2024.07.18. The Fair Economy — Global language artificial intelligence (AI) company DeepL recently announced that it has introduced a next-generation language model into its translation services.
According to DeepL, the new model is built on advanced large language model (LLM) technology specifically designed for translation and writing proofreading. The company said the launch marks a major advancement in enterprise language AI technology and is expected to set a new industry standard for translation quality and performance.
DeepL has further enhanced its translation services through the adoption of this next-generation LLM.
The new solution features an in-house developed LLM specialized for specific languages, exclusive data collected for translation purposes, and model tutoring by language experts.
The language-specialized LLM provides real-world level translation and writing while reducing the risk of hallucinations and mistranslations.
Unlike general-purpose models that simply train on public data, DeepL trains its models using exclusive data collected over more than seven years specifically for content creation and translation. In addition, thousands of language experts directly tutor the language models to lead quality improvement and maintenance efforts.
In a blind test conducted by DeepL among major linguists, the new LLM showed translation performance about 1.7 times better than the previous model for combinations involving English, Japanese, and Simplified Chinese. For English-German combinations, performance improved by 1.4 times. The increased accuracy helps companies reduce the time spent on document proofreading, enabling more efficient communication in business environments.
The LLM-based service is available to DeepL Pro users and will gradually expand the range of supported languages, starting with English, Japanese, simplified Chinese, and German. Users can activate the new LLM by selecting "next-gen model" in the web environment. Meanwhile, DeepL Pro users receive high-level enterprise-grade security that meets compliance requirements such as ISO 27001, GDPR/SOC 2 Type 2, and the data they input is not used for model training.
Jarek Kutylowski, founder and CEO of DeepL, said, "This announcement is only the beginning of DeepL's journey in enterprise LLM-based language AI solutions. As a research-driven company, DeepL is investing heavily to push beyond existing limits in the quality, efficiency, and scope of translation and writing. Our ultimate goal is to help global customers expand and succeed worldwide more easily, without language barriers."
Since its founding in 2017, DeepL has grown rapidly by providing essential translation and writing solutions for global business, establishing itself as a preferred language AI company for enterprises around the world.
DeepL's solutions help resolve foreign-language challenges both inside and outside companies, such as internal communication, customer service support, and global expansion. Unlike general-purpose systems, DeepL uses specialized AI models tailored to specific languages to provide more accurate translations in various scenarios and significantly reduce the risk of errors such as mistranslations and hallucinations. This is why DeepL's solutions are trusted in business environments that demand high accuracy.
In June, DeepL launched DeepL for Enterprise, a customized language technology product designed to address enterprise customers' needs for AI solution integration. The company's customer network is also growing rapidly, with half of the Fortune 500 companies and more than 100,000 enterprises, governments, and other organizations worldwide using DeepL.
In May, DeepL announced that it had secured $300 million (approximately 410 billion won) in new investment led by Indext Ventures.
0 Comments
No comments yet.