Automatic machine translation QA
Learn how CaptionHub automatically improves machine translation quality and readability
CaptionHub applies automatic post-editing to machine translations to help them meet your captioning constraints and also provide the best possible translation within your specific language and team’s grammatical and style context.
Our custom-built MT post-processing engine automatically performs a wide range of quality assurance (QA) checks to improve translation accuracy and readability before human review.
Specifically, this includes (but is not limited to):
- Adjusting line breaks or splitting longer captions when needed.
- Removing multiple spaces, leading or trailing spaces, and unwanted space-like characters (such as tabs or control characters).
- Removing leading punctuation.
- Fixing mismatches in bold or italic tags within captions.
- Detecting when MT engines return misaligned captions (for example, empty, ultrashort, or overlong target captions) and using proprietary algorithmic Natural Captions Technology to re-align them with the original.
- Inserting line breaks in semantically appropriate places when translated captions exceed one line.
- Automatically fixing around 95% of “maximum target length in characters” issues before human review, highlights the remaining cases, and optionally blocking QA approval until they’re resolved.
- Checking and adjusting captions for reading speed, minimum and maximum duration, and minimum frame gaps — areas often missed by manual QA.
Please note that CaptionHub is continually developing its approach to automated QA and MT/AI/LLM transcription and translation management. Please contact your Customer Success Manager for the very latest information.
Superuser controls Superusers have fine-grained control over these behaviours in Team settings > Translation. They can:
- Set maximum character limits and reading speed on a per-language or per-template basis (for example, different presets for YouTube vs TikTok, or for Chinese vs English).
- Choose which detected QA issues block approval and which trigger only a warning.
- Allow the AI to automatically split overlong captions.
- Disable MT post-processing entirely for a specific project template if videos in that category will be reviewed by linguists elsewhere.
You can learn more about these settings here.