Many in the language industry think that whispered interpreting owes its spreading, at least in the political arena, to French General Charles de Gaulle who was known for not grasping any foreign language and always having a French interpreter with him when meeting foreign generals and statesmen. Regardless of any jokes that could be made on the personality of General de Gaulle, who remains a prominent figure in history, this is a testimony of a fundamental issue in translation: trust.
With different nuances, the same issue affected the Nuremberg trials, which saw the introduction of simultaneous interpreting. Many see the trials as a form of victor’s justice and simultaneous interpreting as a fig leaf for fairness and lawfulness. In fact, serious doubts arose as to whether interpretation provided a fair trial for the defendants, for possible mistranslations and transcript errors, and criticism from lawyers and other legal professionals was extensive and severe. Despite the long and accurate recruiting procedures, several questions arose also on the capability and reliability of interpreters and translators.
Trust is a crucial element in game theory and in information asymmetry, which rules the translation industry and should be apparent to anyone in the translation industry: sellers usually know more about their products than buyers, with the consequent imbalance of power and the risk for transactions to go awry. And yet, it is really hard to convince people in this industry that information asymmetry is applicable to translation.
Translation buyers cannot accurately assess the value of their purchase through examination before sale, so offerings are usually compared on price. Also, information asymmetry leads sellers to pass off a low-quality product as a higher-quality one, while decreasing profit margins lead sellers to resort to lower resources, in an endless downward loop.
Information asymmetry leads to a basic mistrust of buyers in the seller’s quality. Also, happy customers usually do not manifest any appreciation for a translation provider, as quality is taken for granted, so ordinary customers can’t see why translation should be any expensive, as well as intrinsically unverifiable and ineluctably faulty.
Anyway, fifty years after the first enunciations of the functionalist approach to translation, something is moving. Imperceptibly, but moving, thanks to Microsoft’s marketing bragging about achieving human parity in machine translation, which led a group of researchers dispute that claim by arguing that quality assessment should be run at a document level instead of comparing output at the sentence level. Essentially, the researchers points to a failure of current best practices in machine translation evaluation that, however, all come from the typical linguistic approach that still constitutes standard practice also in the HT community.
Although Cicero and his “sensum de sensu”—yes, Cicero, not Jerome— came long before Eugene Nida, the error-catching approach at the sentence level still awaits to be amended, although largely superseded. And the red-pen syndrome is still to be cured.
Customers are interested in knowing whether and to what extent the money spent results in a reliable product, which they hardly can assess on their own in all its aspects and beyond any bias. This also explains, at least partially, the enduring success of approximate and seemingly rough automatic measure in MT assessment.
This trend resounds in a recent 39-page report telling the results of a study supervised by Erik Brynjolfsson on how much machine translation affects international trade based on administrative data from eBay. With Andrew McAfee, Brynjolfsson authored two best-selling must-read books on technology, Race Against the Machine and The Second Machine Age.
The most striking number emerging from the study is the 17.5% increase of international trade on eBay following the introduction of machine translation, with a 13.1% increase in revenue. All this against a “modest” 7% increase in human acceptance of machine translation.
The MT engine translates buyers’ search queries and item titles. The study shows that users simply consume the MT output, but otherwise need not change their buying or selling process. While they care about the quality of the translation, it makes no difference whether the translation was produced by a human or machine.
The impact of MT at eBay is equivalent of the export increase from reducing distances between countries by 37.3%. Interestingly, the increase in exports is larger for differentiated products, cheaper products, listings with more words in their title. Machine translation also causes a greater increase in exports to less experienced buyers.
Most translation industry players, notably LSPs, are still struggling to find their own way to automation. Unfortunately, all efforts look disorganized and jammed, as if they were aimed at providing visibility and securing a place in translation history to their promoters rather than at tackling, and possibly winning, the challenges of the near, immediate future. It looks as if they are thinking about how to breed a stronger horse while the universal motor vehicle is already on sale, at a much lower price.
Together with very low entry barriers and market pulverization, rough and obsolete business models and a widespread technological unsophistication are knowable features that make the translation industry the perfect, ideal candidate for disruption.
Anyway, disruption does not happen all of a sudden. Seldom are established market-leading firms, products, and alliances displaced by any single innovation, and disruption usually comes from outsiders, rather than existing market-leading companies. In fact, any innovation, however profitable, requires resources that market leaders are hardly willing to take away from sustaining competition.
Therefore, disruption can easily hurt a company in difficulty as well as a long-stablished one, even with significant research and development, if the market offers very tight profit margins and this company fails in achieving a good growth rate.
Also, the translation industry is supposed to be a global one, but most of its players can hardly think globally and mostly act locally. And this is another reason for the historic pulverization of the industry and the difficulty of growing of its players. And behind the uninterrupted growth of revenues, there is a story that has yet to be told: While volumes have been possibly growing as much as revenues, profits have not been growing at the same pace. In the last twenty-five years, prices have been undergoing an increasing pressure with compensations remaining, at best, unchanged. Over the same period, IT has made volumes increase by at least a 10x factor while productivity has, at most, tripled.
Although pulverized, technologically unsophisticated and trivial—in business line—, there is little interest for disrupting the translation industry because of the difficulty in retaining customers and ensuring suitably large margins—hence profits—for a reasonable time. In any market as mature, open, and aggressive as the translation market, price is king and price competition is not exactly alluring for anyone, especially for newcomers.
Disruption involves process innovation, technological innovation and fast development, possibly via outside funding, and, even taking the not-so-reliable Slator into account, translation does not seem to be an intriguing arena for startups.
What should be an interesting innovation track? The translation industry is notoriously made up of very light players, who rely almost exclusively on outsourcing. Recruiting might then be a good area for innovation and automation. Not crowdsourcing, though, no more and not even freelancer matching. In fact, as Bodo Vahldieck, Quality Manager at VMware, noticed at the TAUS Industry Summit in 2017, the abundance of technology, data and knowledge in a few companies is not enough to attract young talents who have studied languages and translation. The educational system has failed so far in accepting the idea that translation competence is now a three-legged table, less and less a question of language knowledge and more one of knowing how to use it and the right tools to exploit it. And the gap has now become too wide to fill. This tells why most translation industry players share virtually the same pools of vendors.
This tells why vendor management and quality management are two interesting areas of application for machine learning (grossly AI).
Vendor management involves requesting bids, vetting many mixed resources, and scrutinizing and comparing different offers. It is still a labor-intensive human task requiring specific skills and varied abilities, with the end goal to select the vendor consistently providing the highest quality for the best price at the right time. With the right set of historical data, this task can now largely be automated.
ML can also be used to predict translation quality outcomes calculated for all completed jobs, to allow predictive scores to be compared with actual outcomes. Post-factum quality scores can be computed from content profiling and initial requirements (checklists), traditional translation “QA” (i.e. checking for machine-detectable errors in punctuation, numbers, inline tags, capitalization or extra spaces, missing translations or terminology inconsistencies,) correlation and dependence, precision and recall, and edit distance scores. ML algorithms could be implemented to forward on only samples that are questionable thus possibly requiring humans to review.
All this said, the main challenge for the immediate future will be the lack of resources, either because of insufficient turnover, unfitting new entrants, or lack of appeal of the industry and its jobs.
Working at an LSP might not be exactly exciting, challenging or rewarding, especially if the company is small-size and locally oriented. Managing complex projects, figuring out the best workflow, delivering to demanding deadlines, working with people from diverse cultural and linguistic backgrounds is much harder to happen than one might think. And remunerations are regularly below average, especially for a brilliant graduate. To make things worse, in the average micro-sized typical LSP, progression is a mirage, and the chance to learn and develop is virtually non-existing.
One can actually find a multicultural environment when having the opportunity to attend a global industry event, but the associated costs, starting from registration fees, are generally unaffordable for most LSPs and freelancers, and programs just vaguely intriguing. The passion for languages may easily and quickly be frustrated, with the opportunity to use some foreign language occasionally, or on trivial tasks.
The narrative could be different and telling about “a wider breadth of job roles” and “healthy levels of employment globally” while inviting not to underestimate the real and looming risk of losing key talent to competing industries. As a matter of fact, though, the hiring that hit the news are mostly in sales. Curious, isn’t it? After all, there is where turnover of staff carries just a little cost—definitely not for training—one that can be driven down by heeding advice and implementing good people management strategies.
Italians use the expression “fare le nozze coi fichi secchi” (make a wedding with dry figs) for an attempt to accomplish something without spending what is necessary, while Spanish say “bueno y barato no caben en un zapato” (good and cheap don’t fit in a shoe). Both expressions recall the popular adage “There ain’t no such thing as a free lunch”.
This idea is common to virtually every culture, and yet translation industry players still have to learn it, and possibly not forget it.
Bad translators have been driving out the good ones for a long time now.
Translation fundamentalists may have won a few battles, but they have definitely and dramatically lost the war, failing in improving the profession they have been claiming to fight for. Recently, some of them timidly admitted that they would hardly survive in the current dire situation as they have done for a lifetime and that their tiny little business might be doomed. In fact, any new translator who would try to succeed in translation today would experience a very bad time.
The reason, however, is not in free machine translation services but in the competition with other desperate human beings for low-paying jobs (gigs) from we-do-all-in-every-language, even more desperate, LSPs looking at global markets from a local standpoint. Very low pays do not allow anyone to specialize or develop and nurture in their customers the idea of an industry of desperate idiots.
Why then LSPs, even the larger ones, stubbornly persist on paying less and less their vendors even when they insist in defining them highly trained and qualified knowledge workers? Of course, the pressure on prices from their customers is one reason, but the main reason is the traditional business model, which, however faulty, still reigns undaunted.
In the short term, major translation buyers will reduce their total spend. The choice for small- and mid-sized LSPs to contrast this trend will be between lowering costs/prices and look for merging with other LSPs to improve their offering, to achieve a semblance of scale economies. Meanwhile, major translation buyers will be also looking for greater efficiency and lower prices. A new pricing war will make translation and the translation even less attractive, while the increasing pace of innovation will push out the players that have been not investing or that have been investing badly, i.e. too late, too fast, and possibly in the wrong resources.
Democratization of AI will possibly arrive announced, but it will take many by surprise anyway. Also, with content doubling every year, the growth of demand for translation will outstrip the rate at which new translators can be trained and enter the industry.
It takes years to create a professional translator and even more to reshape educational initiatives to meet the new and changing requirements. Unfortunately, as John Maynard Keynes wrote in A Tract on Monetary Reform (1923), “The long run is a misleading guide to current affairs. In the long run we are all dead.”
In the short term, i.e. one to three years, the demand is going to peak for people able to collect, clean, and assemble training data, for others able to rate machine translation services, pick the one that best suits one specific job and fix the output, and for specialists who can support any standardization efforts. Traditional translation education will be justified—and survive—only for quantitatively much fewer editorial, specifically literary jobs and for highly creative trans-cultural copywriting jobs. Forget about ‘transcreation’: It is a scam word invented by imaginative LSPs to apply much higher prices and grant barely higher rates, i.e. rise more lucrative margins.
QBRs and blockchain are just other mass-distraction weapons in an industry where the top 50 companies together barely represent 20% of global revenues, 99% of translation tasks are performed by free machine translation services, and most LSPs hardly have any loyal customers who would be willing to meet a minor vendor every three months to discuss business with them as if they were trusted business advisors, someone those customers might be willing to approach and actively involve in their business plans. Maybe one or two of those top companies may boast one or two of such customers in their portfolios; what about the others?
The same goes for blockchain: The computational power required for the validation mechanism is huge and only attainable through computer farms, which are out of reach for most LSPs. Implementing a blockchain as well as granting access to it is no laughing matter; boasting to do so or just declaring to be willing to do so is a brash and unnecessarily expensive marketing initiative that could prove self-defeating with prospects and customers who are tech savvy enough to be healthily skeptical.
All LSPs, not just Californian, fear that “the added cost of providing full benefits to every single contractor would likely put many [of them] in danger of going out of business.”
Anyway, hiding your head in the sand and pretending that everything is going great is the solution. There’s no need to tell a lie when you can bullshit your way through. What’s the use of hypes otherwise?