The Power of DeepL Translation
Arsylaw Kutylowski, the founder of DeepL, is fluent in German, Polish, and English, and can even order a Coke in French. However, with the help of DeepL’s AI-powered translation tool, he can read and write in approximately 30 more languages. DeepL, founded in 2017, has developed revolutionary translation software that claims to be more accurate than its competitors, including Google, thanks to its powerful artificial intelligence and collaboration with human native language speakers.
The Neural Network Architecture
DeepL’s proprietary neural network architecture, a machine learning technique that mirrors the way the human brain learns, has been trained on an extensive database of bilingual and monolingual text in 31 different languages. While this is a fraction of the languages supported by Google, DeepL’s translations are fine-tuned for nuance and precision by a team of in-house editors and contract-based translators and native speakers worldwide. This human calibration ensures high-quality translations for both casual and formal texts.
Wide Usage and Business Success
DeepL boasts over 10 million monthly active users, including individuals who use the tool to communicate with loved ones in their native languages and enterprise clients such as Mercedes Benz, Fujitsu, and Deutsche Bahn. These enterprise customers utilize DeepL’s software to translate a wide range of content, from websites and legal contracts to marketing materials and customer agreements. In January 2023, the company secured $100 million in funding from global venture capital firms, valuing the startup at $1 billion.
Benefits Over Competitors
Users of DeepL, such as Nina Gafni, a professional translator, appreciate the accuracy and cultural nuance of its translations. Machine translations often lack cultural sensitivity and can be overly literal, whereas DeepL’s translations are praised for their contextual correctness. DeepL’s translators focus on refining translations using movie subtitles, book translations, and forum conversations to provide accurate and human-like translations.
Future Innovations
While DeepL currently does not employ large language models like Transformers for translation, founder Arslaw Kutylowski plans to develop an in-house large language model to enhance results. The company is leveraging Nvidia’s AI-focused data center to bolster its computational resources and aims to explore generative AI, allowing translators to interact more seamlessly with users and provide a dialogue-like experience.
