Using Machine Translation (MT)

Once you have original language captions for a project, you'll be able to translate them into subtitles.

To do that, click on the Translations tab, and click on the Add button. This will pull up an option of translation providers as well as a list of the languages available to translate into.

By default, translations will be generated using the translation provider set as the default in your team settings. However, you can manually select a different MT provider, a translation management system (TMS) if configured for your team, or choose to translate manually.

Next, you'll select the languages for which you want to generate translated subtitles, based on the options available from the chosen translation provider and available for your team. To simplify the selection process, you can use the "Select All" or "Select None" shortcuts

When Add translations is selected, CaptionHub will send your original captions to the selected translation provider, and return translated subtitles. You can then assign linguists to review and improve these translations.

If at any time you wish to overwrite or replace your translations, you can do so via ‘Replace translations’ accessible through the Translations tab. Simply select the three-dot menu for a specific translation and select ‘Replace translation.’ This will allow you to choose another supported third-party MT engine, and/or any TMS platform(s) enabled for your team.

For best MT results, ensure your source captions include accurate punctuation. This helps define sentence boundaries, improving machine translation quality and subtitle segmentation.

You can learn more about how CaptionHub automatically post-edits machine translations to improve quality, readability, and compliance with your captioning constraints here.

Normally, CaptionHub tries to preserve the exact boundaries of each caption during translation in order to maintain the original pacing of the speech. However, this can sometimes reduce translation quality, as caption boundary markers may interrupt the natural processing flow of machine translation engines.

Some MT engines are more resilient to this than others. In our experience, engines such as Amazon Translate are generally less affected, while others, particularly newer LLM-based systems, can be more sensitive to caption boundary interruptions.

Sentence blocking is currently supported only for the following machine translation engines:

Sentence blocking provides an alternative approach. Instead of sending captions individually for translation, CaptionHub groups consecutive captions together into larger sentence blocks, typically until a caption ending with a full stop, question mark, or exclamation mark is reached. The translated output is then automatically split back into captions using CaptionHub’s proprietary algorithms, while attempting to preserve the original timing and structure.

The result is typically:

For example, some words may appear in the preceding or following caption compared to the source version.

You can enable sentence blocking either:

We recommend testing both approaches to compare results. For example, you could create:

This allows you to directly compare translation quality and caption flow.