MT and LLM translation rely on external translation providers, which Smartling integrates with to give you access to the latest, highest-performing MT engines and models. When you translate using MT or an LLM, Smartling sends the content to the external provider, which responds with either a translation or an error.
Errors can be permanent, which require action by you or your team (e.g., unsupported locale pair or invalid credentials), or temporary, which may resolve automatically (e.g., quota limit reached or token-per-minute limit exceeded).
Example:
If you set up a workflow where strings are translated by Google MT, when you authorize content, the strings are sent to the Google MT service. Smartling waits for a response, and any errors returned are displayed in the Smartling dashboard.
Segment-level translation
By default, when strings are sent to a translation provider, they are sent at the segment level. This means that if a string contains multiple segments, each segment is sent individually to the translation provider to optimize translation memory leverage. As a result, translations may not be returned at the same time, and you may see some strings translated before others.
Error handling and retries
Most temporary errors are retried instantly twice, after which the strings enter a queue for retries using exponential backoff. Certain temporary errors, including 400, 401, 403, 404, 429, 502, 503, and 504, are not retried immediately. These typically indicate server issues (e.g., "too many requests" or timeouts) or insufficient permissions. Permanent errors are not retried instantly.
Smartling will retry requests to a provider several times when MT or LLM translation is used in a workflow (as opposed to an Instant MT integration) if the provider returns an error. Retries follow standard exponential backoff practices based on the number of errors returned, which means the time between each retry increases every time an error is received from the provider. This may delay the return of a segment's translation, which can, in turn, delay the translation of the entire string.
If you have a fallback provider configured, strings are sent to this provider or routed to the alternative translation method you specified if the initial translation provider returns an error. This can help prevent translation disruptions.
Troubleshooting MT errors
If any errors occurred on an MT workflow step (e.g., invalid API key, unsupported language, placeholder or tag inconsistency), they’ll show up in the Translation Profiles page.
- From the top navigation of your Smartling dashboard, access the AI Hub.
- Navigate to the Translation Profiles section.
- Enter the MT Profile by clicking the profile name
- If there are any errors associated with the profile, they will show up in the yellow box.
- The error box is collapsed by default, so you’ll need to click on Show Details for more information.
- Click on Retry MT to retry the MT requests
- If the errors persist, they will appear again after a few minutes.
Types of MT Errors
Errors can be found on all strings or on some strings.
All strings
If the MT error was found on all strings, it could mean that the translation provider credentials are wrong or invalid. Ensure all credentials are accurate and then click Retry MT to retry the MT requests.
Another reason why errors can be found on all strings is if the language pairs are invalid. Ensure language pairs correctly mapped from the translation provider and Smartling.
Some strings
If the MT error was found on some strings, it could mean that placeholders and/or tags in the translations do not match those in the source strings. This means that placeholders/tags could be different, invalid, or in a different location in the translations.
Another reason why errors can be found on some strings is if the translation provider failed to respond to the request. This could be due to any number of reasons, including a network error. In this scenario, just wait a couple of moments before clicking Retry MT to retry the MT requests.