Translate

Neural machine translation supporting 75+ language pairs

Amazon Translate is a fully managed neural machine translation service that delivers fast, accurate, and affordable language translation across 75+ language pairs. It uses advanced neural network models trained on large-scale multilingual datasets and supports real-time translation, batch document translation, and custom terminology. For cloud engineers, Translate is the building block for multilingual applications, document workflows, and customer support systems that need to operate globally.

Real-time Translation vs Batch Document Translation

Translate offers two modes: synchronous (TranslateText) for real-time use cases and asynchronous batch jobs (StartTextTranslationJob) for large document sets.

Aspect	Real-time (TranslateText)	Batch (StartTextTranslationJob)
Input	String of up to 10,000 bytes	S3 folder with TXT, HTML, DOCX, PPTX, XLSX, or XLIFF files
Output	Synchronous JSON response	Translated files in S3 output folder
Max size	10 KB per request	Up to 1 GB per document (5 GB per job)
Latency	Milliseconds	Minutes to hours depending on volume
Formatting	Plain text or HTML tags preserved	Document structure (tables, headings) preserved for DOCX/PPTX
Use case	Chat, search, API responses	Content localization, document translation at scale

bash

# Real-time translation
import boto3

translate = boto3.client('translate', region_name='us-east-1')

response = translate.translate_text(
    Text='The deployment pipeline failed due to missing environment variables.',
    SourceLanguageCode='en',
    TargetLanguageCode='es'
)
print(response['TranslatedText'])
# "La canalización de implementación falló debido a variables de entorno faltantes."

# Use 'auto' for automatic language detection (integrates with Comprehend)
response = translate.translate_text(
    Text='Merci pour votre aide.',
    SourceLanguageCode='auto',
    TargetLanguageCode='en'
)
print(response['SourceLanguageCode'])  # fr
print(response['TranslatedText'])  # Thank you for your help.

Custom Terminology - Controlling Brand and Technical Terms

Custom terminology lets you define how specific terms (brand names, product names, technical jargon) are always translated, overriding the default neural translation. This is critical for consistent brand voice and accurate technical documentation.

Terminology files are CSV or TMX (Translation Memory Exchange) format. You define source terms and their required translations into one or more target languages.

bash

# terminology.csv format
# en,es,fr,de
# AWS Lambda,AWS Lambda,AWS Lambda,AWS Lambda
# Auto Scaling,Auto Scaling,Auto Scaling,Auto Scaling
# "blue/green deployment","despliegue azul/verde","déploiement bleu/vert","Blau/Grün-Deployment"

# Upload custom terminology
import boto3

translate = boto3.client('translate')

with open('terminology.csv', 'rb') as f:
    translate.import_terminology(
        Name='aws-technical-terms',
        MergeStrategy='OVERWRITE',
        TerminologyData={'File': f.read(), 'Format': 'CSV'}
    )

# Use the terminology in a translation
response = translate.translate_text(
    Text='Set up a blue/green deployment using Auto Scaling.',
    SourceLanguageCode='en',
    TargetLanguageCode='es',
    TerminologyNames=['aws-technical-terms']
)

💡

Custom terminology works for exact string matches only. It does not handle morphological variations (plural forms, verb conjugations). For highly inflected languages, you may need entries for multiple word forms.

Parallel Data - Customizing Translation Style

Active Custom Translation (ACT) uses your existing human-translated document pairs (parallel data) to adapt the neural translation model to your specific domain and style. This is more powerful than custom terminology - it affects style, tone, and domain accuracy across all sentences.

Feature	Custom Terminology	Parallel Data (ACT)
What it controls	Specific word or phrase translations	Overall style, tone, and domain adaptation
Training required	No - just upload a CSV	Yes - creates a Custom Translation job using parallel data
Input format	CSV or TMX with term pairs	TMX file with parallel sentence pairs
Minimum data	1 term pair	1000+ sentence pairs recommended
Cost	Free to store, charged at translation time	Storage per GB + training time

Translate Pricing

Feature	Pricing	Notes
Real-time translation	$15.00 per million characters	Free tier: 2M characters/month for 12 months
Batch translation	$15.00 per million characters	Same rate as real-time
Custom terminology	Free to store	Charged only at translation time (same character rate)
Active Custom Translation training	$60.00 per million characters of training data	One-time cost per training run
Active Custom Translation inference	$60.00 per million characters	4x premium over standard for customized translations

⚠️

Active Custom Translation inference is 4x more expensive than standard. Measure quality improvement before committing to ACT for all traffic - for many use cases, custom terminology alone achieves sufficient consistency.

🎯

Interview Focus Points

1What is the difference between custom terminology and Active Custom Translation in Amazon Translate?
2How would you build a real-time multilingual customer support chat using Translate and other AWS services?
3What document formats does batch translation support and how is document formatting preserved?
4How does automatic language detection work in Translate and which service powers it?
5When would you choose Amazon Translate vs a third-party translation API like Google Translate?
6What are the cost implications of using Active Custom Translation at scale?