Amazon Translate
Amazon Translate is a neural machine translation service from Amazon Web Services used to translate text between languages.
AI Practitioner focus
- Translate is managed neural machine translation for text/documents and multilingual applications.
- It translates language, while Amazon Comprehend analyzes meaning and Amazon Transcribe converts audio to source-language text. A common pipeline chains Transcribe -> Translate -> Polly.
Key points
- Uses neural machine translation to translate text from one language to another.
- Listed alongside Amazon Comprehend, Amazon Kendra, Amazon Personalize, Amazon Polly, Amazon Rekognition, and Amazon Transcribe as part of the AWS Machine Learning services family in the AWS Management Console.
- Real-time translation — synchronous calls translate short strings of text on demand, suited to interactive uses such as translating a chat message or a page of a website as it loads.
- Batch (asynchronous) document translation — a batch job translates a whole collection of documents (up to 5 GB) stored in an Amazon S3 input bucket and writes the results to an output bucket; intended for bulk workloads such as translating a document repository where an instant result isn’t required.
- Custom terminology — upload a terminology file so specific terms — brand names, product names, character names — translate to an exact desired term instead of a generic literal translation (for example, keeping “Amazon Photos” untranslated in French rather than rendering it as “Photos d’Amazon”). Amazon Translate weighs the custom term against surrounding context, so it isn’t guaranteed to apply on every occurrence.
- Active Custom Translation (ACT) — supply parallel data (pairs of source/target example translations, i.e., translation memory) that Amazon Translate uses at inference time to steer the style, tone, and word choice of new translations, without training a separate custom model.
- Other output customizations: do-not-translate tags (skip specific HTML content), profanity masking, formality level, and brevity (shorter real-time output).
- Automatic source-language detection — setting the source language to
autohas Amazon Translate identify the source language before translating; auto-detection is a separately billable step. - Translates between dozens of languages (75 listed in current AWS documentation, including regional variants such as
es-MX,pt-PT, andzh-TW) — check the AWS Translate docs for the current list, since supported languages are added over time. - Real-time translation supports English paired with any other supported language as the source or target.
- Classic exam pipeline: Amazon Transcribe converts speech to text, Amazon Translate translates that text, and Amazon Polly converts the translated text back into speech.
Pricing
- Standard (real-time) text/HTML translation and batch document translation are billed at the same rate per character translated; real-time translation of richer document formats (e.g., Word-style documents) is billed at a higher per-character rate.
- Active Custom Translation is billed at a separate, higher per-character rate than standard translation, reflecting the added parallel-data customization at inference time.
- No upfront cost or minimum commitment. New accounts get a monthly free-tier character allowance for standard/batch translation for a limited introductory period, plus a smaller free allowance for Active Custom Translation — confirm current limits and rates on the AWS Translate pricing page, since free-tier terms and prices can change.