioritized over MT matches in pre-translation settings.

    • CAT editor

      • In the CAT editor settings for the project or project template, enable Suggest MT only for segments with a TM match below xxx%.

        • Enabling this option controls when MT is suggested in the CAT pane. MT will only be used for the segments with a TM match below the set threshold.

        • If segments are confirmed (and thus saved to TM) in a workflow step, they will have a 101% TM match in subsequent steps. As a result, MT will not be suggested, preserving MTU in those steps.

        • If this option is disabled, MT is suggested regardless of the threshold, leading to MTU consumption.

      • A further optimization of MTU consumption can be achieved by pre-translating a job before opening it in the editor. If MT is used as the target during pre-translation, the CAT pane reuses the MT suggestion without making a new MT request. This optimization is enabled by default (without any settings control) and is not available in the CAT desktop editor.

        MTU are still consumed in the following cases:

        • No pre-translation is performed on the job.

        • MT is not used for a segment during pre-translation (e.g., NT is used instead).

        • A TM match is used in pre-translation, and MT is called if the TM match does not meet the threshold defined for the CAT editor.

        • The source text is changed, or a segment is joined/split.

        • The MT engine is changed after pre-translation.

        • The source text is updated as part of a source update or a continuous job.

      • MTU are consumed both when working on segments within the editor and when segments are pre-loaded:

        • CAT results are pre-loaded to minimize wait times, consuming MTU for the previous segment, the current segment, and the next two segments.

        • When navigating with keyboard arrows or using Go to segment, MTU are consumed for these pre-loaded segments.

        • When jumping upon confirmation, MTU are consumed for the next two segments based on the After confirming a segment, jump to setting in the editor.

        • Pre-loading respects filtering results, ensuring that MTU are consumed only for segments displayed after filtering.

      • MTU are not consumed repeatedly for revisited segments within the same user session. A user session remains active as long as the job is open in the browser.

        • By default, the maximum editor user session lasts three days (unless a shorter session duration is set in TMS). After this period, the user is required to refresh the job.

        • Refreshing the job - whether due to session expiration or manual refresh - ends the current user session.

        • Upon refreshing, a new user session begins, and MTU are consumed again for the relevant segments.

      • User role rights

        • MT can be enabled or disabled, and MTU consumption further optimized, for Linguist and Guest users in their user profile.

    • Phrase Portal

      MT is always used, and third-party engines outside PLAI cannot be configured.

      If not required:

      • Use a manually managed engine with credentials to only incur PLAI fees and avoid engine fees.

    Machine Translation Service Weights

    Type

    Weight

    Amazon Translate

    0.5

    Custom NextMT

    0.5

    DeepL

    0.5

    Google Translate

    0.5

    Microsoft Translator

    0.5

    Phrase Next GenMT

    1.0

    Phrase NextMT

    0.5

    Rozetta Translate

    2.0

    Tencent

    0.5

    Phrase Language AI Service Weights

    Type

    Weight

    With or without additional features

    0.5
    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.

    Articles related to Machine Translation