Main_Translation-Startup-Lengoo-Raises-USD-20-million-in-Series-B-Funding.png

Lengoo Raises USD 20m Series B From Inkef on AI Agency Investment Thesis – via slator.com

On February 10, 2021, Lengoo announced that it had closed a USD 20m Series B round led by Amsterdam-based Inkef Capital. This brings total funds raised by the Berlin-based language […]

-


Lengoo Raises USD 20m Series B From Inkef on AI Agency Investment Thesis

On February 10, 2021, Lengoo announced that it had closed a USD 20m Series B round led by Amsterdam-based Inkef Capital. This brings total funds raised by the Berlin-based language service provider to USD 34m.

The Series B, which closed on January 29, 2021, drew existing investors Redalpine, Creathor Ventures, and Techstars, as well as angel investors Matthias Hilpert and Michael Schmitt. New investors included Polipo Ventures and Volker Pyrtek, Senior Adviser at Inkef.

Lengoo CEO, Christopher Kränzler, described raising funds in the current (Covid) environment as “different, but doable thanks to Zoom and Google Hangouts.” He told Slator, “It still feels strange to not be able to meet new investors in person, but we all hope that we can catch up on that soon.”

Advertisement

Inkef is a Europe-focused VC firm that has backed over 40 healthcare and technology companies. On what attracted Inkef to invest in Lengoo, Frank Lansink, Operating Partner at Inkef said, “Lengoo is a tech company at its core. Hence, for us the experience of investing in technology companies was far more important than the experience in the field of language services.”

According to Lansink, Inkef views recent developments in machine learning models “as another form of computation that requires new business models where humans and machines partner up to deliver its potential to the fullest capacity.” Lansink’s investment case is reminiscent of the AI Agency thesis, which Redpoint Ventures Partner, Tomasz Tunguz, discussed at SlatorCon San Francisco in 2019.

Lengoo said in a press statement that it will use the funds to “build a global presence for […] globally active clients” and further develop their proprietary machine translation (MT) system to improve speed and scalability.

“We find that on output from well-trained [MT] systems, Lengoo translators make no changes to approximately 40% of new material”

Kränzler told Slator, “Existing [MT] frameworks do not allow us to respond to our customers’ needs with sufficient flexibility and speed. We need a more coherent system where the parts cater to potentially various input signals — a Lengoo nervous system if you will. We have more ambitious goals than just training a customer translation model in isolation and then serving that.”

He added, “Historically, machine translation was the main part of our industry; going forward, it will be in almost all interactions.”

All Boils Down to Edit Distance

CEO Kränzler said Lengoo primarily targets multilingual enterprises. He said they combine “a machine learning approach on our client’s language data with human revision of machine translated output.” The result, translations that are “50% cheaper and three times faster than traditional providers.”

“Humans take center stage in our workflow”

Asked how they calculate cost savings, Kränzler said they base it on a client’s previous translation cost: “We literally ask our clients what they paid before and compare that to what they spend on translation with Lengoo.”

As for measuring translation speed, he said, “We run frequent in-house experiments comparing the revision for generic MT, domain-specific MT, and Lengoo custom-trained MT.”

Kränzler pointed out that measuring how effective machine translation (MT) output is requires data tied to a use case. For Lengoo, it means measuring the time a translator needs to edit “machine-generated draft translations and the level of correction required” to meet a client’s quality and terminology requirements. In short, the time and effort it takes to post-edit machine translations (a.k.a. PEMT or MTPE).

“To measure the level of correction, we use Edit Distance, which counts the number of insertions, deletions, and substitutions made to the MT output by the translator,” Kränzler said.

“Ensuring that [translations] are correctly rendered in the target language is a highly cognitive task”

He added, “We find that on output from well-trained [MT] systems, Lengoo translators make no changes to approximately 40% of new material.”

Augmented Translation, Two Ways

Kränzler described Lengoo’s approach to training MT models as “ultra-customization.” He said, “We need about 150,000 translated words, or 1,000 A4 pages per use case and per language, to initialize the system — usually no problem for our enterprise customers.”

Kränzler summarized the approach into three “automated” steps.

  • Cleaning / preparing language data from the client, augmenting it with publicly available data
  • Modeling data so it adheres to the client’s rules
  • Creating and deploying the models at scale

Asked where humans sit in the loop, Kränzler said, “Humans take center stage in our workflow. Translators make changes to the machine translation output, the machine keeps learning from the human and, in turn, the machine translation becomes better and better” — something he referred to as “augmented translation.”

In-Line_Translation Startup Lengoo Raises USD 20 million in Series B Funding

He said, “Optimally, the augmentation goes both ways,” allowing translators to “focus on highly nuanced parts of language […] what they are truly interested in: the meaning of language.”

He further pointed out, “Ensuring that [translations] are correctly rendered in the target language is a highly cognitive task.” And while MT “often provides surprisingly fluent translations […] we cannot be confident yet that those translations are actually accurate. There still needs to be an expert revision. It’s very necessary to have translators who understand all the semantic and stylistic nuances of source content and ensure that it is present in the translation. That remains and will remain a critical skill.”

[MT] often provides surprisingly fluent translations, but we cannot be confident yet that those translations are actually accurate. There still needs to be an expert revision”

According to Kränzler, “We have focused mainly on the processes and steps in the field of data preparation and modeling in the past and have manipulated existing [machine learning] technology accordingly. In the next step, we will now develop a proprietary [MT] framework that brings everything together.”

‘Very Large Clients’

Lengoo increased its ARR sixfold over the past 12 months, according to the press statement, and onboarded “50+ enterprise customers in Europe and the US including National Instruments, Sunrise Communications, Sixt, and the WWF.”

The same statement said the company also expanded into the US, the UK, Scandinavia, and Poland and tripled its headcount, which currently stands at 100 across several locations in Germany, Switzerland, Poland, the UK, and Sweden.

Kränzler told Slator, “We focus on very large clients that have an inherently high demand for translation and localization, such as global software corporations, e-commerce businesses, manufacturing companies active in exporting, or highly technical companies with a tremendous amount of documentation.”

He said near- to mid-term plans include further expanding within Europe as well as building a presence in North America to deliver “a 24/7 offering for all of our globally active clients.”

Images: Lengoo founders (from left) Philipp Koch-Buettner, Christopher Kränzler, Alexander Gigga; inline photo courtesy of Lengoo



Source link

20mAgencyfeedInkefInvestmentLengooRaisesSeriesslatorcomThesisUSD