![]() |
Machine Translation |
Machine translation (Wikipedia) is a technology where a cloud service automatically translates sentences from one language into another. Machine translation lets you quickly translate your strings for testing or for post-editing, and its quality gets better every year.
Soluling Desktop and SoluMake use the machine translation services that you configure. In Soluling Cloud, Translate shows machine translation suggestions while you translate in the browser.
No single machine translation engine is the best in all languages. Each engine has good and less good language pairs. This is why Soluling supports several machine translation services. The prices and free usage limits of the services change, so check the current pricing on each provider's site.
![]() |
Amazon TranslateAmazon Translate (home page) is a neural machine translation service provided by Amazon. It can use custom glossaries. Amazon Translate is a paid service, and new AWS customers get a free tier. See the pricing. To use Amazon Translate, configure it. |
![]() |
Azure AI Translator (Microsoft Translator)Azure AI Translator (Wikipedia) is a neural machine translation service provided by Microsoft. It supports more than 100 languages. It has a free tier with a monthly character limit. If you translate more, you pay per character. See the pricing. To use Azure AI Translator, configure it. |
![]() |
ClaudeClaude (Wikipedia) is a large language model made by Anthropic. Its translation quality is very good. Claude is a paid service. You create your own API key and fund it. See the pricing. Claude uses the context of the row: its comment, maximum length, and glossary terms. To use Claude, configure it. Soluling also uses a large language model for AI validation. |
![]() |
DeepL TranslatorDeepL Translator (Wikipedia) is a neural machine translation service. Its translation quality is very good. It can use custom glossaries, which you manage in the DeepL configuration dialog. In Soluling Cloud you don't need to manage them: Soluling imports the account's terminology as a glossary to compatible machine translation engines and updates it automatically whenever the terminology changes. Soluling also gives DeepL the comment of the row as context, which helps it choose the right meaning of short strings (see context-aware translation). Soluling uses the DeepL API, which has a free plan with a monthly character limit and a Pro plan that you pay for by usage. See the API plans. To use DeepL, configure it. |
![]() |
Google Cloud TranslationGoogle Cloud Translation (Wikipedia) is a neural machine translation service provided by Google. It provides good translation quality and supports more than 100 languages. It has a free monthly character limit. If you translate more, you pay per character. See the pricing. To use Google Cloud Translation, configure it. |
![]() |
ModernMTModernMT (home page) is an adaptive neural machine translation service. Its translation quality is good. ModernMT is a paid service that you pay for by usage. You need a ModernMT plan that includes API access. See the pricing. To use ModernMT, configure it. |
![]() |
MyMemoryMyMemory (home page) is a translation memory and machine translation service. It is free up to a daily character limit, which is higher if you give your email address. See the usage limits. The quality is good if there is a close match in its translation memory. If not, the quality is lower. To use MyMemory, configure it. |
When you install Soluling, all the above machine translation engines are added, but most of them are not configured. To configure an engine, get an account key from the provider and enter it in the configuration dialog of the engine. Choose Options | Machine Translators to configure the common machine translation engines. Each project can use the configured common engines, its own project-specific engines, or both. Use the Machine Translation sheet of the project options to edit the machine translation settings of a project.
You can use machine translation in two ways:
User interface strings are often short, and the same English word can have several meanings. "Open" can be a command (open a file) or a status (the ticket is open). "Close" can be a command or a distance. "Free" can be a price or unused disk space. Most languages need a different word for each meaning. A human translator sees the comment next to the string, but a traditional machine translation engine gets only the text, so it has to guess.
A large language model (LLM) such as Claude can use the same information as the translator. When Soluling translates a row with a LLM, it also gives LLM the context of the row:
Unit1.SStatusOpen or Form1.OpenButton.Caption. Often the id alone tells a verb from an adjective.//loc comment, a .resx comment, an ARB description, or an Angular i18n description, and the comment you added in Soluling.//loc MaxChars=12), the translation must not be longer.LLM also keeps the placeholders (%s, {0}, {name}), the markup tags, ICU plural and select syntax, and the hotkey markers unchanged.
The more comments your source code has, the better the translations are. Add comments to short and ambiguous strings in particular. For example:
resourcestring SStatusOpen = 'Open'; //loc Status of a support ticket (adjective): the problem is not solved yet. Not a command. SOpenCommand = '&Open'; //loc MaxChars=12 Button (verb): opens a closed support ticket again. SNear = 'Close'; //loc Distance of the store (adjective): near. Not the Close command.
Context-aware translation works both in the interactive and in the batch mode. The other machine translation engines translate as before, without the context.
In addition to the context of each row, you can give instructions that apply to the whole project: the tone, the formality, the words that are never translated, the spelling, and so on. Soluling sends this style guide with every string LLM translates. For example:
Use an informal tone and address the user with "sinä". Never translate the product names Soluling and SoluMake. Use American spelling.
Keep the style guide short and specific. The terminology belongs to the glossary, which LLM also gets.
In Soluling Desktop, enter the style guide in the Style guide for AI translation field on the Machine Translation sheet of the project options. In Soluling Cloud, enter it in the Style guide field on the project's manage page. The style guide is stored in the project file, so Desktop and Cloud share it: when you upload a project file that has a style guide, Cloud uses it, and a project file downloaded from Cloud contains the style guide of the cloud project.
A translation that LLM made with the context of the row, or that DeepL made with the comment of the row, fits the row better than a match of the same text from another row. That is why the candidate window shows it at the top of the list, above the duplicates and the translation memory matches.
If the project uses the Optimize usage option, Soluling normally does not call the machine translation engines when the translation memory contains a 100% match. A 100% match that was saved from another row may have a different context, so Soluling still calls the engines that use the context (e.g. Claude, and DeepL when the row has a comment). It does not call them if the match was saved from the same row.
When the row has a comment or a maximum length, the candidate window also shows a 100% match that was saved only from other rows with the other context info instead of the percentage, after the matches saved from the same row, the same way as a duplicate with a different context. Check that it fits before you use it. A match whose rows are not known, for example one saved by an older version of Soluling or imported from a file, is shown as before.
The duplicates have a similar issue: a row with the same original value but a different comment may need a different translation. See Duplicates with a different context.
There are several machine translation technologies. Phrase-based (PBMT) and other statistical machine translation (SMT) models were the mainstream for many years. Then neural machine translation (NMT) became the mainstream because it provides better quality. Today, most machine translation services use neural machine translation or a large language model (LLM). Some services use a hybrid model that combines several technologies. Some engines let you train your own model and use it instead of the standard one.
Machine translation is not always accurate. The results are getting better, but they do not yet match the quality of human translations. To improve the quality, Soluling can translate each sentence with several engines (e.g., DeepL and Google), analyze the results, and select the best translation. The translation status and match percentage depend on how well the translations match each other. If all the engines give similar results, the translation status and match percentage are higher. This helps you filter out bad machine translations. You can use Quality-Translate if you have configured at least two machine translation engines on the Machine Translators sheet.
To use Quality-Translate, choose Machine Translate | Use Quality-Translate instead of an engine-specific menu (e.g., Use Microsoft Translator).

Because machine translations are not always good enough to be the final translations, you most often use them during development and testing, when you need real translations quickly. Later in the process, translators edit them or replace them with human translations. To make this easier, Soluling can remove the machine translations: right-click the language column header and choose Machine Translate | Remove Machine Translations.
If you edit a machine translation, it becomes a real translation, and Soluling does not remove it when you remove the machine translations. Editing means adding or removing characters or changing the translation status.
By default, Soluling uses the original language as the source language of machine translation. You can change this in two ways:
In both cases, if the selected source language or alternative original language does not have a translation, Soluling uses the original value and language as the source.
In interactive mode, Soluling uses only the machine translation engines that are enabled for the project. Right-click the project root in the project tree, choose Options, and select the Machine Translation sheet. The following image shows a situation where several engines are configured in Soluling, but only Microsoft is enabled for this project.

The lower half of the sheet lets you configure machine translation engines for this project only.
Soluling also uses machine translation in back-translation, and a large language model in AI validation.
What is the difference between a termbase, translation memory, existing translations, and machine translator?