Machine translation (MT): Frequently Asked Questions (FAQ)

Introduction

The following article aims at answering common questions you might have, or resolving issues you might experience, about the machine translation (MT) module in XTM Cloud.

Before reading the article below, make sure you also read the article that handles common TM-related issues: Machine translation (MT) – most common issues & troubleshooting!

Before reading the article below, make sure you also read the article that handles common TM-related issues: Machine translation (MT) – most common issues & troubleshooting!


Questions and answers

How can matches from MT become ICE matches?

There must be a proper project’s workflow applied in which the Approve TM step (one of the automatic steps offered in XTM Cloud) is the first one:

Following the above workflow, MT matches populate into segments during the project analysis. Then, the translations from MT are sent off to the TM base when the first Automatic step (the approve TM step, as shown below) is completed. Finally, when reanalysis is performed, the segments are populated with those MT translations, which are in the form of TM matches, coming straight from the TM base, in XTM Cloud UI.

Why are there no Microsoft Translator matches in XTM Workbench when setting up the project with Microsoft Translator?

There has to be a configuration finished on the back-end side to obtain matches from Microsoft Translator. For this purpose, contact the XTM International Support team.

Is it possible to receive matches from MT only, without TM?

The Do not search for an MT match when one of the selected matches exists option (Project Editor → General infoMachine translation) refers to only enabling searching for MT matches even when a given segment already has other TM matches. If you have, for example ICE and Leveraged options selected, then MT will be omitting those segments where ICE and Leveraged were first found, and will look for another segment where either the rest of TM matches are present or the segments are empty.

Other TM matches, like ICE or Leveraged, will take priority over MT matches, and if you would like to disable the population of TM matches, then you will have to configure it in the global settings, in Configuration → Settings → Translation → TMICE matches|Leveraged matches (administrative privileges required!), and set Populate target with match to Never.

You can also do that in the project template or change the setting during the project creation. There is also an option to create a new customer, whose TM is empty, and then create a project.

Is there a way in XTM Cloud to have active MT only for a certain range of languages within a project and not all of them?

The option to enable MT only for selected languages is available for some of the MT engines; for example, Systran, Intento, DeepL Translator, Google AutoML, or Omniscien. To configure this, click on the Choose profile button in the settings of your project (Project Editor → General infoMachine translation), available in the project editor or project template editor after enabling the said MT:

In the pop-up, you will be able to select the profile used for each available language pair, as well as the option to not use the MT for the language:

Is it possible to set up options for TM-enhancing and Terminology-synchronization, together with Custom MT in XTM Cloud?

Currently, it is not possible to send TM matches or terms in an API request using Custom MT.

Does MT support language variants?

Currently, most MTs, including Microsoft Translator, use general language codes and do not recognize language variants, therefore you will probably not see any major difference in such translation. This means that if you have a combination of, for example, English (UK) to English (USA), the output will be general English. It strictly lies on the MT side and the XTM team can not do anything about it.

Is it possible to monitor Google MT usage in XTM Cloud? I would like to check our costs for using Google MT via XTM Cloud.

As far as monitoring Google MT usage, we do not maintain and do not log information on how many characters exactly have been sent for machine translation within XTM Cloud. However, you should be able to check how many characters are sent to Google MT through the Google Cloud Platform. If you are using the API key only in XTM Cloud, then, in theory, the results should be accurate on their side, although XTM Cloud cannot guarantee how their feature works. Therefore, this should be clarified with the Google support team.

Is Amazon ACT (Amazon Translate Active Custom Translation) feature supported in the XTM Cloud Amazon MT connector?

Currently, in the current implementation, the Amazon ACT is not supported.

How does XTM Cloud treat inline tags for right-to-left languages like Arabic? Does XTM Cloud reverse the order of the tags, or does it keep them as they were returned from the MT provider?

XTM Cloud does not correct the order of inline tags after returning them from custom MT. MT is responsible for changing the order of tags. In some of the MT engines, we can decide if XTM Cloud should use our algorithms to place inline tags. Intento, for example, does that automatically.

Why do I not see any matches after enabling Systran MT?

To receive matches, you will need to perform a project’s reanalysis (if, in Configuration → Settings → Translation → Machine translationSYSTRAN MT, you set Send text for matching to After analysis) or open XTM Workbench in the edit mode and move through segments (if you set Send text for matching to In XTM Workbench).

Before this, in Configuration → Settings → Translation → Machine translationGeneral options, you can set the Populate target with match option to Always, fixing inline differences. It can also be set on the project level as well, in Project Editor → General info, or during the project creation.

Why is there the same text in source and target language in XTM Workbench when using Systran MT?

First of all, make sure that the Populate target with source text setting was not enabled in XTM Workbench (Settings General).

The XTM International Support team can search for the logs on the back-end side. However, if there are no errors associated with the analysis itself and the match is populated, then the whole process is successful. If Systran returned text in English, it means that the issue is with the engine itself.

Why do I not see any matches while using the Microsoft MT?

The XTM International Support team can surely verify the logs, but it is worth checking the Category ID field, in Configuration → Settings → Translation → Machine translationMicrosoft Translator.

Removing or changing the Category ID might help with resolving the issue.