SlatorCon 2019 — Your Definitive Guide and Key Takeaways

SlatorCon 2019—您的最终指南和关键 Takaways

2019-12-31 18:20 slator

本文共4366个字,阅读需44分钟

阅读模式 切换至中文

The SlatorCon tour took in three global cities across two continents in 2019 and featured a total of 28 speakers representing the full breadth of the language industry — from startups, language technology and language service providers to the buy-side and finance. Here, we revisit the key takeaways from our events calendar in 2019. SlatorCon London – May 2019SlatorCon San Francisco – September 2019SlatorCon Amsterdam – November 2019 Jean Senellart, CEO, Systran When Systran’s Jean Senellart took the stage at SlatorCon London, it was not the first time. The machine translation (MT) veteran made his debut at SlatorCon London 2018, taking participants on a whistle-stop tour of MT history; from rules-based MT to current-state neural MT (NMT), and much more. In May 2019, Senellart drew attention to Google’s self-attentional transformer NMT model. First introduced in 2017, it has since become the leading model in both research and deployment, and no technology has managed to supersede it in two years, he said. Senellart also highlighted the current hot topics for machine translation researchers, including the interaction between MT and TM, post-editing, and low resource languages. The field of unsupervised learning, Senellart said, has the potential to greatly impact MT output quality. The Systran CEO gave a word of advice to the MT community, reminding the SlatorCon audience that the MT touch-point for customers is the product rather than the technology: “We’re talking about the technology like it’s the thing that the user wants,” he said. “They don’t want the technology, they want the product.” Mark Brayan, CEO, Appen A layer removed from MT is the world of language data, a booming niche of the language industry, also known as “AI support services.” Language data is used to train AI and machine applications such as MT engines and much more. At the heart of the language data space is Australia-listed company Appen, which makes use of its huge crowd of remote workers to collect, process, and label massive amounts of data. “The more data the better,” Appen CEO Mark Brayan said, adding that one needs to keep in mind that “data is expensive, complicated to collect, and complicated to work with.” While Appen’s operations rely on a vast army of human annotators to create and check data that will eventually be used for speech and image recognition tools, it is more than a numbers game. The breadth, variety (and, ultimately, quality) of the data also matters greatly. To get the best results from speech data, for example, factors such as accent, age, acoustic setting, and so on must suit the specific use case of the customer, Brayan said. Andrew Bredenkamp, Founder and CEO, Acrolinx Also well versed in the rapidly evolving data labeling market for AI is Acrolinx’s Bredenkamp, who told SlatorCon Amsterdam participants that the market is valued at around USD 0.5bn and set to double in the next four years. While it is true that data has become highly sought-after and hugely valuable, it is not all roses in the market for AI-related services, Bredenkamp cautioned. In his view, while more and more language service providers (LSPs) are exploring the market, the middleman is starting to get squeezed. The growth of the direct-to-client freelance market is putting pressure on intermediaries, as are the new machine-first alternatives that the AI community is exploring in an attempt to remove bottlenecks. According to Bredenkamp, although there are limits to what AI can achieve, it would be a mistake to impose human restrictions on its potential. Indeed, he expects and welcomes the possibility of AI bettering human-level intelligence: “Why should it stop there? We aren’t the limit,” Bredenkamp said. “I think this will be good for us. Let’s all get used to it. It will be a brave new world.” John Fennelly, CEO, Lionbridge Competing with Appen in the language data space, LSP Lionbridge is making an aggressive bet on the AI support services business. In an on-stage interview with Slator Co-founder Andrew Smart, Lionbridge CEO John Fennelly said that this line of business is nothing new for them and that they have been operating in the AI market for more than a decade. Of the company’s core localization business, Fennelly said that he sees the market shifting away from services toward a tech-enabled services model, which may leave a few smaller competitors flailing. Lionbridge returned to growth in 2018, Fennelly said, as the company’s focus on expanding verticals such as gaming and life sciences appears to have paid off. “You can’t do everything,” he pointed out, adding that Lionbridge is “focusing on key markets where we think there’s a lot of potential” and “committing more capital” to these priority verticals. Smith Yewell, CEO, Welocalize At SlatorCon London, Welocalize CEO Smith Yewell spoke on how to execute M&A in the language industry. Yewell, who founded Welocalize more than two decades ago, had 17 acquisitions under his belt as of May 2019 (and another one since). There are many potential pitfalls to avoid and, sadly, “half of M&A fails,” Yewell said. Though a lack of vision, culture clashes, and an unclear path to execution can lead to M&A fails, according to Yewell, “the biggest threat to M&A is when the talent leaves.” And this is all the more true for service companies. Yet the key lies in being able to demonstrate that staying on presents an opportunity: “The majority of our leadership team across the company, all came through acquisitions,” Yewell said, pointing to Welocalize’s track record for retaining and developing acquired talent. Phil Shawe, CEO, TransPerfect Growth was front and center at TransPerfect CEO Phil Shawe’s presentation at SlatorCon San Francisco. Shawe described the journey of building the company he co-founded more than 25 years ago from its humble beginnings into the language industry juggernaut it is today. Growth equals survival, Shawe said. And growth allows an LSP to give pay raises to employees without going out of business. Yet growing to a certain size brings with it a challenge of its own: “It’s very, very difficult, once you get larger, to maintain a high growth rate,” Shawe said. According to the TransPerfect CEO, the answer is in strategically thinking small rather than thinking big: “If we run 20 or 30 different small businesses and we run them all responsibly, and they all have a high growth rate, and they all have high profit margins, at the end of the day, we’re going to end up with a USD 700m business that’s going to have those same characteristics,” he said. To succeed in this, Shawe said it is important to measure the minutiae “like crazy,” hire entrepreneur-minded people, and be as transparent as possible across the business. Jack Welde, CEO, Smartling Concluding the day’s speaker lineup at SlatorCon San Francisco was Jack Welde, CEO of language services and technology provider Smartling. Welde gave his take on global content production, describing its importance and evolution. From the perspective of global companies, “content is how you reach prospects, reach customers,” Welde said. This need to engage customers wherever they are and whatever language they speak, means that companies are increasingly required to create and update large volumes multi-language content across multiple platforms, Welde explained. Yet customers still have budgets to manage, and often need help to discern which specific content to localize and how to go about it. Whether it is providing a retailer with details of trending items so they can decide which items to translate next, or helping a manufacturer reduce localization outsourcing costs by differentiating between content types, data is informative and powerful, Welde said. Yaron Kaufman, Co-founder and CMO, One Hour Translation As customers and LSPs alike explore the use of machine translation, many are unsure whether now is the time to invest, said Kaufman speaking at SlatorCon Amsterdam. For customers, the appeal of MT is either driven by a desire to save costs or by the need to process huge volumes of content. Customers want to know whether MT is “good enough.” Kaufman believes it is, and backed this assertion by pointing to One Hour Translation’s NMT report. OHT tracks the performance of popular MT engines, including Amazon Translate, Baidu, DeepL, Google Translate, and Yandex, by testing the output of 100 general phrases. The results from the engines are analyzed and given an NMT score, which is published in a publicly-available quarterly report. The performance of the MT engines tested has improved rapidly, even from quarter to quarter, Kaufman added, adding however that generic engines do not perform as well for domain-specific content such as gaming. Language pairs, of course, also play a role. Although still imperfect, Kaufman said NMT is indeed good enough and, on average, customers using machine translation (or MT plus post-editing) can expect to save around 30% of translation costs. Tomasz Tunguz, Partner and Managing Director, Redpoint VenturesSri Chandrasekar, Partner, Point72 VenturesCharles Stubbs, Partner, MSouth Equity Partners Led by Slator Co-founder Florian Faes, the SlatorCon San Francisco investor panel featured three language industry investors representing both venture capital and private equity. Tomasz Tunguz, whose VC firm is invested in tech-enabled startup-LSP Lilt, spoke of industry transformation. The Redpoint Ventures Partner said he sees an opportunity for AI-powered agencies to disrupt the industry. Though such agencies may not look very different from traditional LSPs on the surface, “under the hood they’re going to be much more efficient,” he said. The result? They will either choose to undercut the market or run the business “at a higher net income margin,” he added. Sri Chandrasekar, who led VC firm Point72’s investment in NLP startup PolyAI, agreed and elaborated on the idea of technology transforming the industry: “Perhaps just as importantly,” he said, “technology is enabling a whole new class of language translation and actually expanding the size of the market.” Two weeks post-SlatorCon, Chandrasekar also led the USD 60m Series C investment in Unbabel. Offering a private equity perspective, Charles Stubbs of MSouth Equity Partners outlined what he described as a slightly more simplistic strategy. Stubbs said that MSouth, which teamed up with language industry veteran Jeff Brink to form Big Language Solutions in 2019, saw three main reasons for investing in the language industry: fragmentation, customer demand, and revenue growth. According to Stubbs, language translation is “an industry that’s growing 7% year over year, double or triple GDP — that always interests us — and low capex investment. We traditionally love business services. Patrick Prokesch, Senior Director, I5Invest From an M&A advisory perspective, Prokesh walked SlatorCon Amsterdam participants through the various investment models common to the language industry. Of these, private equity and venture capital are the most dominant forms of investment and, according to Prokesch, were responsible for USD 0.3bn worth of language industry investment in 2019. The right choice of investor would depend on what a company is looking to do, since there are differing expectations from different types of investor, Prokesch said. Private equity, which focuses on mature companies and largely relies on buy-and-build strategies, aims to generate 2-3x return on investment; while venture capital investors have a comparatively high-risk appetite and are looking for higher returns in the region of 10x. Prokesch believes that the purpose of taking on a partner should be to gain access to opportunities that would otherwise be unavailable to a business; whether this means penetrating new markets or building faster. Securing investment is a lengthy process and one that relies on being able to tell a compelling story to potential investors as well as building relationships, Prokesch added. His perspective in a nutshell: “We believe that great companies are bought and not sold.” Amanda Smith, Director of Language Strategy and Content Business Operations, EMEA, Discovery NetworkJustin Walton, Global Content Operations Manager, ITVJan-Hendrik Hein, Director of Media Operations, A+E Networks UK Led by Slator Co-founder Andrew Smart, the SlatorCon London media panel featured three senior media localization buyers from Discovery Network, ITV, and A+E Networks UK. The panel discussion focused on challenges relating to managing increasing volumes of media entertainment content. Discovery localizes “over 100,000 hours of content per annum in EMEA alone,” Smith said, and to support these volumes, they use 55-plus in-territory vendors. Over at A+E Networks UK, on the other hand, Hein said that they localize into 20 languages: “All of the dubbing and subtitling is done by language service providers,” he said, “although we have an in-house team of Language Supervisors — all qualified linguists — who manage those relationships, specific guidelines, and run the RFPs.” ITV’s volumes are “more modest,” according to Walton. The team localized around “800 hours of dubbed content” in 2018, he said, all of which was outsourced. Most of ITV’s content is high-end drama (think Downton Abbey). Smith, Walton, and Hein also walked the SlatorCon participants through the complexities of current media localization realities; from linear scheduling, licensing and multilingual shows, to localizing archive content and, of course, emerging technology. Mario Tarantino, Content and Merchandising Manager, Huel Mario Tarantino of UK-based food startup Huel outlined the process of launching localization operations from scratch. Starting out with a blank slate in early 2018, Tarantino had rolled out localization into eight languages when he spoke at SlatorCon London in May 2019. Tarantino described the initial process of deciding what content to translate and into which languages based on factors including current and potential market opportunity. He detailed Huel’s localization operating model; from utilizing freelance resources and in-house employees through to using a single vendor (LSP) for localization and a third-party app for delivering translations. True to startup form, Huel monitors conversion rates before and after the launch of localized websites. Tarantino said that localization had driven an increase of more than 1% in conversion rates. While this may not sound like a huge boost, relative to typical e-commerce conversion rates of 1–3%, the figure takes on more meaning. Adelina Lear, Client Services Manager, ICON plc From her vantage point of working with a huge contract research organization or CRO, Adelina Lear spoke to SlatorCon London participants about the complex process of servicing the language needs of life sciences customers. ICON’s typical customers include the likes of pharma giants, who play a crucial role in researching, creating, and distributing drugs, with the aim of bringing breakthrough pharmaceuticals to market. Lear described the wide variety of document types generated as part of the clinical trials process, many of which require translation in order to comply with regulations governing the industry. She then discussed what could be one of the language industry’s most operationally complex areas: linguistic validation. The process, which involves a series of steps such as multiple forward and back-translations, as well as extensive quality assurance, leaves zero margin for error. Claire Tsai, Head of Globalization, Cloudflare Similarly to Huel’s Tarantino, Claire Tsai was her own one-person team when she started at cloud-based web security and network services company Cloudflare in 2017. From there, Tsai embarked on a journey of helping Cloudflare in “going global.” Cloudflare now localizes between 10–17 languages for consumer apps, Tsai said, and they plan to add more. In implementing her data-based strategy for expanding Cloudflare internationally, Tsai said she learned to steer clear of “vanity metrics,” which, for example, measure mere page views rather than actual viewer conversions. Instead, looking at the right metrics helps a company localize selectively, by prioritizing the items (such as check out) that have the most impact on the bottom line. Chip Huyen, Senior Deep Learning Engineer, NVIDIA Working at AI hardware production company NVIDIA, Chip Huyen bridges the gap between academia and enterprise. Huyen’s own experience includes stints at Netflix and teaching at Stanford. Explaining the differences between the state of neural machine translation (NMT) in academia and “the real world,” she outlined ways in which “real-world applications are more complex than academic datasets.” For one thing, academia provides a controlled setting for NMT. Not so in the real world of industry, where there are no limits to sentence length and no enforced definitions of content type, she said. According to Huyen, NMT research is currently exploring areas such as document length, unsupervised training, and quality estimation. The latter is more of a preoccupation for industry researchers, who have a vested interest in gauging the level of effort required to take NMT output to a deliverable standard. However, she said, quality estimation remains “underexplored” in academia. Meredith Stegall, Director of Language Services, Parkland Memorial Hospital With 882 beds and more than one million outpatient visits each year, Dallas, Texas-based Parkland is no small hospital. At SlatorCon San Francisco, Meredith Stegall described why and how she led an initiative to insource the vast majority of Parkland Memorial Hospital’s considerable interpreting needs. From spending some USD 9m of tax-payer money on interpreting in a 70%-outsourced model back in 2017, Parkland now employs more than 80 interpreters at its onsite call center and has delivered massive savings to boot, Stegall said. At current rates, interpreting through the Parkland call center costs USD 0.55 per minute compared to third-party vendors at USD 0.74 per minute. Totted up, the insourcing program has delivered some USD 1.5m in annual savings in addition to creating jobs locally. Vinicius Britto, Global Localization Manager, Bose CorporationAndrea Guisado Munoz, Senior Localization Architect, KAYAKAlvaro Villalvilla Merelo, Senior Global Localization Manager, NikeBalázs Benedek, Co-founder and CTO, Easyling Joining the three localization buyers on stage at SlatorCon Amsterdam was Balázs Benedek of Easyling as well as Slator’s Florian Faes, who moderated the enterprise and e-commerce panel. First tackling the complex activity of SEO localization, the panel also discussed their views on the single biggest value-add a language service provider can bring to a client’s organization, and much more. For Britto, global account management was top of mind: “What account management can you provide to make my job easier?” he asked the LSPs. For Nike’s Merelo, the ideal LSP is a “Swiss Army vendor,” focused on problem-solving. He also said that good account management skills are key to developing a partner-style relationship with customers — and the industry could do better in this area. At KAYAK, Muñoz’s team favors working with freelancers alongside the in-house team and rarely uses LSPs, “but I could see where they could help in situations when we need scalability very quickly,” she added. On the topic of MT, Muñoz said it is something the team considers whenever new translation requests come in. They assess factors including content type, scope, volume, and visibility, she said, to see whether MT makes sense. Easyling’s Benedek described MT as being used to remove the communication barrier in online fora, where users post around the same topic but in different languages. He also believes that MT is a good option for low-budget, high-volume scenarios, “when there are more words than dollars.” Michal Antzcak, Head of Localization Technology, Paypal For a group of people focused on language and communication, LSPs do not always do the best job of talking to customers in a way that really “speaks” to them, Antzcak told SlatorCon participants in Amsterdam. Many suppliers are quick to emphasize their competitive prices, quality commitments, and tech stacks, Antzcak said, yet they fail to demonstrate how localization can impact things that customers really care about: revenue metrics, KPIs and, ultimately, making their vision a reality. For Antczak, what makes more sense is for LSPs to instead focus on helping customers communicate with end users in a more personalized way. For example, LSPs could use what their customers know about the end users and tailor localization accordingly: Where do they live? How old are they? What kind of communication methods do they prefer? How do they feel about the company / product / service? If LSPs could then also discern how best to use technology such as MT and NLG to create highly targeted content, “that would be true innovation,” Antczak said. Jimena Almendares, Global Expansion VP, Intuit At SlatorCon Amsterdam, Jimena Almendares delivered a keynote speech tackling the question of how a big, public company like Inuit perceives and executes international expansion. Almendares described how Intuit, a US-based company, managed to bring their Quickbooks accounting product to the new and complex market of Mexico in just inside of three months. Fast forward, and the product has one of the highest revenues per customer. When it comes to expansion projects like this one, Almendares said, localization is just a drop in the ocean, and there are myriad factors at play; from HR and legal considerations to the regulatory environment in a new market. Conducting research up front and finding the right local partners to support you is crucial, she added. Her top success tips for other businesses looking to expand internationally? Understand your target customers and their particular pain points. Define your brand voice based on local preferences and likely impact. Secure cross-functional buy-in internally. And embrace the local culture for hiring and management practices. Hartmut von Berg, Head of Localization, LogMeIn With his eye firmly fixed on fueling international growth at listed SaaS provider LogMeIn, von Berg explained his team’s evolution from internal translation provider to globalization consultancy, a journey which is still ongoing. In von Berg’s mind, change was needed because “to fuel International growth, localization needs to be transformed away from providing translation services towards influencing business decisions about investments into international markets.” In an initial move, LogMeIn formed a dedicated localization team by bringing together the company’s localization employees, who were housed in two separate functions after a 2017 merger. Von Berg then dropped all his engineering responsibilities to head the newly created localization team. One crucial step was to centralize localization budget management, moving budget responsibility away from internal customers such as marketing, product, and customer care. While budget ownership still lies with these functions and “they still decide what to localize […], we manage it,” von Berg said. Looking forward, von Berg pinpointed the areas of SEO localization, MT, and terminology management as current priorities. LogMeIn is currently developing an SEO localization solution in-house since “we looked into the industry and we couldn’t identify a really good solution,” he said. As regards MT, the team is using it to improve scalability and serve unaided markets, he said. Meanwhile, strong terminology management, von Berg believes, is key to delivering “global customer experience with consistent terminology.” Florian Faes, Co-founder and Managing Director, Slator Opening proceedings at all three SlatorCons in 2019, Slator’s Florian Faes gave his trademark take on trends impacting the language industry. Faes referenced Slator’s own data and insights stemming from research conducted for the inaugural Slator 2019 Language Industry Market Report, published just hours before the SlatorCon London event in May 2019. In London, Faes described Slator’s approach to sizing and segmenting the market; which, by its very nature, operates cross-border. From one angle, Faes described how Slator defined and sized the 10 primary industry verticals that make up the USD 23.2bn language industry, based on end-customer segments. Examined from a supplier perspective, Faes said, Slator identifies four distinct groups of players that service the addressable market for language services: Super Agencies, Leaders, Challengers, and the Longtail, a large chunk which comprises freelancers, boutique agencies, as well as internal corporate and public sector translation functions. With 10% market share, the five Super Agencies lead the global full-service LSP contingent from a revenue perspective: TransPerfect, Lionbridge, RWS, SDL, and Welocalize. As customer spend drops, accounts become less readily serviceable by large LSPs, Faes explained, largely because of the escalating cost of sale. Among the likely winners in an evolving language industry, he singled out the Super Agencies, localization technology platforms, and tech-centric service providers as being well positioned to succeed. With a spotlight on media localization at SlatorCon London, Faes identified strong growth in this language industry niche, also found in technology and gaming. These three markets, followed by life sciences, will be the most attractive verticals for translation and localization providers over the next three years, he said. Skip forward four months to the other side of the Atlantic, and Faes’s message at SlatorCon SanFrancisco reaffirmed many of his assertions from London, as he spoke of the language industry’s main markets, players, technology developments, and growth opportunities. Expanding on the role and positioning of Super Agencies, Faes also outlined the current M&A environment, touching on newly-completed sizable deals, such as Big Language Solutions’ acquisition of ProTranslating and the BTI Studios and IYUNO merger. According to Faes, what is currently dominating the focus in much of the industry is machine translation. MT, he said, is becoming more integrated into the supply chain as the entire industry retools in response to technology becoming more available, and as clients grow more interested in leveraging MT to achieve productivity gains and cost efficiencies. Developing this idea at the third leg of the SlatorCon tour, in Amsterdam, Faes told conference attendees that there is still room for the market for machine translation to grow and mature. The industry is trying to figure out how MT can complement and work together with other translation tools that are well-embedded in company tech stacks, like TMS and productivity (CAT) tools. The opportunity for providers lies in developing a product that makes it easier for translators and linguists to interact with machine translation, Faes said. Revisiting growth hotspots in the industry, Faes highlighted pockets of disruption that are influencing the media and game localization verticals, as subscription-based streaming platforms are shaping the competitive landscape. He also gave an update on the state of funding in the language industry, pinpointing a shift in the nature of investments. Five years ago, Faes said, investors were pitched on the idea of uberizing the marketplace, whereas now it is very much AI-based. We are pleased to announce the dates and locations for SlatorCon 2020.
SlatorCon 巡回赛于2019年在两大洲的三个全球城市举行,共有28名演讲者代表了语言行业的方方面面——从初创企业、语言技术和语言服务提供商,到买方和金融。 在这里,我们回顾了2019年活动日历中的关键内容。 SlatorCon London –2019SlatorCon 旧金山–2019SlatorCon Amsterdam –2019年11月 Jean Senillart , Systra 首席执行官 当希斯勒的让·塞内拉在伦敦的 SlatorCon 演出时,这不是第一次。这位机器翻译( MT )老手在2018年伦敦 SlatorCon 首次亮相,参加了 MT 历史的告密之旅;从基于规则的 MT 到当前状态的神经 MT ( NMT ),等等。 2019年5月,塞内拉提请人们注意谷歌的自关注变压器 NMT 模型。他说,自2017年首次推出以来,它已成为研究和部署领域的领先模式,两年内没有任何技术能够取代它。 Senellar 还强调了机器翻译研究人员当前的热门话题,包括 MT 和 TM 之间的交互、后编辑和低资源语言。塞内拉说,无监督学习领域有可能对 MT 输出质量产生重大影响。 Systrar 首席执行官向 MT 社区提供了一句建议,提醒 SlatorCon 的受众,客户的 MT 触点是产品,而不是技术:“我们谈论的是技术,就像用户想要的一样,”他说。“他们不想要技术,他们想要产品。” Mark Brayan , Appen 首席执行官 从 MT 中删除的一层是语言数据的世界,语言行业的一个蓬勃发展的细分市场,也被称为“ AI 支持服务”。语言数据用于培训人工智能和机器应用程序,如 MT 引擎和更多。语言数据空间的核心是澳大利亚上市公司 Appen ,它利用其庞大的远程工作人员群体来收集、处理和标注大量数据。“数据越多越好,” Appen 首席执行官马克•布兰扬( Mark Brayan )表示。他补充说,人们需要记住,“数据昂贵、收集起来复杂,而且工作起来复杂。” 尽管 Appen 的操作依赖于大量的人类注解来创建和检查最终将用于语音和图像识别工具的数据,但它不仅仅是一个数字游戏。数据的广度、多样性(以及最终的质量)也很重要。Brayan 说,要从语音数据中获得最好的结果,例如口音、年龄、声音设置等因素必须适合客户的具体使用情况。 AndrewBredenkamp , Acrolinx 创始人兼首席执行官 Acrolinx 的 Bredenkamp 也精通人工智能快速发展的数据标签市场。他告诉 SlatorCon 阿姆斯特丹的参与者,该市场的价值约为5亿美元,未来四年将翻番。Bredenkamp 警告说,虽然数据确实变得非常受欢迎,而且非常有价值,但并非所有人工智能相关服务市场的玫瑰。 他认为,虽然越来越多的语言服务提供商( LSP )正在开拓市场,但中间商开始受到挤压。直接面向客户的自由职业市场的增长给中介机构带来了压力,人工智能领域正在探索新的机器优先替代品,以消除瓶颈。 根据 Bredenkamp 的说法,虽然人工智能所能达到的目标是有限度的,但把人类对其潜能的限制强加给他人是错误的。的确,他期望并欢迎人工智能改进人类智能的可能性:“为什么它应该停止在那里?”我们不是极限,” Bredenkamp 说。“我认为这对我们有利。我们都习惯了。这将是一个勇敢的新世界。” John Fennell , Lionbridge 首席执行官 在语言数据领域与 Appen 竞争, LSP Lionbridge 正积极押注于人工智能支持服务业务。在对 Slator 联合创始人 Andrew Smart 的现场采访中, Lionbridge 首席执行官 John Fennelly 表示,这一业务对他们来说并不新鲜,他们已经在人工智能市场运营了十多年。 芬内利表示,在公司的核心本地化业务中,他认为市场将从服务转向技术支持的服务模式,这可能会让一些规模较小的竞争对手破产。 芬内利表示, Lionbridge 在2018年恢复了增长,因为该公司专注于拓展游戏和生命科学等垂直领域的业务似乎已获得回报。他指出:“你不能做任何事情。”他补充说, Lionbridge “专注于关键市场,我们认为这些市场潜力巨大”,并“投入更多资金”这些优先垂直市场。 Smith Yewell , Welcalize 首席执行官 在伦敦 SlatorCon , Welcalize 首席执行官 Smith Yewell 谈到了如何在语言行业实施并购。Yewell 在20多年前创立了 Welcalize ,截至2019年5月,他已经进行了17笔收购(此后又进行了一笔收购)。有许多潜在的陷阱要避免,可悲的是,“一半的并购失败了,” Yewell 说。 Yewell 表示,尽管缺乏远见、文化冲突和执行途径不明确可能导致并购失败,“但对并购的最大威胁是人才离开。”对服务公司来说,这更真实。然而,关键在于能够证明继续留任是一个机遇:“我们整个公司的大部分领导团队都是通过收购实现的,” Yewell 说,他指的是 Welcalize 在留住和培养人才方面的过往记录。 Phil Shawe , TransPerfect 首席执行官 成长是 TransPerfect 公司首席执行官 PhilShawe 在旧金山 SlatorCon 的演讲的中心和前沿。 Shawe 描述了25年前他共同创立的公司从一个不太起眼的语言行业巨头发展到今天的历程。 Shawe 说,增长等于生存。而增长使得 LSP 能够在不退出业务的情况下提高员工的工资。然而,增长到一定规模带来了自身的挑战:“一旦规模扩大,保持高增长率非常困难,” Shawe 说。 TransPerfect 首席执行官表示,答案是从战略角度考虑小企业,而不是考虑大企业:“如果我们经营20到30家不同的小企业,我们都负责任地经营它们,它们的增长率都很高,最终它们都有很高的利润率。我们最终将拥有7亿美元的业务,具有同样的特点,”他说。为了成功做到这一点, Shawe 表示,重要的是要衡量细节“像疯了一样”,雇用有企业家头脑的人,并在整个企业中尽可能透明。 JackWelde , Smartling 首席执行官 最后,语言服务和技术提供商 Smartling 的首席执行官杰克•韦尔德( Jack Welde )在旧金山 SlatorCon 结束了当天的演讲阵容。Welde 介绍了他对全球内容制作的态度,描述了它的重要性和演变。 Welde 说,从全球公司的角度来看,“内容是你如何接触潜在客户,接触客户。”Welde 解释说,无论客户身处何地,无论他们说什么语言,这都意味着企业越来越需要在多个平台上创建和更新大量多语言内容。 然而,客户仍然需要管理预算,并且经常需要帮助确定本地化的具体内容以及如何进行本地化。Welde 说,无论是为零售商提供趋势项目的细节,以便他们能够决定下一步要翻译哪些项目,还是通过区分内容类型帮助制造商降低本地化外包成本,数据都是信息丰富且功能强大的。 Yaron Kaufman ,联合创始人兼首席营销官,一小时翻译 考夫曼在阿姆斯特丹的 SlatorCon 说,由于客户和 LSP 都在探索机器翻译的使用,许多人不确定现在是否是投资的时机。对于客户而言, MT 的吸引力要么是出于节省成本的愿望,要么是因为需要处理大量内容。 客户想知道 MT 是否“足够好”。考夫曼认为这是事实,并支持这一说法,指出一小时翻译的 NMT 报告。OHT 通过测试100个通用短语的输出来跟踪流行的 MT 引擎的性能,包括 AmazonTranslat 、 Baidu 、 DeepL 、 GoogleTranslate 和 Yandex 。对发动机的结果进行分析并给出 NMT 评分,该评分在公开发布的季度报告中公布。 考夫曼补充说,测试的 MT 引擎的性能已经快速提高,即使是从季度到季度,但通用引擎的性能并不像游戏等特定领域的内容那样好。 当然,语言对也起着作用.考夫曼说,虽然 NMT 还不完善,但它确实足够好,平均而言,使用机器翻译(或 MT 加编辑)的客户可以节省大约30%的翻译成本。 红点投资合伙人兼董事总经理 Tomasz Tunguz Sri Chandrasekar , Point72 Ventures 合伙人 Charles Stubbs , MSouth Equity Partners 合伙人 在 Slator 联合创始人 Florian Faes 的带领下, Slator Con 旧金山投资者委员会的三位语言行业投资者代表了风险投资和私人股本。 TomaszTunguz 是一家风险投资公司,投资于科技创业公司 LSP Lilt ,他谈到了行业转型。RedpointVenturesPartner 表示,他认为 AI 支持的机构有机会颠覆这个行业。尽管这些机构表面上看起来可能与传统的 LSP 并无太大不同,“但它们的效率将大大提高,”他说。结果是什么?他补充称,他们要么选择削弱市场,要么“以更高的净收入利润率”经营业务。 风险投资公司 Point72对 NLP 初创公司 PolyAI 的投资负责人斯里·钱德拉塞卡( Sri Chandrasekar )同意并阐述了技术改变行业的想法:“或许同样重要的是,技术正在促成一种全新的语言翻译类别,并实际上扩大了市场规模。”两周后的 SlatorCon , Chandrasekar 也领导了6000万美元的 C 系列投资 Unbabel 。 MSouth Equity Partners 的查尔斯?Stubbs 表示, MSouth 与语言行业资深人士 JeffBrink 合作,于2019年成立了 BigLanguage Solutions 。该公司认为,在语言行业投资的三个主要原因是:碎片化、客户需求和收入增长。根据 Stubbs 的数据,语言翻译是“一个每年增长7%的行业,两倍或三倍的 GDP ——总是让我们感兴趣——和低资本支出。我们传统上喜欢商务服务。 Patrick Prokesch , I5Invest 高级总监 从并购咨询的角度来看, Prokesh 通过语言行业常见的各种投资模式,带领 SlatorCon 阿姆斯特丹的参与者。其中,私人股本和风险资本是最主要的投资形式,根据 Prokesch 的数据,它们在2019年对语言行业投资价值3亿美元。 Prokesch 说,投资者的正确选择将取决于公司希望做什么,因为不同类型投资者的预期不同。专注于成熟企业、主要依赖于收购和建设战略的私人股本,旨在产生2-3倍的投资回报;而风险资本投资者的风险偏好相对较高,正寻求在10倍左右的地区获得更高回报。 Prokesch 认为,与合作伙伴合作的目的应该是获得企业无法获得的机会;这是否意味着打入新市场或加快建设速度。Prokesch 补充说,确保投资是一个漫长的过程,依赖于能够向潜在投资者讲述一个令人信服的故事以及建立关系。简单地说,他的观点是:“我们认为,大公司是被收购而不是被出售的。” Amanda Smith , EMEA 语言战略和内容业务运营总监, Discovery NetworkJustin Walton , ITVJan-Hendrik Hein ,英国 A + E Networks 媒体运营总监 在 Slator 联合创始人 Andrew Smart 的带领下, SlatorCon London 媒体小组展示了来自 Discovery Network 、 ITV 和 A + E Networks 英国的三家高级媒体本地化买家。小组讨论的重点是与管理日益增多的媒体娱乐内容有关的挑战。 史密斯表示, Discovery 将“仅在 EMEA 每年的内容就超过10万小时”,为了支持这些内容,他们使用了55多家境内供应商。另一方面,在英国 A + E 网络公司( A + E Networks UK ),海因表示,他们将本地化为20种语言:“所有配音和字幕都是由语言服务提供商完成的,”他说,“尽管我们有一个由语言主管组成的内部团队——所有合格的语言学家——负责管理这些关系、特定指导原则,并运行 RFP 。”沃尔顿表示, ITV 的销量“较为温和”。他表示,该团队在2018年将“800小时的配音内容”本地化,所有这些都是外包的。ITV 的大部分内容都是高端戏剧(如唐顿庄园)。 Smith 、 Walton 和 Hein 还向 SlatorCon 参与者介绍了当前媒体本地化现实的复杂性;从线性调度、许可和多语言展示,到存档内容本地化,当然还有新兴技术。 Mario Tarantino , Huel 内容和营销经理 英国食品初创公司 Huel 的马里奥•塔兰蒂诺( Mario Tarantino )概述了从零开始启动本地化业务的过程。从2018年初的一张白纸开始,塔伦蒂诺于2019年5月在伦敦 SlatorCon 演讲时,推出了8种语言的本地化。 塔伦蒂诺描述了根据当前和潜在市场机会等因素决定翻译内容和翻译语言的初始过程。他详细介绍了 Huel 的本地化运营模式,从利用自由职业资源和内部员工,到使用单一供应商( LSP )进行本地化,以及第三方应用程序进行翻译。 对于启动形式, Huel 监控本地化网站启动前后的转化率。塔伦蒂诺说,本地化推动了1%以上的转化率的提高。虽然这听起来不像是一个巨大的推动,相对于典型的1-3%的电子商务转化率,这个数字更有意义。 Adelina Lear , ICON plc 客户服务经理 Adelina Lear 从与一家大型合同研究组织或 CRO 合作的角度,向 SlatorCon London 的参与者讲述了为生命科学客户的语言需求服务的复杂过程。ICON 的典型客户包括制药巨头等,它们在药品研发、生产和销售方面发挥着关键作用,目的是将突破性的药品推向市场。 Lear 描述了作为临床试验过程的一部分而产生的各种文档类型,其中许多需要翻译,以符合规范行业的法规。然后,她讨论了语言行业最复杂的操作领域之一:语言验证。这一过程包括一系列步骤,如多个前向和后向翻译,以及广泛的质量保证,为错误留有零余地。 Claire Tsai , Cloudflare 全球化总监 与 Huel 的塔兰蒂诺( Tarantino )类似, Claire Tsai 在2017年开始在基于云的网络安全和网络服务公司 Cloudflare 工作时,也是她自己的一人团队。从那以后,蔡英文开始了帮助 Cloudflare “走向全球”的旅程。蔡崇信说, Cloudflare 现在将消费者应用程序的语言定位在10到17种之间,他们计划增加更多。 蔡英文在实施她的基于数据的全球云端扩展战略时表示,她学会避开“虚荣度指标”,例如,衡量的仅仅是页面视图,而不是实际的浏览者转换。相反,查看正确的指标有助于公司有选择地进行本地化,方法是对对底线影响最大的项目(如签出)进行优先排序。 Chip Huyen , NVIDIA 高级深度学习工程师 在 AI 硬件生产公司 NVIDIA 工作, Chip Huyen 弥补了学术界和企业之间的差距。Huyen 自己的经历包括在 Netflix 任职和在斯坦福( Stanford )任教。在解释学术界神经机器翻译( NMT )与“现实世界”之间的差异时,她概述了“现实世界的应用比学术数据更复杂”的方式。首先,学术界为 NMT 提供了一个可控的设置。她说,在真实的工业世界里,没有对句子长度的限制,也没有对内容类型的强制定义。 Huyen 表示, NMT 研究目前正在探索文档长度、无监督培训和质量评估等领域。后者更多地是行业研究人员的关注,他们对衡量将 NMT 输出提升到可交付标准所需的努力水平有着既得利益。然而,她表示,学术界对质量的评估仍“未充分展开”。 Meredith Stegal ,帕克兰纪念医院语言服务主任 每年有882张病床和100多万人次的门诊就诊,德克萨斯州达拉斯市的帕克兰医院不是小型医院。在旧金山的 SlatorCon , MeredithStegal 描述了她为什么以及如何领导一项计划,以满足绝大多数 Parkland 纪念医院的大量口译需求。 Stegal 说,早在2017年, Parkland 就花费了大约900万美元的纳税人资金,在70%的外包模式下进行口译服务。如今, Parkland 在其现场呼叫中心雇用了80多名口译员,并为启动提供了大量节省。按照目前的费率, Parkland 呼叫中心的口译费用为每分钟0.55美元,而第三方供应商为每分钟0.74美元。总体而言,内包项目除了在当地创造就业机会外,每年还节省了150万美元。 Vinicius Britto , Bose Corporation Andrea Guisado Munoz 全球本地化经理, KAYAKAlvaro Villalvilla Merelo 高级本地化架构师, NikeBalz Benedek 高级全球本地化经理, Earling 联合创始人兼首席技术官 参加在阿姆斯特丹 SlatorCon 舞台上的三个本地化买家是 Earling 的 Balz Benedek 和 Slator 的 Florian Faes ,他们主持了企业和电子商务小组。 首先,针对 SEO 本地化的复杂活动,专家小组还讨论了他们对语言服务提供商能够为客户组织带来的最大增值的看法,以及更多的观点。 对英国人来说,全球客户管理是最重要的思想:“你能提供什么客户管理来让我的工作更轻松?“他问伦敦警察局。对于耐克( Nike )的梅洛( Merelo )来说,理想的 LSP 是一家“瑞士陆军供应商”,专注于解决问题。他还表示,良好的客户管理技能是与客户建立合作伙伴关系的关键,行业可能在这方面做得更好。在 KAYAK , Mu ñ oz 的团队倾向于与自由职业者一起工作,与内部团队一起工作,很少使用 LSP ,“但我可以看到,在我们需要快速可扩展性的情况下,他们可以帮助什么,”她补充说。 关于 MT 的话题, Mu ñ oz 说,每当新的翻译请求出现时,团队都会考虑这个问题。他们评估了一些因素,包括内容类型、范围、容量和可见性,她说,看看 MT 是否有意义。Earling 的 Benedek 将 MT 描述为用于消除在线论坛中的通信障碍,用户在论坛上发布的主题相同,但语言不同。他还认为, MT 是低预算、高容量场景的一个很好的选择,“当词多于美元时”。 Paypal 本地化技术总监 Michal Antzcak Antzcak 告诉阿姆斯特丹的 SlatorCon 参与者说,对于一群专注于语言和沟通的人来说, LSP 并不总是以一种真正“与他们交谈”的方式来最好地与客户沟通。 Antzcak 表示,许多供应商很快就会强调自己的竞争价格、质量承诺和技术优势,但他们未能证明本地化如何影响客户真正关心的事情:收入指标、 KPI ,以及最终实现他们的愿景。 对于 Antczak 来说,更有意义的是 LSP 将重点放在帮助客户以更个性化的方式与终端用户沟通上。例如, LSP 可以使用他们的客户对最终用户的了解,并相应地定制本地化:他们住在哪里?他们有多老?他们喜欢哪种沟通方式?他们对公司/产品/服务有何看法?Antczak 说,如果 LSP 也能发现如何最好地利用 MT 和 NLG 等技术来创建高针对性的内容,“那将是真正的创新。” JimenaAlmandaares , Intuit 全球扩展副总裁 在阿姆斯特丹的 SlatorCon , JimenaAlmendaares 发表了主旨演讲,讨论了像 Inuit 这样的大型上市公司如何感知并实施国际扩张的问题。Almandaares 描述了总部位于美国的 Intuit 公司如何在短短三个月内将其 Quickbook 会计产品带到了墨西哥这个新的复杂市场。快速前进,并且该产品具有最高的收入每个客户。 谈到像这样的扩张项目, Almandaares 说,本地化只是海洋中的一种下降,而且还有很多因素在起作用:从人力资源和法律方面的考虑到新市场的监管环境。她补充说,提前进行调查并找到合适的当地合作伙伴来支持你是至关重要的。 她的成功秘诀为其他寻求国际扩张的企业?了解目标客户及其特定的痛点。根据当地偏好和可能的影响来定义您的品牌声音。确保内部跨职能收购。并接受当地雇用和管理实践的文化。 LogMein 本地化主管 Hartmut von Berg 冯•伯格( von Berg )坚定地致力于推动 SaaS 上市公司 LogMeIn 的国际增长,他解释了自己的团队从内部翻译提供商向全球化咨询公司的演变,这一过程仍在进行中。 冯•伯格( von Berg )认为,变革是必要的,因为“为了促进国际增长,本地化需要从提供翻译服务转变为影响有关投资进入国际市场的商业决策。” LogMeIn 最初通过将公司的本地化员工聚集在一起,组建了一个专门的本地化团队,他们在2017年的合并后被安置在两个独立的职能部门。然后 VonBerg 放弃了他的所有工程职责,领导新创建的本地化团队。 一个关键步骤是集中本地化预算管理,将预算责任从内部客户(如市场营销、产品和客户关怀)转移。虽然预算所有权仍然掌握在这些职能上,“他们仍然决定将哪些工作本地化,但我们对此进行管理,”冯•伯格( von Berg )表示。 展望未来, vonBerg 指出 SEO 本地化、 MT 和术语管理领域是当前的重点。LogMeIn 目前正在内部开发一个 SEO 本地化解决方案,因为“我们研究了这个行业,我们无法确定一个真正好的解决方案,”他说。他说,在 MT 方面,该团队正在使用它来提高可伸缩性,并为未开发的市场提供服务。同时, von Berg 认为,强大的术语管理是“以一致的术语提供全球客户体验”的关键。 Florian Faes , Slator 联合创始人兼董事总经理 Slator 的 Florian Faes 在2019年的三场 Slator Cons 大会上首次亮相,标志着他对影响语言行业的趋势的认同。Faes 引用了 Slator 自己的数据,以及在2019年5月 SlatorCon 伦敦活动前几个小时发布的首份《 Slator 2019语言行业市场报告》( Slator 2019 Language Industry Market Report )研究得出的结论。 在伦敦, Faes 描述了 Slator 对市场规模和细分的方法;从本质上讲,这是一种跨境经营。Faes 从一个角度描述了 Slator 如何根据终端客户细分市场定义和调整10个主要行业垂直行业,这些行业构成了232亿美元的语言行业。 Faes 说,从供应商的角度来看, Slator 确定了为语言服务的目标市场服务的四个不同的参与者群体:超级机构、领导者、挑战者和龙尾,其中很大一部分包括自由职业者、精品机构以及内部公司和公共部门翻译职能。 凭借10%的市场份额,五家超级机构从收入角度领导了全球全面服务 LSP 队伍: TransPerfect 、 Lionbridge 、 RWS 、 SDL 和 Welcalize 。Faes 解释说,随着客户支出下降,大型 LSP 的客户服务变得不那么容易,这主要是因为销售成本不断上升。在一个不断发展的语言行业可能的赢家中,他特别指出了 SuperAgencies 、本地化技术平台和以技术为中心的服务提供商,以及成功的有利条件。 Faes 在 SlatorCon London 关注媒体本地化,发现该语言行业在技术和游戏领域也有强劲增长。他说,这三个市场,其次是生命科学,将是未来三年翻译和本地化供应商最具吸引力的垂直市场。 在谈到语言行业的主要市场、参与者、技术发展和增长机会时,费斯在 SlatorCon SanFrancisco 的讲话重申了他在伦敦的许多主张。 Faes 还扩大了超级机构的作用和定位,概述了当前的并购环境,涉及新完成的大量交易,如 Big Language Solutions 收购 ProTranslating 和 BTI Studios 以及 IYUNO 合并。 根据 Faes 的说法,目前该行业的主要焦点是机器翻译。他说,随着整个行业因技术变得更加可用而进行重组,以及客户越来越有兴趣利用 MT 来提高生产率和成本效率, MT 正越来越融入供应链。 费斯在阿姆斯特丹的 SlatorCon 巡回赛的第三站开发这一理念,他告诉与会者,机器翻译市场仍有成长和成熟的空间。 该行业正试图弄清楚 MT 如何与其他嵌入在公司技术堆栈中的翻译工具(如 TMS 和 CAT )进行互补和协作。Faes 说,供应商的机会在于开发一种产品,使译者和语言学家更容易与机器翻译交互。 Faes 重新审视了该行业的增长热点,强调了影响媒体和游戏本地化垂直市场的部分中断,因为基于订阅的流媒体平台正在塑造竞争格局。 他还介绍了语言行业融资状况的最新情况,指出了投资性质的转变。Faes 说,5年前,投资者被灌输市场优势化的理念,而现在,这在很大程度上是基于人工智能的。 我们很高兴地宣布 SlatorCon 2020的日期和地点。

以上中文文本为机器翻译,存在不同程度偏差和错误,请理解并参考英文原文阅读。

阅读原文