Today, with even the smallest enterprise potentially serving a global client base, the need to communicate across languages and cultures is growing rapidly. However, cross-context communication is hard and costly. Unless great care is taken, many things can be lost in translation due to translation errors and/or differing interpretations of even correctly translated communications.
The costs of translation failures are often more than just financial. Miscommunication can lead to loss of reputation, legal exposure, physical harm, or even industrial disasters. For this reason, clear, accurate and effective communication – between cultures, languages, disciplines, and industries – is an increasing priority.
In response, many companies spend significant resources to ensure communication within their networks of agents, partners, customers, and government agencies. This need to accurately share information between and among diverse trading partners has evolved into the business function called localization.
Localization is defined as technologies and processes that adapt products and services for use in and by specific countries, regions, or groups. It is often a highly complex and expensive process that includes translating text and audio material, modifying documents and software to reflect localized conventions (such as how decimal points and dates are represented), and continually analyzing and incorporating into products and services the regulatory, compliance, and tax requirements of specific countries, states, and cities.
This article focuses on one specific function within the broader localization picture: translation. Across businesses worldwide, there is an increasing demand for translation, driven by:
- an increase in demand for non-English languages,
- increase in products and services from non-English countries reaching foreign markets,
- an in vertical-specific translation use cases, and
- a reduction of translation, driven by improvements in AI technology and the rise of cloud-based translation platforms. These lower costs support a growing “long tail” of businesses that can profitably offer services in multiple languages.
This article discusses this new future, in which there is an increased demand and opportunity for translation. We also cover nuances that your organization needs to keep in mind to succeed in this changing market.
Translation services at a turning point
Annual enterprise spending on translation services is expected to grow to US$45 billion by 2020, primarily driven by increasing globalization and an increasing amount of text being generated worldwide.
This growth is also being stimulated by new technology. Many organizations are using artificial intelligence (AI) in the form of machine translation (MT) to reduce the costs of translation. AI-enabled automated translation platforms like Google Translate, Microsoft Translator, and the recently released Amazon Translate have in the last 24 months taken a great leap forward in accuracy. This is for two reasons: one, they build on recent breakthrough improvements in neural machine translation (NMT) algorithms, and, two, they have access to a much larger amount of language data from search engines, social networks, and e-commerce sites.
For less-demanding consumer (B2c) use cases, such as translating a web site for a casual browser, the accuracy of these fully-automated AI-based systems has recently become “good enough” for a large number of use cases. Typically these translations are offered for free and supported by ads, so the users are happy with whatever quality they can get, and the consequences of errors are low.
In contrast, the accuracy of these existing systems is not adequate for many business use cases, such as creating a user interface in a new language, translating a tax document, or creating a user manual for a product in a new language. Yet AI is also having a big impact here, where human-in-the-loop uses cases allow the AI system to do an initial translation that is then refined by a human expert. Although this isn’t driving translation pricing all the way to zero, this technology is, nonetheless, having a profound impact on the translation marketplace, which is transforming in shape as a result of these forces.
AI and the end of translation jobs?
The recent acceleration in machine translation sophistication and reliability leads some observers to speculate that machines will essentially remove the need for expensive human translation even in the enterprise market, eliminating tens of thousands of jobs in product and service localization, publishing, marketing, and myriad other fields, even as the demand for translation explodes.
However, this is a false extrapolation of current success. Although the hype around recent improvements is largely justified, the idea that machines will destroy language services as an industry and drastically reduce the need for translation and globalization teams is not. There are a number of reasons:
- As described above, the bar for successful language translation in the enterprise is substantially higher than for consumer applications.
- Even within the enterprise, the bar is rising for these reasons:
- More languages and dialects must be handled.
- More specialty vertical markets, such as the law and healthcare, must be served.
- There is an increase in the need for more specialty horizontal document types, such as for documents describing decisions, requirements, and systems.
- Translating the functional aspects of a product (e.g. menus and documentation) is a specialized practice and one for which consumer approaches to translation do not readily apply.
- Current language translation technologies will not improve at the current pace unabated. The biggest recent advances have come from leveraging massive corpora of already-translated materials to learn translation models that can translate similar content in the future. Many enterprise cases are much more specific in terms of context and discipline, and also have lower volumes of already-translated data for these narrower contexts. These are technical challenges that AI algorithms are only today beginning to address, and new technology transfer ―if not also new R&D―are required to reach the next level in driving business value.
- The number of languages that can be profitably translated is increasing with the new lower-cost, AI-supported approach, as we describe in more detail below. Hence, even as the costs for translating higher-priority languages might come down, the volume of emerging-priority languages continues to rise. These less-translated languages have less training data, making the automation problem harder, as noted above.
Translating the long tail
The number of languages and language pairs now handled by the most advanced translation platforms represents only a small fraction of the languages spoken across even the developed world. But translating content into languages beyond the 40 or so supported by the largest language service providers (LSP) and by enterprise software vendors has, to date, been difficult or impossible to cost-justify: for most companies, the cost and time required to add just one new language to a product has been measured in the millions of dollars and years of time.
Those barriers are about to be shattered by the combination of scale efficiencies enabled by cloud-based platforms and translator productivity improvements enabled by machine translation.
While some of these long-tail language markets are growing quickly, few will ever represent enough revenue to justify the cost under the existing on-premises deployment, non-AI technology model. However, with platform economics and AI-enabled efficiencies dropping the cost, translation providers will be able to recoup the financial and time investment required to add new languages while still keeping translation prices affordable for a much larger set of customers. The economics will also allow providers to vastly increase the volume of translation they can handle, helping to maintain revenue and margins even as prices drop.
The mix of languages that need to be translated is shifting. Today, while English is the top language used on the internet, less than a third of an estimated 4 billion Internet users are English speakers.
Going forward, the number of new language opportunities is substantial and represents a new market for many businesses. According to Common Sense Advisory (CSA), enterprises will need to translate content into a steadily increasing number of “niche” languages in order to reach small but fast-growing economies. Where approximately 14 languages are sufficient to reach about 75% of global Internet users today, reaching the next 20% requires adding about 40 more. By 2027, the firm estimates that enterprises will need to translate into more than 60 languages in order to reach 96% of the online population.
More data on this point: while English remains the primary language of international business and the Internet, a commonly cited forecast that appeared as early as 2005 projected that the next billion Internet users would not be native English speakers. Visual Capitalist traces the entire status of language usage today. World language authority Ethnologue estimates that English is a second language for about 60% of all English speakers. If you have ever learned a foreign language, you will appreciate how much easier it is to understand your native language. That translates into more effective communication and higher business value for whoever does the translation.
While no one expects every linguistic group will be covered any time soon, new patterns of trade will soon drive requirements for some unexpected languages and language pairs. China’s Belt and Road initiative to build a new Silk Road – roads, rail, bridges, sailing routes, pipelines, and trade alliances – is expected to connect 70 countries in Asia. Defaulting to English as a common language in place of local languages will become less viable as these markets are opened up.
Also consider that there are more than 20 major languages in India, written in a dozen different scripts, and estimates of over 720 dialects. Twenty-six regional languages in India are spoken by over 1 million people who don’t speak Hindi. And although it may not be obvious, some of these regional languages are more valuable than Hindi in terms of the speakers’ literacy and economic status.
Beyond adding entirely new languages, the move to a platform model will also ultimately make translating withinlanguages – between dialects and specialist jargons – affordable, further extending the long-tail opportunity.
Altogether, this combination of today’s decreasing hegemony of English along with non-Western centric developments in global business means that there is an increasing demand for translation between many new language pairs, and this demand will last for years to come.
High-value and high-volume domain-specific translation
While these long-tail enterprise language services opportunities develop, the ability to translate and interpret domain-specific vocabularies that convey highly precise meaning is at least as important as the number of languages translated. For an example, a document may describe a medical patient, an important strategic decision, or the rationale for a new government or management policy. In a medical context, for instance, the word “protocol” has a very specific meaning referring to a set of standard steps used to treat a condition. For example, a particular chemotherapy drug may be associated with a protocol that says it should be administered at a certain dosage every two weeks. In contrast, “protocol” in a telecommunications context has an entirely different meaning referring to how data must be changed as it is exchanged between two telecommunications companies like Verizon and Orange.
In medicine, as more medical documents are digitalized, there will be an increasing global demand for fast but accurate translation for doctor-patient communication. The disparity between the availability of specialists in developed countries and in the developing world (including in refugee camps) represents one important gap for which automated translation systems can be particularly helpful.
In domain-specific contexts like these, AI systems require domain-specific language examples for training. However there is a power-versus-generality tradeoff here: It is a time-consuming process to gather medical, insurance, and other specific language information in 100 languages at once. For this reason, new AI technologies like transfer learning are promising because they allow one system to bootstrap another that is targeted for a related situation.
From vertical to horizontal domains
It is less widely recognized that every industry and many industry sectors use language in ways that, while not as specialized as in highly technical fields, differ in important ways from common usage.
Companies both within and between industries and/or segments may share those semi-specialized vocabularies. For instance, many industries share the same project management vocabulary, with well-understood meanings for words like “resource” (which usually means a person working on a task) and “dependency” (which specifies which tasks must be done before others) independent of whether the company offers insurance or healthcare.
Enterprise software vendors, especially those furthest along in the transition to delivering solutions via cloud-based platforms rather than via on-premises implementations, have access to the best and most extensive source of this semi-specialized training data – their own products and the unending stream of transactions that move across their business systems.
As with machine translation, the resulting systems will augment human expertise, not replace it. Companies will refine the resulting “horizontal domain” translation tools by identifying subtle differences in otherwise common business language usage across industries, sectors, and regions. The resulting mass-customization capabilities will allow cost-effective product development that can be leveraged across multiple companies and industries.
This can get particularly complicated when vertical- and horizontal-specific semantics overlap. For instance, a word like “deliverable” may only be completely disambiguable when both definitions are known. This need for sophisticated understanding will challenge humans and advanced AI systems for years to come.
Translating a threat into an opportunity
As if the changes in the industry wrought by AI weren’t enough, in addition, the rise of cloud-based machine translation platforms also promises to further drive down the cost of enterprise translation. Platforms consolidate functionality needed by multiple ecosystem partners in a system that is analogous to Amazon’s providing marketing functionality for its network of product suppliers.
While the two forces of AI and cloud-based platforms might appear on the surface to create an existential threat for language services providers (LSPs) and for human translators, the truth is that they will actually create a massive opportunity for those with the foresight and agility to pivot quickly into the new reality.
Lower costs will increase the number of enterprises that can consider sophisticated translation services they haven’t previously been able to justify, driving a sharp increase in demand. Rather than eliminating the need for human translators with specific language, cultural, and vertical domain expertise, cloud-based, AI-powered platforms will instead enable massive improvements in human translators’ capacity (languages, verticals, and horizontals, as above), efficiency, and accuracy.
Providers prepared to transition away from high-cost, hard-to-scale translation models and take advantage of enterprise-focused translation platforms will be able to handle more volume, continue to offer the value-added expertise that differentiates the enterprise and consumer translation markets, and more easily and quickly add new languages.
From an employment point of view, although there will be a massive shift in the work performed by a typical translator, for the reasons given above, we do not foresee a substantial decrease in the need for their services.
This article has explored a massive shift in the translation industry, primarily driven by machine translation (MT) technology and the shift from a premises-based to a platform model. It has examined some commonly held assumptions about how MT developments are shaping the opportunities and challenges facing language services providers and their enterprise software vendor partners, customers, and competitors.
As we have discussed, rapid machine translation improvements and the development of platforms like Amazon Translate, Google Translate, and Microsoft Translator have lowered to near-zero the cost of translation output that is good enough for many consumer and simple business-to-consumer applications. For this reason, many industry players and observers worry that those same trends will reduce the value of enterprise language services providers, independent translators, and corporate localization teams; decrease margins for translation services to unsustainable levels; and eliminate jobs.
In contrast, the continued need for domain expertise and very high accuracy, combined with machine translation and the transition to a platform-based model, actually holds great promise.
Even small enterprises now have global reach previously reserved for only the largest brands. Many must execute global marketing campaigns in multiple languages. This has been prohibitively expensive to date. But, much like Amazon enables a “long tail” of small sellers, emerging cloud translation platforms will consolidate many functions, allowing translators to remain profitable, even when they serve only niche markets.
AI tools will augment rather than replace humans at the high end of the enterprise market, increasing providers’ ability to handle vastly increased volume while simultaneously meeting the stringent requirements of highly specialized translation in healthcare, law, engineering, and other technical verticals. Growth will remain healthy in that segment of the market and prices will drop more slowly than for less-specialized translation.
Enhanced by the increase in the digitalization of business documents worldwide, these trends will drive new demand from enterprises previously unable to justify the cost of enterprise-quality translation and related services and will open up a long tail of opportunities to provide native-language services targeted to small but rapidly growing emerging markets.