There was no need to go to Dublin and attend the MT Summit 2019 to see whether MT has become so good that not using it makes you a monk seal—or a bald eagle, if you look for glory.
The interest in translation technology goes beyond the translation industry. In fact, the global market for MT is expected to reach USD 983.3 million by 2022. It is still a trifle with respect to translation industry market size, regardless of contradictory estimates, but CAGR is twice as much.
No David for MT Goliaths
Globalization has considerably widened the scope of MT and its market for a whole lot of content that would have left untranslated and for previously improbable language pairs that are now within reach for zero-shot NMT.
A big boost to the revitalization of research in MT has come from the rise in computing power and the consequent explosion of data in the last decade of the last century and of the first decade of the current one.
As Mariana Mazzucato brilliantly explained in The Entrepreneurial State, all major innovation and technological developments have come from public funding, with states being the primary risk takers. The private sector only popularized government-created technologies.
The US National Science Foundation funded the algorithm which helped create Google’s search engine, which constituted the base for its subsequent developments. The same goes with MT, whether rule-based, statistical or neural, with the heavy involvement of public organizations, being the Pentagon, the Canadian government, or the EC. And it is not a case if everyone has been using the same technologies, including MT open-source engines.
Not only are Google’s and Microsoft’s engines the dominant players for being easily accessible and, most importantly, free: Like it or not, Google’s NMT is cutting-edge and it is the result of the company’s commitment to all-out research and development and of massive investments.
It takes much money to hire the rightly skilled specialists and arrange for even the basic hardware resources, nothing within everyone’s reach, especially small businesses. Due to the perpetual self-breeding technology hype, innovation, especially in Europe, will still come from public funding, with venture capitalists pouring money only into fast-growing initiatives thus leaving the lead to giants and their increasingly huge expenditure capacity.
As Mariana Mazzucato explained in her latest book, The Value of Everything, this is an effect of Milton Friedman’s shareholder theory that has been poisoning economies for fifty years and that only a few weeks ago the Business Roundtable has declared mistaken. Ironically, some of the members of this organization are those who have been profiting more of the Friedman doctrine.
In fact, expectations are such that virtually none of MT businesses can keep focusing on technology only, thus turning to services to make ends meet. Seemingly purely-tech players depend heavily on direct or indirect public funding and, in many cases, have released nearly no workable product or are struggling. This explains why, in the end, excluding giants, purely-tech players of some reliability can be counted on the fingers of one hand.
Indeed, open-source engines are an invaluable resource for academic institutions that can access public research programs and the relevant funding and attract scholars as well as brilliant and inexpensive PhD students who are possibly eager to sell their knowledge, experience and discoveries to the industry.
As a matter of fact, the relentless improvement of MT technology says that while translation will survive, the translation industry, however young, may not. And it is quite unnerving that more and more translation industry players of any size look for mutual comfort with peers on social media and at events. On the other hand, the saying still goes that “ even growth that is slower than prior years will still mean another good year”. But an increase in turnover, possibly due to a sharp increase in volumes, does not necessarily mean an increase in profits, however slight.
Just think of the recurring issue of MT and IP. The recent indictment of Anthony Levandowski for alleged theft of trade secrets from Google shows that IP actually is a bloody battlefield. And yet, as in the past, trading secrets does not necessarily require sophisticated tricks, and it certainly happens well before translation industry players are involved. The same applies for the arrest of Aleksandr Korshunov on charges of attempting to steal trade secrets from General Electric.
Put the Blame on Mame
As the ‘Mame ’ of the iconic song in Gilda , the translation industry has a culprit for any cataclysmic event that has been affecting it. And a solution too, of course, regardless it is workable or not.
The ‘Mame’ of most translators is the industry itself and its incarnation, LSPs. The biggest LSPs are the ‘Mame’ of smaller LSPs, and so on. It should be obvious, then, that the translation industry is irremediably sick.
Not surprisingly, no one in the industry will ever pick technology as the ‘Mame’ to put the blame on. On the contrary, everyone blatantly claims not to be afraid of technology and shows blind and unwavering faith in the superiority of human beings and the impossibility for machines to take over them.
Ironically, all solutions are quite imaginative, ranging from translators signing their own translations—as if all translations were the same length as a restaurant menu—to packaging them—as if all translations were for printing and presentation can really make the difference in a zero marginal cost society. Not to forget the direct-client Eldorado, the educating-the-client claim, and the delusion of long-term continuous education. And what about the fear of free online MT engines stealing confidential data from segments submitted for suggestions, thus exposing—dumb, ça va sans dire—translators to breaches of contracts and other scaring violations?
The stories of Lewandowski and Korshunov should be enough to show that confidentiality and IP are minor issues when it comes to machine translation, with humans being far more dangerous than machines.
Academics may be even funnier than translators. From their ebony towers they look downward to us mere mortals and dispense wisdom pills, regardless of timing or credibility, with the latest academic delusion of teaching “low-level programming” in translation courses. A (very late) giant leap that just adds to the earlier and greater delusion of EMT.
Of course, AI is the hype of the moment. And everybody wants to be picking at or monkeying around it , possibly more than ever now that an AI system has passed an eighth-grade science test , while expectations were that no AI system could match the language and logic skills that students are expected to have to enter high school.
New Dogs, Old Tricks
Indeed, AI researchers start wondering whether ability measurement tests for AI should base on typically human models. And possibly, also the same tests for human beings should be remodeled. After all, the Turing test measures the ability of a machine to simulate/emulate a human being, not the achievement of the purpose for which it is designed. And school tests require the application of logic to solve problems that follow old logic.
Not surprisingly, a major field of application of ML and DL in translation automation is quality estimation (QE), which still follows the same old error-based approach, though. Another important downstream technique is APE (Automatic Post-Editing), for correcting machine translation output by learning from human corrections, thus falling again in the bias of using the same old error-based approach.
The irony, again, is in the same people claiming that machines are not going to replace humans who also claim that harvesting, aligning and cleaning data have become major activities in the translation industry, which is blatantly false.
Getting Rid of the Delusion
Getting rid of the delusion of an endlessly growing industry must be terribly hard. The recent renewed deal-making frenzy just shows that it takes money to make money, and that acquisitions are the only way to growth and, with BCE’s quantitative easing, they are going to boom where leveraged buyouts are made easy, possibly as a means to support the national economy.
All the money poured in these acquisitions will hardly be used to implement cutting-edge technologies, while remuneration will be subject to higher and higher pressures to repay debts, thus triggering a harsher Gresham’s-Law effect and making the Bodo dilemma even more intricate, with the already lagging educational system still too slow and stubborn to re-adapt and overcome.
On the other hand, in a desperate effort to enrich their offerings with the most advanced technology and win over competition in a race that is getting harder and harder every day, more and more translation business players—especially the larger ones—have been getting caught in AI washing, i.e. labelling their technology as Artificial Intelligence (AI) when it is not even high-end.
Ironically, a primary reason making hard for many translation business leaders to effectively implement AI is that they cannot even tell between machine learning (ML), deep learning (DL) and AI. And this is also why the blockchain dodge virus could still infect the translation industry.
The Weak Points (of a Whole Industry)
Paradoxically, most LSPs and freelancers can perfectly identify the weak points of their industry, which have been spotted and analyzed from time immemorial at events and on trade media. And left unsolved. Unfortunately, most of them are fated to remain unsolved. In fact, the prediction mania affecting the industry always results in medium-term guesses to look forward-thinking, i.e. smart. Unfortunately, today, the medium term does not exist any longer and everything is short-term because “in the long run we are all dead”. This is why it is now late, we are running short of time, and only those who have already faced and solved these problems or are entering the scene with ready-made solutions have some chance of surviving.
The first of these weaknesses is a stagnating productivity for the inability to take full advantage of technology, which is the legacy of the ‘artistic’ approach to translation; this, in turn, results in the inability of tackling the infamous iron triangle and differentiate the offering. These two weaknesses result in and combine with process inefficiency, which hampers the rate of improvement and development, and with the expensiveness and zero scalability of the traditional business model. Recently, the ineptness in handling data security and the modest profitability of certain language combinations turned up the heat, leaving additional room for new entrants in the relevant markets and thus increasing fragmentation, locally and globally, because the strongest point of the industry is still the old “buy low, sell high” model that does not work for stock markets only.
What to Expect
Nobody in the enterprise technology world argues against the benefits of AI. However, given the skills and amount of data to build and work models, the technology might not exactly be for small businesses.
Indeed, harvesting, storing, handling, knowing, and eventually interpreting and understanding data is just one part of AI, the others involving automation and integration, and translation industry businesses, especially the smaller ones but also, and more culpably, the larger ones, have traditionally been lacking in this respect.
It is not late for automation and setting off the digital transformation journey. Translation industry players can start with AI and implement third-party platforms to simplify and integrate workflows across teams and functions.
Do not forget AI is a big race, though, kind of an arms race, with big powers pioneering and being eager to come up first. Do not even think of catching up with them. They are going to win the race, which is going to be long anyhow and a decent position in the end will still bring real benefits.
However, seldom have translation industry businesses the experience and/or in-house expertise to start working with AI. So, what is AI going to bring to translation industry businesses?
According to a recent McKinsey Global Institute report, “What lies ahead is not a sudden robot takeover but a period of ongoing and perhaps accelerated change in how work is organized”. Indeed, in the last 60 years, only one job has been entirely lost to automation, elevator operators. So, a sudden robot takeover is not going to happen any time soon, but a slow robot war of attrition is maybe already underway, meaning that machines are slowly taking individual tasks over, with fewer and fewer employees running the remains.
When automobiles and trucks replaced horses, blacksmiths and farriers went to mechanic’s schools and learnt to apply the same skills to new vehicles. Unfortunately, algorithmic coding and data science do not lend themselves to quick training for an orderly shift in employment and, besides being costly and time-consuming, re-training could not be enough to ensure professional survival for translators and LSP workers, while continuing education would surely prove ineffective as well, as even STEM workers, undoubted beneficiaries of technological innovations, are under threat from constant technology evolution.
Rather than trying to guess which jobs will remain relevant, or delude yourself that yours will not be, perhaps the best survival strategy is to develop transferable skills and remain agile.
What are such transferable skills? Ask yourself this question then try and give yourself an answer: do tagging and categorization require specific skills? Or, like in war-times factories, the new slaves are going to be crowdworkers manually sorting and annotating data?
The alternative to sorting and annotating data might still be found in a niche, as usual, be it terminology or analytics. In both cases, as long as terminology is still going to be reduced to translation-oriented terminography, as in translation education, the niche will be just getting narrower and narrower. Data analytics, on the contrary, could prove much more interesting and susceptible of re-cyclable skills, but LSPs, no matter their size, still do not collect valuable data for use with analytics, e.g. for vendor management, project management, business administration, etc.
With information asymmetry still dominant and automation prevailing in midterm, translation schools will keep focusing on writing their epitaph, churning frustrated graduates with little or no chance of making a living with translation. As usual, these will have to invent something on their own to supplement the useless education received and rely on themselves after trusting some bombastic synapsids.