What is Marco
Marco is the latest large-scale commercial translation large model launched by Alibaba International, supporting 15 mainstream global languages, including Chinese, English, Japanese, Korean, Spanish, French, and more. It surpasses competitors such as Google Translate, DeepL, and GPT-4 on the BLEU evaluation metric, providing context-based precise translation and avoiding ambiguity caused by literal meaning. Marco is based on multilingual data filtering technology and parameter expansion methods to ensure translation quality and reduce service costs. The Marco translation large model has now achieved large-scale commercial use, providing efficient and precise translation services for global users and meeting diverse translation needs. Marco has been released on Alibaba International's AI official website Aidge and is open for use by global users.
Main Features of Marco
- Multilingual translation support: Provides mutual translation capabilities for 15 mainstream global languages.
- Precise contextual translation: Performs precise translation based on context to avoid ambiguity.
- Cross-border e-commerce optimization: Especially suitable for cross-border e-commerce, optimizing product titles and descriptions.
- Diversified style translation: Supports different translation styles to meet personalized needs.
- High-concurrency processing capability: Large-scale commercial use, supporting high-concurrency calls.
Technical Principles of Marco
- Driven by large language models: Based on deep learning technology, a large language model trained with a large amount of multilingual text data can deeply understand language structure and semantics, providing more natural and accurate translations.
- Multilingual data filtering: Based on advanced data filtering technology, it ensures that high-quality data is used in model training. Through multilingual mixed corpus screening and multi-dimensional data quality assessment, it obtains precise, large-scale multilingual data.
- Mixture of experts and parameter expansion: Combines a multilingual mixture-of-experts system, where each expert is responsible for a specific language pair or language feature, improving translation quality in specific domains. Based on parameter expansion methods, it optimizes the overall performance of the model.
- Model quantization and acceleration: Uses model quantization technology to reduce the model's computational requirements and improve the speed of the translation process. Based on acceleration technologies such as parallel computing and distributed training, it further improves the model's response speed and processing capability.
Marco Project Address
- Experience address：aidc-ai.com
Application Scenarios of Marco
- Localization of product information: Accurately translates information such as product titles, detailed descriptions, attributes, and image phrases uploaded by merchants into the language of the target market, helping products better adapt to the local market.
- Multilingual customer service: Provides real-time chat translation services, helping merchants communicate smoothly with customers who speak different languages and improving the customer service experience.
- Cross-border e-commerce platforms: Integrated into e-commerce platforms such as AliExpress, Lazada, and Trendyol to provide automatic translation services for merchants and buyers, eliminating language barriers.
- Personal document translation: Provides high-quality document translation services for individual users, including but not limited to resumes, academic papers, legal documents, etc., supporting multiple languages.
- Content creation and publishing: Helps content creators translate articles, blogs, press releases, and other content into different languages, expanding the global audience of the content.