Connect with us


With AI translation service that rivals professionals, Lengoo attracts new $20M round



Most people who use AI-powered translation tools do so for commonplace, relatively unimportant tasks like understanding a single phrase or quote. Those basic services won’t do for an enterprise offering technical documents in 15 languages — but Lengoo’s custom machine translation models might just do the trick. And with a new $20M B round, they may be able to build a considerable lead.

The translation business is a big one, in the billions, and isn’t going anywhere. It’s simply too common a task to need to release a document, piece of software, or live website in multiple languages — perhaps dozens.

These days that work is done by translation agencies, which employ expert speakers to provide translation on demand at a high level of quality. The rise of machine translation as an everyday tool hasn’t affected them as much as you might think, since the occasional Portuguese user using Google’s built-in webpage translation on a Korean website is very much a niche case, and things like translating social media posts or individual sentences isn’t really something you could or would farm out to professionals.

In these familiar cases “good enough” is the rule, since the bare meaning is all anyone really wants or needs. But if you’re releasing a product in 10 different markets speaking 10 different languages, it won’t do to have the instructions, warnings, legal agreements, or technical documentation perfect in one language and merely fine in the other nine.

Lengoo started from a team working on automating that workflow between companies and translators.

“The next step to take obviously was automating the translation itself,” said CEO and founder Christopher Kränzler. “We’ll still need humans in the loop for a long time — the goal is to get the models to the level where’s they’re actually usable and the human has fewer translations to make.”

With machine learning capabilities constantly being improved, that’s not an unrealistic goal at all. Other companies have started down that road — DeepL and Lilt, for instance, which made their cases by showing major improvements over Google and Microsoft frameworks, but never claiming to remove humans from the process.

Lengoo iterates on their work by focusing on speed and specificity — that is, making a language model that integrates all the jargon, stylistic preferences, and formatting requirements of a given client. To do this they make a custom language model by training it not just with the customer’s own documents and websites, but by continually adding in feedback from the translation process itself.

Illustration showing an infinity sign on which are various components of the machine learning feedback process.

A fanciful representation of the self-improving model process.

“We have an automated training pipeline for the models,” said Kränzler. The more people contribute to the correction process, the faster the process gets. Eventually we get to be about three times faster than Google or DeepL.”

A new client may start with a model customized on a few thousand documents from the last couple years. But whenever the model produces text that needs to be corrected, it remembers that particular correction and integrates it with the rest of its training.

Diagram showing how fewer corrections are needed after the AI receives additional feedback.

An exciting bar graph. After 30 iterations, the segments requiring no corrections have doubled, and those requiring few are much increased.

While the “quality” of a translation can be difficult to quantify objectively, in this case there’s no problem, because working as a human translator’s tool means there’s a quality check built right in. How good the translation is can be measured by “correction distance,” essentially the amount of changes the human has to make to the model’s suggested text. Fewer corrections not only means a better translation, but a faster one, meaning quality and speed both have objective metrics.

The improvements have won over customers that were leery of over-automation in the past.

“At the beginning there was resistance,” admitted Kränzler. “People turn to Google Translate for everyday translations, and they see the quality is getting better — they and DeepL have been educating the market, really. People understand now that if you do it right, machine translation works in a professional use case. A big customer may have 30, 40, 50 translators, and they each have their own style… We can make the point that we’re faster and cheaper, but also that the quality, in terms of consistency, goes up.”

Although customizing a model with a client’s data is hardly a unique approach, Lengoo seems to have built a lead over rivals and slower large companies that can’t improve their products quick enough to keep up. And they intend to solidify that lead by revamping their tech stack.

The issue is that due to relying on more or less traditional machine learning technologies, the crucial translator-AI feedback loop is limited. How quickly the model is updated depends on how much use it gets, but you’re not going to retrain a large model just to integrate a few hundred more words’ worth of content. It’s expensive computationally to retrain, so it can only be done sporadically.

But Lengoo plans to build its own, more responsive neural machine translation framework that integrates the various pipelines and processes involved. The result wouldn’t improve in real time, exactly, but would include the newest information in a much quicker and less involved way.

“Think of it as a segment by segment improvement,” said applied research lead Ahmad Taie (segments vary in size but generally are logical “chunks” of text). “You translate one segment, and by the next one, you already have the improvements made to the model.”

Making that key product feature better, faster, and easier to implement customer by customer is key to keeping clients on the hook, of course. And while there will likely be intense competition in this space, Kranzler doesn’t expect it to come from Google or any existing large companies, which tend to pursue an acquire-and-integrate approach rather than an agile development one.

As for the human expert translators, the field won’t replace them but may extend their effectiveness by, eventually, as much as an order of magnitude, which may shrink the workforce there. But if international markets continue to grow and with them the need for professional translation, they might just keep pace.

The $20M round, led by Inkef Capital will allow Lengoo to make the jump to North American markets as well as additional European ones, and integrate with more enterprise stacks. Existing investors Redalpine, Creathor Ventures, Techstars (out of which program the company originated), and angels Matthias Hilpert and Michael Schmitt all joined in the round, along with new investors Polipo Ventures and Volker Pyrtek.

Lyron Foster is a Hawaii based African American Musician, Author, Actor, Blogger, Filmmaker, Philanthropist and Multinational Serial Tech Entrepreneur.

Continue Reading


Snowflake latest enterprise company to feel Wall Street’s wrath after good quarter



Snowflake reported earnings this week, and the results look strong with revenue more than doubling year-over-year.

However, while the company’s fourth quarter revenue rose 117% to $190.5 million, it apparently wasn’t good enough for investors, who have sent the company’s stock tumbling since it reported Wednesday after the bell.

It was similar to the reaction that Salesforce received from Wall Street last week after it announced a positive earnings report. Snowflake’s stock closed down around 4% today, a recovery compared to its midday lows when it was off nearly 12%.

Why the declines? Wall Street’s reaction to earnings can lean more on what a company will do next more than its most recent results. But Snowflake’s guidance for its current quarter appeared strong as well, with a predicted $195 million to $200 million in revenue, numbers in line with analysts’ expectations.

Sounds good, right? Apparently being in line with analyst expectations isn’t good enough for investors for certain companies. You see, it didn’t exceed the stated expectations, so the results must be bad. I am not sure how meeting expectations is as good as a miss, but there you are.

It’s worth noting of course that tech stocks have taken a beating so far in 2021. And as my colleague Alex Wilhelm reported this morning, that trend only got worse this week. Consider that the tech-heavy Nasdaq is down 11.4% from its 52-week high, so perhaps investors are flogging everyone and Snowflake is merely caught up in the punishment.

Snowflake CEO Frank Slootman pointed out in the earnings call this week that Snowflake is well positioned, something proven by the fact that his company has removed the data limitations of on-prem infrastructure. The beauty of the cloud is limitless resources, and that forces the company to help customers manage consumption instead of usage, an evolution that works in Snowflake’s favor.

“The big change in paradigm is that historically in on-premise data centers, people have to manage capacity. And now they don’t manage capacity anymore, but they need to manage consumption. And that’s a new thing for — not for everybody but for most people — and people that are in the public cloud. I have gotten used to the notion of consumption obviously because it applies equally to the infrastructure clouds,” Slootman said in the earnings call.

Snowflake has to manage expectations, something that translated into a dozen customers paying $5 million or more per month to Snowflake. That’s a nice chunk of change by any measure. It’s also clear that while there is a clear tilt toward the cloud, the amount of data that has been moved there is still a small percentage of overall enterprise workloads, meaning there is lots of growth opportunity for Snowflake.

What’s more, Snowflake executives pointed out that there is a significant ramp up time for customers as they shift data into the Snowflake data lake, but before they push the consumption button. That means that as long as customers continue to move data onto Snowflake’s platform, they will pay more over time, even if it will take time for new clients to get started.

So why is Snowflake’s quarterly percentage growth not expanding? Well, as a company gets to the size of Snowflake, it gets harder to maintain those gaudy percentage growth numbers as the law of large numbers begins to kick in.

I’m not here to tell Wall Street investors how to do their job, anymore than I would expect them to tell me how to do mine. But when you look at the company’s overall financial picture, the amount of untapped cloud potential and the nature of Snowflake’s approach to billing, it’s hard not to be positive about this company’s outlook, regardless of the reaction of investors in the short term.

Continue Reading


A first look at Coursera’s S-1 filing



After TechCrunch broke the news yesterday that Coursera was planning to file its S-1 today, the edtech company officially dropped the document Friday evening.

Coursera was last valued at $2.4 billion by the private markets, when it most recently raised a Series F round in October 2020 that was worth $130 million.

Coursera’s S-1 filing offers a glimpse into the finances of how an edtech company, accelerated by the pandemic, performed over the past year. It paints a picture of growth, albeit one that came at steep expense.


In 2020, Coursera saw $293.5 million in revenue. That’s a roughly 59% increase from the year prior when the company recorded $184.4 million in top line. During that same period, Coursera posted a net loss of nearly $67 million, up 46% from the previous year’s $46.7 million net deficit.

Notably the company had roughly the same noncash, share-based compensation expenses in both years. Even if we allow the company to judge its profitability on an adjusted EBITDA basis, Coursera’s losses still rose from 2019 to 2020, expanding from $26.9 million to $39.8 million.

To understand the difference between net losses and adjusted losses it’s worth unpacking the EBITDA acronym. Standing for “earnings before interest, taxes, depreciation and amortization,” EBITDA strips out some nonoperating costs to give investors a possible better picture of the continuing health of a business, without getting caught up in accounting nuance. Adjusted EBITDA takes the concept one step further, also removing the noncash cost of share-based compensation, and in an even more cheeky move, in this case also deducts “payroll tax expense related to stock-based activities” as well.

For our purposes, even when we grade Coursera’s profitability on a very polite curve it still winds up generating stiff losses. Indeed, the company’s adjusted EBITDA as a percentage of revenue — a way of determining profitability in contrast to revenue — barely improved from a 2019 result of -15% to -14% in 2020.

Continue Reading


The owner of Anki’s assets plans to relaunch Cozmo and Vector this year



Good robots don’t die — they just have their assets sold off to the highest bidder. Digital Dream Labs was there to sweep up IP in the wake of Anki’s premature implosion, back in 2019. The Pittsburgh-based edtech company had initially planned to relaunch Vector and Cozmo at some point in 2020, launching a Kickstarter campaign in March of last year.

The company eventually raised $1.8 million on the crowdfunding site, and today announced plans to deliver on the overdue relaunch, courtesy of a new distributor.

“There is a tremendous demand for these robots,” CEO Jacob Hanchar said in a release. “This partnership will complement the work our teams are already doing to relaunch these products and will ensure that Cozmo and Vector are on shelves for the holidays.”

I don’t doubt that a lot of folks are looking to get their hands on the robots. Cozmo, in particular, was well-received, and sold reasonably well — but ultimately (and in spite of a lot of funding), the company couldn’t avoid the fate that’s befallen many a robotics startup.

It will be fascinating to see how these machines look when they’re reintroduced. Anki invested tremendous resources into bringing them to life, including the hiring of ex-Pixar and DreamWorks staff to make the robots more lifelike. A lot of thought went into giving the robots a distinct personality, whereas, for instance, Vector’s new owners are making the robot open-source. Cozmo, meanwhile, will have programmable functionality through the company’s app.

It could certainly be an interesting play for the STEM market that companies like Sphero are approaching. It has become a fairly crowded space, but at least Anki’s new owners are building on top of a solid foundation, with the fascinating and emotionally complex toy robots their predecessors created.

Continue Reading