Alibaba’s Translation Large Model “Marco”Achieves Large-Scale Commercial Use

On October 16, Alibaba International unveiled its newly developed Marco Translation Mega-Model. This advanced model supports 15 of the world’s most commonly spoken languages, including Chinese, English, Japanese, Korean, Spanish, and French. It has been made publicly available on Alibaba International’s official AI website, Aidge, for global users.

The Marco model, when evaluated against the widely used open-source dataset Flores, outperforms other leading translation products in the market, such as Google Translate, DeepL, and GPT-4. This puts Marco at the forefront of AI translation capabilities, making it an industry leader.

One of the key strengths of the Marco model is its ability to provide precise translations based on context, driven by a large language model. This approach avoids the pitfalls of literal translations that can often result in laughable ambiguities. It also allows for more authentic translations of popular social media buzzwords.

The Marco model employs various techniques to ensure high-quality translations. It uses multilingual data screening technologies to obtain high-quality, large-scale multilingual data. It also uses a combination of multilingual mixed experts and parameter expansion methods to maintain the performance of dominant languages such as Chinese and English while enhancing the quality of other languages. Furthermore, through model quantization, acceleration, and other optimization strategies, the service cost of the mega-model is significantly reduced, making it cost-effective even when compared to traditional smaller models.

Currently, Marco is being used on a large scale in commercial settings. Its performance in the cross-border e-commerce field, an area where Alibaba International has years of experience, is especially noteworthy. For individual users, Marco offers high-quality translations with good readability based on context. It even supports different translation styles, catering to a variety of translation needs.

Last year, Alibaba International established an AI team that has since tested AI capabilities in over 40 scenarios. This has helped optimize processes for 500,000 small and medium-sized businesses and improved the quality of 100 million products. The demand for AI from businesses continues to grow. Recent data shows that the volume of AI usage doubles every two months, with the daily number of AI calls surpassing 100 million.

SEE ALSO: BYD Tang L SUV Spy Photos Exposed