Innovating Translation: How DeepL Competes and Excels Beyond Industry Giants

Founded in 2017, DeepL started as a modest translation website and has since evolved to cover 31 languages, serving over 20,000 clients, including law firms and consulting companies, particularly in Asia.

For years, the prospect of a real-life universal translator, akin to the sci-fi translation gadget seen in countless books and movies, has been a tantalising promise from the tech community. Recent strides in artificial intelligence, particularly with the introduction of ChatGPT in 2022, have brought this futuristic concept closer to reality. Amid the contenders aiming to provide such a breakthrough, Alphabet Inc.’s Google, with its extensive Translate service, has been a front-runner. Additionally, OpenAI, the creator of ChatGPT, and Spotify have unveiled plans to utilise advanced AI technology for translation purposes.

However, there’s a less prominent yet formidable candidate in the race – the 700-person startup DeepL SE based in Cologne, Germany. Despite its lower profile, DeepL has been making waves in the field of machine translation, with an impressive valuation of €1 billion ($1.1 billion) after a funding round earlier this year. DeepL is set to introduce its first voice interpreter in December, a feature that automatically captures spoken words, translates them and transcribes them in another language. The company envisions integrating this feature into its app and other services like Zoom, envisioning a future where language barriers become inconsequential in business meetings.

In contrast to its sprawling competitors, DeepL is carving its niche by focusing exclusively on machine translation. As the influence of AI expands across businesses, the question arises: will a few general-use models dominate the market, or will various organizations thrive by offering specialized tools? DeepL’s continued success suggests the latter scenario.

Founded in 2017, DeepL started as a modest website and has since evolved to cover 31 languages, serving over 20,000 clients, including law firms and consulting companies, particularly in Asia. The startup, which competes with tech giants like Google and Amazon, distinguishes itself by its dedication to machine translation. DeepL claims that “tens of millions” of users utilize its service monthly. The company is gearing up to establish its first US office in January 2024, signalling its global ambitions.

Ajay Vashee, a partner at IVP, a venture capital firm that has invested in DeepL, draws parallels between the startup and other industry giants like Dropbox and Slack, emphasizing their potential to become household names in the software industry.

According to Intento Inc., a company tracking machine translation, DeepL has been ranked alongside Google and Amazon for the best overall offerings, outperforming in sectors such as education, health care, and finance. Academic studies have shown that DeepL exhibits fewer errors than Google Translate in certain language pairs. However, it faces challenges in maintaining the gender of words during translations compared to ChatGPT.

DeepL’s success can be attributed to its reliance on neural networks, the cutting-edge technology in translation software. Neural networks excel in providing translations that accurately consider the context of words and phrases. Eva Vanmassenhove, an assistant professor at Tilburg University in the Netherlands, notes that advanced translation systems must discern between multiple meanings of a word based on its usage, requiring a contextual understanding.

Jarek Kutylowski, DeepL’s founder and CEO, describes the company’s technology as a “contextual engine,” capable of processing entire paragraphs at once. While DeepL utilises OpenAI tech for certain experiments, it is also developing its own large language model. Kutylowski emphasizes that although DeepL’s system is a “grade smaller” than more general-use models, its efficiency is unmatched as it is tailor-made for translation. The company employs proprietary web crawlers to locate translations online and assess their quality, supplemented by human feedback from thousands of contractors to refine its models.

Despite its success and recognition, DeepL remains somewhat secretive about its technical details, citing concerns about intellectual property. This opacity is common in the tech industry, with the landscape continually evolving. Vanmassenhove acknowledges the challenge of comparing in this dynamic field and cautions that existing research is still relatively limited.

The global translation services market is projected to exceed $44 billion in the next decade, driven by multinational corporations seeking cost-effective alternatives to human translation. To secure such business, DeepL understands the need to outperform behemoths like Google Translate. Kutylowski expresses confidence that DeepL has demonstrated its ability to compete with larger competitors through focus and determination.

As DeepL continues to redefine the landscape of machine translation, its specialised approach suggests that a diverse array of organizations may flourish by excelling in specific tasks rather than a few dominating the market. DeepL’s journey highlights the evolving dynamics of AI and its potential to reshape industries, making once-distant sci-fi concepts a tangible reality.