The Evolution of AI-Powered Translation for Businesses

The majority of individuals utilizing AI-driven translation services do so for simple, everyday needs, such as deciphering a phrase or quotation. However, these fundamental tools are insufficient for organizations requiring technical documentation in numerous languages – potentially up to fifteen. Lengoo, with its bespoke machine translation models, presents a viable solution, and a recent $20 million Series B funding round positions them for substantial growth.

A Substantial and Enduring Market

The translation industry represents a multi-billion dollar market with consistent demand. The necessity of adapting documents, software, and websites into multiple languages – sometimes dozens – remains a prevalent requirement.

Currently, this work is largely handled by translation agencies. These agencies employ skilled linguists to deliver high-quality translations on demand. The emergence of machine translation as a common tool hasn’t significantly impacted them, as casual use cases like translating webpages or social media posts are typically not outsourced to professionals.

The Need for Precision Beyond "Good Enough"

In these common scenarios, a basic level of accuracy is often sufficient. However, when launching a product in multiple international markets, it’s crucial that instructions, warnings, legal agreements, and technical documentation are not merely adequate in some languages, but consistently perfect across all of them.

Lengoo's Approach: Combining Human Expertise with Machine Learning

Lengoo initially focused on streamlining the workflow between companies and human translators.

“Our subsequent step was naturally to automate the translation process itself,” explained CEO and founder Christopher Kränzler. “While human involvement will remain essential for the foreseeable future, our aim is to refine the models to a point where they are genuinely practical, reducing the workload for human translators.”

Continuous advancements in machine learning make this goal increasingly attainable. Companies like DeepL and Lilt have demonstrated significant improvements over Google and Microsoft’s frameworks, while acknowledging the continued need for human oversight.

Speed and Specificity: The Lengoo Advantage

Lengoo differentiates itself by prioritizing both speed and specificity. They develop language models tailored to each client’s unique jargon, stylistic preferences, and formatting requirements. This is achieved by training the models using the client’s existing documents and websites, and continuously incorporating feedback from the translation process.

“We have an automated training pipeline for our models,” Kränzler stated. “The more contributions to the correction process, the faster the improvement. Ultimately, we achieve speeds approximately three times faster than Google or DeepL.”

A new client benefits from a model initially customized using documents from the past few years. Whenever the model’s output requires correction, that specific adjustment is learned and integrated into the model’s overall training data.

Quantifying Translation Quality

Determining the “quality” of a translation can be subjective. However, Lengoo’s approach, as a tool for human translators, incorporates a built-in quality check. Translation quality is measured by “correction distance” – the extent of changes a human translator must make to the model’s suggested text. Reduced corrections indicate both a better and a faster translation, providing objective metrics for both quality and speed.

These improvements have convinced clients previously hesitant about extensive automation.

“Initially, there was some resistance,” Kränzler admitted. “People are familiar with Google Translate for everyday use and have observed its improving quality – alongside DeepL, they’ve been educating the market. There’s now a general understanding that, when implemented correctly, machine translation is viable for professional applications. A large client might employ 30, 40, or 50 translators, each with their own style… We can demonstrate that we are faster and more cost-effective, while also enhancing consistency.”

Future Development and Expansion

While customizing models with client data isn’t novel, Lengoo has established a competitive edge over larger, slower companies that struggle to rapidly improve their products. They plan to solidify this position by upgrading their technology infrastructure.

Currently, their reliance on conventional machine learning technologies limits the speed of the translator-AI feedback loop. Retraining large models is computationally expensive and therefore infrequent.

Lengoo intends to develop its own, more responsive neural machine translation framework that streamlines the various processes involved. This wouldn’t result in real-time improvements, but would integrate new information much more quickly and efficiently.

“Consider it as incremental improvement, segment by segment,” explained applied research lead Ahmad Taie (segments vary in size but generally represent logical text units). “You translate one segment, and the model incorporates the improvements by the time you reach the next one.”

Enhancing this core product feature, making it faster and easier to implement for each client, is crucial for client retention. Kränzler anticipates competition will likely come from agile startups rather than established giants like Google, which tend to favor acquisition strategies.

The Future of Translation and Translators

The role of human expert translators won’t be eliminated, but their effectiveness may be significantly amplified – potentially by an order of magnitude. This could impact the size of the translation workforce, but continued growth in international markets and the demand for professional translation may offset these changes.

The $20 million funding round, led by Inkef Capital, will enable Lengoo to expand into North American and additional European markets, and integrate with more enterprise systems. Existing investors Redalpine, Creathor Ventures, Techstars, and angels Matthias Hilpert and Michael Schmitt participated, alongside new investors Polipo Ventures and Volker Pyrtek.