We wanted to know where the innovation is happening along the complex supply chain in language services and technology. So we reached out to six of the most intriguing startups for an exclusive look at what their founders think is the next big thing. What we discovered is that startups are targeting pain-points throughout the supply chain.
We saw innovation that seeks to accelerate human translation speed, cut out the project manager, guess what the translator wants to type next, make life easier for app developers, and package many existing innovations into a neat new bundle of buyer joy. We intentionally left out pure-play machine translation startups and have saved them for future coverage.
It is too early to tell which of these models if any is going to break through and define the business in the years to come. What we do know, however, is that the incredible pace of innovation bodes well for the sector’s growth in the future.
Skuuper: Do-It-Yourself Translation for Bilingual Staff
Skuuper is the youngest startup listed here. Founded in June 2015, the Tallinn, Estonia-based startup also has a presence in San Jose, Moscow, and St. Petersburg. It was founded by CEO Raul Malmstein and CTO Oleg Urzhumtsev, who co-own the company with its investors. They currently employ eight full-time equivalents. The company’s product is still in prototype stage and they have so far raised over USD 225,000 in seed funding from undisclosed angel investors.
Skuuper makes translation management systems (TMS) and computer-aided translation (CAT) tools accessible for Do-It-Yourself use. Targeting the long-tail of translation and potential clients that already employ bilingually proficient staff, Skuuper provides the tools to do small scale translation and localization tasks.
The startup allows companies to leverage bilingual employees for DIY translation through affordable TMS and CAT tools. “Skuuper is a Software-as-a-Service (SaaS) with no learning curve and a modest subscription fee,” Raul told Slator.
“Our target is to make [translation tools] affordable and intuitive, so users do not get confused by an abundance of hard-to-learn functions, and to decrease the price from USD 500 to 1,000 to a small subscription fee of EUR 2 to EUR 30,” he said.
Skuuper’s key technologies are translation memory (TM) and fuzzy matching. It has its own translation bank with five million translated phrases, although users can utilize their own TM. Skuuper currently supports translations into six languages including the pairs English-Estonian, English-Russian, and Estonian-Russian. Raul said the next pair will be Chinese-English.
He said Skuuper supplies “demand from the underserved segment of the market [that] cannot afford to invest money in luxuriously priced mainstream tools and staff training.”
They are not directly challenging major vendors. At least, not yet.
Although Skuuper currently focuses on providing DIY translation tools to bilingual people, their ambition is a market-matching, work-intermediation platform connecting clients to freelancers. Their current toolset remains core, with clients being able to outsource work to freelancers.
Skuuper is in a few preliminary agreements for alpha and beta testing with some big financial corporates that the team is hoping will stay on board as actual clients. They are convinced the intuitive interface and affordable pricing differentiate Skuuper from competitors and, also, as Raul put it, “Because we reveal the concept of translation memory to corporates, turning their dead content into a profitable asset.”
Making TMs accessible to corporates is not earth-shattering innovation, and a low-cost subscription-based CAT tool on the cloud will need a lot of marketing dollars to reach potential users. But like others, Skuuper’s product is targeting the long tail of demand, which is so vast that even a small piece of the pie could turn out to be big business.
Fluently: Cut Out the Middle-Man
The startup business is majority owned by co-founders CEO Karin Nielsen and CTO Giovanni Charles. The London-based company has so far received support from startup bootcamp Entrepreneur First and an undisclosed amount from unnamed angel investors.
Fluently’s goal is to use technology to allow users access to direct market-matching sans the need for what they call “middlemen agencies”―traditional language service providers. As is fashionable these days, Karin likened their centralized marketplace to Uber; they set the rates and standards, and then shortlist the most suitable, vetted translators for client tasks.
Fluently will launch officially in March 2016, Karin said and so far their key technology is focused on making sure there is a perfect match between ideal translator and translation task.
“For the time being, our focus is on market-matching since this is, in itself, a difficult nut to crack, and one where computation has significant advantages over humans,” Karin said. Their hypothesis was that many translation-quality issues can be avoided by focusing on perfect market-matching upfront. Fluently uses many data points, including demographic data, to make inferences about a translator’s suitability for a given task. Karin claims this allows higher quality while also enabling translators to work on their ideal projects.
Fluently’s private beta supports English to five European languages, and the startup is planning 20 language pairs for the public beta launch.
Karin explained that many platforms, which have emerged in the language industry in the past two to three years, have focused on workflow. “The real value lies in the translation,” Karin argued, “everything else is a vitamin, not a pill. I remain less optimistic about the long-term viability of any business model that focuses exclusively on workflow.”
Fluently has been working with both startups and established companies, and supply-side interest is also high. “The waiting list for vetting translators is now running into the thousands,” Karin said.
Asked how Fluently would approach project management if it wants to get rid of the “middlemen agencies” that provide it, Karin said most customers project manage their agencies and are thus effectively paying for project management twice in the first place. “The assertion that customers are incapable of managing their own translation projects is, frankly, plain arrogant,” she asserted.
However, she noted that “highly regulated verticals such as life sciences are segments that can genuinely benefit from agency competence in their niche.” But, in general, Karin believes that “we are already seeing a seismic shift in purchasing behaviors, which will ultimately spell the end for many small-to-medium-sized agencies that are not outcome-focused, agile, or technologically competent enough to provide a unique user experience.”
“I’d like to think that we are doing our bit for empowering customers and translators to become more independent so they can work together to achieve great results,” she said, “We want to demystify localization, making it a more accessible endeavor for the layman.”
Market-matching is an intriguing concept that has yet to prove it works at scale. That is what Duolingo realized, as it pulled out of translation saying it would “have to hire people focused on quality control, sales people, etc.” to scale it.
Lilt: I Know What You are Thinking. MT Powered Predictive Typing
Back in July 2015, we reached out to Spence Green, a PhD student in Computer Science at Stanford University, about one of the papers he wrote with peers, “Human Effort and Machine Learnability in Computer Aided Translation and Translation Memory: A Mixed-Initiative System for Human Language Translation.”
We never had that discussion. It turns out he was busy starting up Lilt with colleague John DeNero, an assistant professor of computer science at University of California, Berkeley.
Established in March 2015, the privately-owned startup has so far raised USD 650,000 in pre-seed funding from XSeed Capital. Based in Palo Alto, California, Lilt employs five full-time equivalents.
Lilt offers predictive CAT―sort of like the iPhone’s autocorrect feature―with the technology to train (in real-time) the underlying TMS to “learn” user input. Not only will Lilt’s CAT tool predict what translation users want, the entire TMS will also adapt to user specialization, vertical-specific translation patterns, and style as they work.
“Newer companies are certainly more interested in machine translation technology,” Spence said about demand for Lilt’s product, at the core of which is predictive typing and model adaptation.
Asked about the core tech that is obviously key to Lilt’s potential, Spence explained the technologies further: “Lilt offers two products: a browser-based CAT tool, and an [application programming interface] API, through which TMSs can access the adaptive machine translation backend.” These technologies allow translation predictions at typing speed as well as real-time model personalization; the platform gets smarter as it is used.
Currently, Lilt supports the language pairs English-German, -French, -Spanish, and -Portuguese, with English-Italian and -Dutch in the works.
The tech Lilt is developing seems to be something of interest to direct clients, language service providers and translators alike. At its core, it is meant to improve a translator’s productivity. So it is no surprise that the company is busy making a name for itself in the freelance translator community through sponsored webinars and promoted tweets.
Spence highlighted what he thinks is their differentiator: “Lilt is the first and only production MT system that learns from feedback in real-time.” He tempered this enthusiasm, however, with realistic expectations as regards competition: “We’re the first, but certainly not the last, company to deploy this technology.”
He said neither translators nor vendors appreciate the coming improvements to machine translation over the next couple of years, which he likened to the vast improvements introduced by phrase-based systems in the early 2000s.
“We’ve been evaluating interactive machine translation versus conventional TM as implemented by the legacy CAT tools. The productivity improvements are significant,” Spence said. He added, “We’ll release a case study in the next few weeks that, I hope, will shift discussions about CAT systems toward empirical results and away from marketing pitches.”
If those empirical results actually do prove a significant productivity boost―think 600-700 words per hour without the translator going blind―Lilt could find subscriptions flying off the shelf.
Unbabel: Going All-In on MTPE
Unbabel is a bit older than the first three startups on this list. Founded in August 2013, Unbabel has five co-founders who are also majority shareholders: CEO Vasco Pedro, PhD, CTO João Graça, PhD, CMO Sofia Pessanha, Bruno Silva, and Hugo Silva. The company is headquartered in San Francisco with another office Lisbon. It employs 26 full-time equivalents and have built a freelance pool of 40,000 translators, 15% of whom can access paid customer requests.
Unbabel has so far raised USD 1.5m from international venture capitalists and business angels, including Y Combinator, Matrix Partners, IDG, and Digital Garage.
“At a basic level, we’re a machine translation post-editing company,” Sofia said. Unbabel uses a crowdsourcing model that market-matches clients with translators, but the intelligent translation engine underlying their product allows machine learning to continuously improve the TMS the more it is used.
It is machine translation post-editing of crowdsourced translations where the TMS gets increasingly smarter and more customized to the user’s style and field of specialization.
Unbabel provides clients with a sandbox API that can be customized for systems that the team has not built a ready-made API for, such as Zendesk and MailChimp. The APIs help Unbabel focus on companies with online, digital content. The company translates mostly from English into French, Italian, German, Spanish, Portuguese, and Brazilian Portuguese, although the tech supports 22 languages and 45 language pairs.
Clients send them translation jobs and Unbabel turns these into micro-tasks, which are then run through machine translation. Post-editing work is distributed to ideal editors (translators), who can work even on their smartphones. All the while, the Smartcheck feature allows the translation engine to get smarter from experience.
Unbabel says it has over 400 paying customers, including some listed on their website such as Eventbrite, Trello, and Weebly. According to the company, revenues grew 300% from 2014 to 2015. In September 2015, CEO Vasco Pedro went on record as saying it was the first month Unbabel surpassed USD 100,000 in monthly revenue; 80% of its revenue comes from North America and Europe. Unbabel charges monthly fees of USD 110 for 2,000 words, USD 450 for 9,000 words, and USD 900 for 20,000 words, with respective excess word pricing.
As for growth, Sofia said Unbabel has grown 50% month-over-month since September 2015, expects 2016 to be much busier, and that they are open to partnerships and acquisitions. “Unbabel is uniquely positioned to deliver MTPE services at scale and we’re open to being the MTPE arm of more traditional LSPs,” Sofia said. “We’re very open to talking about it and making it happen―email me!” she added.
In a sense similar to Lilt, Unbabel’s promise it to radically enhance the speed and efficiency of human translation. If they succeed, word prices will come tumbling down and cause a headache for less tech-savvy incumbents.
El Loco: Visualize That Multilingual App
El Loco is another startup that is not so young, although it only became very active in 2014 (the founding year indicated on its LinkedIn page). Founded in December 2011, the California-based startup is an LLC with seven staff. It has so far raised USD 3.5m from unnamed individual investors.
El Loco CEO Kee Nethery conceived the startup when he saw that apps localized in the five top languages saw significant sales increases in their sales platform, Kagi, where Kee is also CEO.
El Loco encourages iOS mobile app developers to localize by trying to make the process easier. Developers send requests to the El Loco portal and translators work on translations in browsers, while also having access to real-time, simulated views of the translated mobile app for visual context.
El Loco says it speeds up translation work with less back-and-forth communication thanks to improved visual context. They want to enable developers to tap into localized mobile app sales with low overhead costs and minimal effort. El Loco also analyzes and adjusts source codes to support the localization lifecycle.
El Loco’s technology has three components: the Mac app, the iOS library, and the website. The library is embedded in the app to capture screens and strings as they are displayed by the Simulator. Those captured screens and strings are what the Mac app uploads to the website for translators for a real-time visual context of what they are translating.
“Our goal is to enable the translators to complete the translation within a single round trip,” Kee said. “To do it correctly the first time, not the three or more round trips that are common when the developer has to check the translator’s work for text that gets truncated by the app interface.”
Kee said current users typically translate to Simplified Chinese and Spanish. He said, “Apps that generate revenue want access to mainland China. Apps that support existing businesses in the US tend to want to service their Spanish-speaking customers.”
Currently, El Loco users are developers who work for software companies. The idea is that they try the app free of charge for private projects, but eventually go on to recommend the tool for their company’s mobile app localization efforts.
Their free tier allows one user one app, two languages, and 10,000 managed words. Their paid tiers are USD 20 monthly for 10 users, three apps, and 20,000 words in unlimited languages, and USD 100 monthly for 10 users, unlimited apps and languages, and 60,000 words.
“Developers don’t bother with localization because it’s a major hassle,” Kee said, “As far as I’m concerned, any translation is better than none.”
Kee argued that users would welcome any translation better than machine translation if they truly want an app, even if it was localized by non-professional translators. “El Loco provides the infrastructure so that a developer can manage their translators [even if they are not professionals]” he said.
This is easier said than done, however. Kee explained, “The first really big challenge was gathering the visual context from the app. After lots of trial and error, we came up with a solution, and then went out for funding to build the business. The current big challenge is getting developers past their fear of localizing.”
El Loco hopes their freemium model can show developers the true potential of mobile app localization. “We believe developers should start with free translations from their family and friends in any language; just start,” Kee said. “When the app earns enough to prove the monetary benefits of additional languages, we feel developers will shift to professionals.”
What makes El Loco somewhat different from many other startups in the space is that, if its technology works and breaks through with developers, the company can create significant demand where there may not have been any before.
TextMaster: What If You Had No Legacy Technology?
The most mature company and also the one with probably the broadest model on the list is TextMaster. Started in 2011, the company is owned by its founders, Thibaud Elziere, Quentin Nickmans, and Alexandre Ponsin, the current CEO, Thibault Lougnon, as well as a pool of investors including venture capital firms Serena Capital, Alven Capital, and eFounders. It is also supported by the founders of OLX, Empruntis, and Webedia, according to CEO Lougnon.
The company’s 35 full-time equivalents are spread across its main offices in Paris and Brussels, and others in US, Germany, and Spain. TextMaster has so far raised USD 7.82m from the above individual investors and companies.
TextMaster claims it represents the way a full-service LSP should be if it were built using the tech of today. No legacy tech. No highly bespoke processes that prove hard to change. Thibault said TextMaster improves upon the traditional translation process by automating “useless and potentially faulty manual processes.”
TextMaster offers the same services as a traditional vendor, with a technological edge. They use proprietary and cloud-based TMS and CAT tools, a market-matching platform, APIs, and plugins. TextMaster focuses on translation services and partners up with other vendors for localization, such as PhraseApp, Transifex, Localize, and LingoHub. Thibault declined to specify verticals and languages, but said they mostly focus on European languages and clients.
The Market Potential
TextMaster offers two tiers of pricing: standard for USD 0.066 per word and enterprise for USD 0.132 per word. Additional options such as expedited delivery and additional proofreading are available for extra per-word fees.
TextMaster claims it has grown 150% year-on-year for three years and saw 130 million words translated in the company’s history, almost half of which was in 2015. While Thibault did not directly name key clients, he disclosed that TextMaster was working with “one of the leading global petroleum companies, several famous luxury brands, some major banks and leading tech companies.” Some large clients translate over five millions words per year, he said, adding that the next step for them is to secure multimillion-dollar yearly contracts.
TextMaster’s clients are typically companies that have been using other vendors before, Thibault said. He claimed they get the same quality and project management, but with added transparency and efficiency because of TextMaster’s technology.
Asked about plans for growth, Thibault said they plan to double the size of their team in the next 18 months to support expansion plans, which are focused on areas in Europe other than France, Italy, Germany, and Spain where the company is already established. Asia and the US are also on their radar. Thibault said they were also open to acquiring companies to accelerate expansion in Europe.
A challenge to TextMaster’s model may come once the company starts to sign up the USD 1m+ a year accounts. How would they sacrifice a major new client that requests full customization and use of competing technology for the sake of process purity?