Machine translation (MT), as the name suggests, refers to translations performed by a computer program, specifically a neural network. MT is an excellent tool for translating simple texts or pre-translating your strings before further refinement.

Currently Lokalise supports two MT providers:

See the full list of languages supported by Google Translate in the official Google Cloud documentation.

Google Translate supports a wide range of languages and language combinations.

On Lokalise, Google Translate can be used:

See the full list of languages supported by DeepL in the official DeepL documentation.

DeepL provides machine translation using its own language models optimized for translation.

On Lokalise, DeepL can be used:

There are multiple ways to use machine translations on Lokalise. Lokalise supports both automated and manual workflows for implementing machine translation, allowing users to streamline bulk tasks or manually adjust individual translations as needed.

You can use Standard AI/MT when creating a translation task.

Create a new task and select Machine translation (Standard AI/MT) as the task type. Then choose the content and target languages you want to translate and configure the task as usual.

Translations generated through this task type consume your Standard AI/MT allowance. Standard AI/MT usage is measured in characters, and the resulting translations may also count toward your processed words usage.

Inline machine translation provides suggestions from MT engines as you work in the Lokalise translation editor. Suggestions appear in the right-side panel under the Suggestions tab:

In the screenshot above, the user is presented with suggestions from Google Translate and DeepL. To choose one, simply click on it. You can enable or disable inline MT providers in your personal profile under the General tab.

If you can't see these suggestions, go to Project settings (or ask your project admin to do so) and enable the Inline machine translations option under Miscellaneous:

Note that the chosen MT values will also be saved to your translation memory for future use.

To populate all empty values for a given key with machine translations, click on the Google-translate empty values button under the key name.

As the name implies, this feature always uses Google Translate. If some languages aren't supported by this provider, the corresponding empty values will remain unaltered. To provide machine translations for multiple keys simultaneously, create a new translation order.

You can't Google-translate empty values for all keys within the project at once. To do that, create a new translation order and choose Google as the provider. Alternatively, create a new Automation with the Use Standard AI action and apply it on demand by clicking Save and apply.

Translation orders allow you to request translations or reviews from professional linguists or neural networks (note: neural networks can only perform translations, not reviews).

To create a new translation order, click on the Orders button in the left menu:

Then click New order and choose a translation provider:

If you want a machine translation, choose either Google Translate or DeepL (the last two icons). MT services are cheaper ($0.001 per word) but might not always be as accurate.

Choose your project, the source and target languages (note that certain languages might not be supported by the chosen provider), and which keys to include in the order. You’ll see the order summary and total price.

It's recommended to enable the Mark target translations as unverified option so your linguists can double-check the translations generated by the neural network.

Automations allow you to set up custom actions on the per-project basis that execute automatically whenever the translation value for a selected language changes.

To create a new automation rule, open your Lokalise project, click on the More button and choose Automations.

Automations offer a variety of actions, including Use machine translation. The available providers are Google Translate and DeepL. This action is useful for automatically populating target translations when content in the monitored language changes. By default, automations apply only to empty target values.

Many languages have regional variations, such as English, which is spoken in Great Britain, the US, Australia, and other countries. How are these specifics handled by machine translation (MT)?

Languages are mapped progressively, starting with the base language and then moving to the region if the MT provider supports these more specific variants.

For example, with DeepL:

I cannot see inline machine translation suggestions. Why?

If you don't see any suggestions from machine translations, there could be a few potential causes:

If you can't access the translation editor at all, please refer to the troubleshooting section in the Translation editor article.

Will machine translation engines use terms from my glossary or TM entries?

No, machine translation engines do not use glossary terms or translation memory entries. However, Lokalise AI supports working with glossary terms.

My source text contains many placeholders and inline tags. Should I use MT in this case?

We recommend to use MT with caution when translating texts that contain HTML markup and placeholders. After translation, your HTML tags and placeholders may have corrupted or misplaced brackets.

Can I use my custom trained translation engines with Lokalise?

No, currently this is not possible.