Overview:
Smartling built an AI-powered translation solution on Amazon Web Services (AWS) to help enterprises localize content at scale with better quality, lower cost, and minimal latency. The platform — part of Smartling’s “LanguageAI” offering — uses generative AI to produce translations that preserve brand voice, terminology, and cultural nuance while reducing manual post-editing work. The system supports over 450 languages and locales, serving hundreds of enterprises and processing billions of words annually.
The solution addresses key challenges for global content teams: foundational large-language translations were fluent but often missed subtlety (tone, terminology, style), leading to high post-editing overhead; and latency from AI inference threatened smooth automated workflows. Smartling’s AWS-based architecture enables fast, high-quality translations while significantly lowering total translation cost.
Key Features:
Ingests source content and applies a generative AI foundation model (Amazon Nova Pro) for initial translation.
Employs a retrieval-augmented generation (RAG) framework that integrates translation memory, customer-specific terminology, and style guides to ensure brand consistency and linguistic nuance.
Utilizes prompt-engineering and optimized tokenization — reducing prompt size from ~3,000 to ~200 tokens — for efficient, cost-effective translation without sacrificing quality.
Stores translation data in scalable cloud storage (Amazon S3) and employs real-time monitoring (via Amazon CloudWatch) to track performance and maintain low latency.
Integrates with the broader content and localization ecosystem (through AWS services), enabling enterprises to embed translation workflows into content pipelines and deploy globally translated content rapidly.
Results & Impact:
Achieved a 26% improvement in translation quality (measured by BLEU score) compared to prior machine translation engines.
Reduced human editing (post-edit) effort by 30%, cutting down manual labor for linguists.
Lowered translation costs by up to 15× relative to traditional trained machine-translation engines.
Delivered translation with sufficiently low latency to support fully automated workflows, eliminating inference delays and improving throughput for global content deployment.
Enabled enterprises to scale multilingual content production efficiently across hundreds of languages locales — supporting global expansion and faster time-to-market for localized content.
AI Technology:
AI Model Types: Generative AI (foundation model — Amazon Nova Pro), Retrieval-augmented generation (RAG)
AI Purpose: Translate, Adapt to terminology and style, Automate localization workflows
Application Type: Translation & Localization, Content Operations
Target Users:
- Localization / Localization Managers
- Content operations teams (e.g., marketing, product content, documentation)
- Global marketing & communications teams
- SaaS / App development teams requiring multilingual support
- Enterprises expanding into new markets needing efficient, consistent localization
Sources:
https://aws.amazon.com/solutions/case-studies/smartling-case-study