Translate
Amazon Translate is a neural machine translation service supporting real-time and batch translation.
Language support
Section titled “Language support”Translation is between a source and a target language, a combination Translate calls a language pair. The service supports 75 languages and language variants, including regional variants distinguished by country code — Mexican Spanish (es-MX) separately from Spanish (es), Canadian French (fr-CA) separately from French (fr), Brazilian and European Portuguese separately.
Not every feature covers every language. Real-time document translation supports English to any supported language and any supported language to English; brevity, profanity masking and formality each cover their own subsets.
- Real-time translation — synchronous API call, for interactive use such as chat or on-the-fly page translation
- Batch translation — asynchronous jobs over collections of documents in S3
- Real-time document translation — translates a document while preserving its formatting
Customisation
Section titled “Customisation”- Custom terminology — a supplied glossary that forces particular translations for particular terms, so product names, brand terms and regulated vocabulary come out consistently
- Active custom translation — adapts output using your own parallel data without training a separate model
- Formality — selects a formal or informal register in languages that distinguish them
- Brevity — produces shorter output where length matters
- Profanity masking — masks profane words in the output
- Automatic language detection — where the source language is unknown
Typical uses
Section titled “Typical uses”- Localising content and product catalogues
- Translating inbound customer support messages and outbound replies
- Multi-language chat
- Document translation
- Serving a website in the visitor’s language
Practical notes
Section titled “Practical notes”- Translating between identical source and target languages is not charged, but automatic detection is charged when the source is set to
auto. - Machine translation is a draft, not a sign-off. Anything with legal or safety consequence needs human review.
- Custom terminology is the cheapest large improvement available on domain text — build the glossary before concluding the output quality is inadequate.