Pages

Monday, July 27, 2020

Observations on the Translation Industry

This is a guest post by a frequent contributor on this blog: Luigi Muzii. Here he shares observations on some key trends in the professional translation industry. His observations are presented as pieces of a jigsaw puzzle and readers can connect them or not as they wish. His opinions are his own, but I like to include them on this platform as they often ring true and show a keener sense of observation than we typically find in the localization media.

 He and I have both been saying for many years that disintermediation and disruption are coming to the industry, but we have yet to see a real fundamental change in the way things are done. This may be because the industry is highly fragmented and the inertia requires much more force to enable the needed structural change. There has been some change, but it has been slow and incremental. Or, quite possibly it may simply be that we are both wrong on this prediction of inevitable disruption.

After considering his observations here again, I think that it is perhaps, that the timing is hard to predict. MT has taken over a decade to even moderately penetrate the industry, and it is my opinion that it is still most often sub-optimally or wrongly used in the localization world. For real disintermediation to take place tools, processes, and solutions all have to evolve and align together in a meaningful way.  

Luigi often points to the practice of emphasizing the wrong aspects of the business challenges in the industry in many of his observations. This little clip makes this clear for those who still find his observations somewhat opaque.




“The reason why it is so difficult for existing firms to capitalize on disruptive innovations is that their processes and their business model that make them good at the existing business actually make them bad at competing for the disruption.”

'Disruption' is, at its core, a really powerful idea. Everyone hijacks the idea to do whatever they want now. It's the same way people hijacked the word 'paradigm' to justify lame things they're trying to sell to mankind."
'Disruption' is, at its core, a really powerful idea. Everyone hijacks the idea to do whatever they want now. It's the same way people hijacked the word 'paradigm' to justify lame things they're trying to sell to mankind.
Read more at https://www.brainyquote.com/topics/disruption-quotes
'Disruption' is, at its core, a really powerful idea. Everyone hijacks the idea to do whatever they want now. It's the same way people hijacked the word 'paradigm' to justify lame things they're trying to sell to mankind.
Read more at https://www.brainyquote.com/topics/disruption-quotes
Clay Christensen


“Life’s too short to build something nobody wants.”
Ash Maurya

“If you always do what you always did, you will always get what you always got.”
Albert Einstein

In the last week or so, there has been much clamor about the "magical" and "astounding" GPT-3 capabilities that can "create" and generate text by drawing from a HUGE language model. More data equals better AI, right? They say that GPT-3 is different because it creates. GPT-3 is a text-generation API. You give it a topic, and it spits back a (hopefully) coherent passage. It learns over time, tracking not just what it thinks your topic is about, but how you talk about that topic. 

Some of the examples of GPT-3 intelligence being shared in the Twitterverse are truly remarkable, but while I am indeed impressed, I think we should also maintain some skepticism about this "breakthrough" until we better understand the limitations. I will not be surprised to see overenthusiastic feedback from the LSP industry just as we saw with NMT. This thread has some interesting discussion and varied viewpoints on GPT-3.   



My initial impression is that is indeed a great leap forward, but it has two very serious flaws that come immediately to mind:
  1. It lacks common sense as does all deep learning based AI that I have seen,
  2. It is unable to admit that it does not know.
However, GPT-3 already appears to have the potential to displace mediocre marketing content producers, just as MT displaced some mediocre or bad translators. As more competent people test it and play with it, we will uncover the problems it is best suited to address. I look forward to hearing more about the production use of the technology and real use cases.


The difference between stupidity and genius is that genius has its limits. 




========


A Jigsaw Puzzle

Over the last week or two, several topics have jumbled together in my mind. While they may seem unconnected, I do see a thread that binds them together. Commenting on each of these subjects separately would have meant breaking that thread, so they are presented together here as jigsaw tiles, that the reader may wish to combine to build an overall picture.


Ethnocentrism

Ethnocentrism is the original sin of globalization and one of the capital sins of internationalization. Most often, incorrect localization is like the fruit of the poisonous tree.

Writing full strings with as few variables as possible should be the most basic lesson in a Software Internationalization 101 course.

Context helps, syntactic gimmicks don’t.

Using an active voice is always better than using a passive one.

Gender issues should be left to localizers. Paying too much attention to use gender-neutral forms and words from strings (and content in general) won’t help translators do their job. On the contrary, they will make it harder, forcing translators to develop solutions that hardly sound as natural as the neutral English source material does. These translator modifications are not necessarily as neutral in another language, especially when an ending vowel can make a difference

Beyond being a silly stereotype, “thinking outside the binary box” to prevent using gendered language does not necessarily lead to effective communication.

Removing pork or cow meat from menus will not help per se increasing restaurant sales in Muslim or Hindu countries. However, redesigning the menu probably will. And this is a fundamental lesson in globalization 101.

All this reminds me of the launch of Windows 95 when you consider the initial localization attempts of the “Start” button and the sudden abandonment of the “Start Me Upguitar chord as the accompanying jingle, which of course, makes much less sense in non-Anglo cultures.

Ethnocentrism could appear even in a theoretically unbiased approach to writing. Being a linguist does not necessarily mean that one is also open, inclusive, and global. The editor of a historically-popular trade magazine, who was also a translator, was also a prominent figure in the formation of the not so inclusive UKIP.

Inclusive language is something localizers and translators need no specific guide for. Sexist, racist, or otherwise biased, prejudiced language and ideas cannot be prevented from spreading, and translators have to deal with this daily. And they know how to cope with this phenomenon. Most importantly, they know how not to be influenced by this in doing their job. It’s called ethics.

It is wise, though, to request that vendors notify customers whenever they find language that isn’t inclusive, at least when inclusiveness is a pre-requisite. A customer’s task requirements guidelines should clarify whether a translator should keep the non-inclusive language intact — requirement specifications: such strange stuff.

Guidelines on using inclusive language may be useful for authors when machine translation is going to be involved. Much too often, people prefer to ignore that bias in AI and MT doesn’t come from algorithms, but from the people who developed the technology, and it reflects their values. Biased preference comes from training data even more than from input data. Training data are examples from which computers learn patterns and build predictive models. And this historical data is usually coming from real examples of human/social attitudes in the past.


Pandemic Crisis ‘Secondary’ Effects

The effects of the ongoing pandemic may have different readings — some of these readings concerns the broadening of the gig economy.

According to recent reports, the gig economy is taking over the enterprise. That more employees opt for a more flexible work structure may be one reading. Another one is that it is invaluable for organizations seeking to streamline and reduce costs.

Gig jobs are no longer limited to lower-paying work performed on-demand, and it seems that organizations have started taking advantage of more valuable employees. Gig jobs in the white-collar world has significantly increased in the past few years. 72 percent of all gig jobs worldwide between 2018 and 2019 were in large enterprise and professional services firms, and, according to Deloitte, gig workers in the US are going to triple to 42 million workers in 2020.

Quoting Gigster’s CEO Chris Keene, “Companies have always valued the ability to increase capacity without increasing costs.”

The impact of the gig economy on professionals that very few seem to see is that it exploits the demand for jobs to push remuneration lower and lower. No one pays attention to building a meritocracy: performance ratings and rankings are just truncheons.

What remains of the gig economy is a blessing for post-pandemic corporate recovery who can avoid hiring back thousands of full-time employees laid off or furloughed. Quoting Chris Keene again, “Coming out of this pandemic, there are a lot of jobs that people are not going to be able to come back to.” The pandemic crisis has had the gig economy jump a decade forward and pushed capitalism and its mission to a peak, i.e., increase profits and reduce costs to the maximum possible level.

This cost reduction focus is an unrelenting mission, as recent German slaughterhouse outbreak cases of the coronavirus showed. The specific impact of the cost-reduction focus, in this case, was to force close contact amongst workers in feverish working conditions needed to produce cheap meat. Of course, reports showed that otherwise despised migrants provided almost all the cheap labor. The German NGG union spoke of “shameful and inhumane conditions.”

The usual justification is that better working conditions involve higher prices. But are low prices really low? Higher prices always hide behind low prices.


Mainstream

Now that machine translation is finally mainstream [in the translation services industry,] nobody questions its use anymore. But still, the debate around MT use has taken on the same quagmire issues as those around localization translation in general. This means that, as Kirti Vashee, wittily notes, “the quality discussion remains muddy.”

Translation industry attention focuses mostly on edit distance, post-editing effort assessment, post-editing practices, and overall effectiveness measurement. Not surprisingly, discussions focus primarily, if not exclusively, on assessing the quality of machine translation output rather than on how to improve overall MT system capabilities, and shoddy tools like DQF receive all too much consideration.

Indeed, data and its understanding draw little or no interest, despite the clear enterprise market interest in an MT offering. This lack of focus is due not only to the fact that the LSP MT offering is not transparent, is unconvincing, and often poorly focused. Despite the interest of enterprise customers in MT, the relatively good performance of (almost) free online MT engines create a disincentive for LSPs to invest. LSPs are reluctant to explore a territory that seems outside their traditional scope of business and expertise.

Helping machine translation systems handle inclusive language is not just a matter of focus on training data, just as producing good content downstream is not just a matter of effective post-editing practices.

Preemptive quality assessment (or a priori risk assessment, as some call it) is only as effective as the training data is useful. Also, error detection and correction capabilities are crucial, at least as long as quality assessment still heavily depend on inspections.

Information asymmetry also applies to machine translation. Estimating risk only for the output without taking into account the source data, process conditions (especially buyer requirements), and the expected results do not raise high hopes per se. If you are unable to measure these three parameters according to consistent and parallel metrics and produce a weighted mean, you will face misleading estimates. Last but not least, insistence on segment-based rather than document-based analysis will not get you out of the narrow enclave in which the translation community has been basking for centuries.


Disintermediation Is Not A Vending Machine


And no ATM either.

At the WWDC 2020, Apple revealed that version 14 of iOS would come with a translation app specifically designed to translate conversations in 11 languages. An on-device mode will also be available to allow offline translations.

Should this be interpreted as another sign of the imminent end of the translation industry? The industry is most probably doomed, but its end is not set to come tomorrow.

The end of the industry will come from disintermediation. Some, including “yours truly,” have been writing (and talking) about this happening for a decade. Others are speaking more quietly about this more recently. More precisely, the usual suspects made some enthusiastic, although scanty, comments when Lionbridge launched its BPaaS platform, onDemand, five years ago or so.

Recently, though somewhat belatedly, SDL has struck back with its self-service, on-demand platform, SLATE.

Disintermediation is almost inexorable in the evolution of the global (digital) village, where intermediaries are generally seen as the villain. However, they are everywhere online, despite the common belief that they are not (e.g., Airbnb, Amazon, Booking.com, eBay, Expedia, Instacart, Uber, the food delivery companies, the app stores, just to name a few). Following the typical marketing model of rechristening old things by giving them glamorous or more palatable names, they are simply renamed as two-sided markets.

Incidentally, Lionbridge’s OnDemand was quickly, abruptly and mysteriously discontinued despite its boasted growth of 68 percent in one year with reportedly impressive scores of 99.8 percent on-time delivery rate, 99.4 percent revision-free project rate, and 99 percent of users satisfied or very satisfied (85 percent) with their customer care.

Lionbridge onDemand’s should have turned language services into items that could be bought through an e-catalog via a procure-to-pay system.

This “productization” approach involved standardizing options and making pricing instant. The idea behind it was to entice business stakeholders with 24/7 access, faster turnaround times and lower prices, while providing higher visibility into total-cost-per-output and rate-card negotiations, thus curbing the vendors’ role and their ability to add fees and lengthen lead times.

The pricing model was the traditional word rate model, while for its self-service platform, SDL offers a subscription model (SLA anyone?).

Today’s fundamental question is the same as then: Who and what are these platforms for?

As Semir Mehadžić brilliantly noted, beyond the aim of ‘cutting out the middleman’ childishly coming from typical middlemen, a BPaaS should come up with a better value proposition than the one currently used, i.e., “fewer clicks” and “avoiding the use of Google Translate.”

In the projected perspective, self-service translation platforms may entice consumers, but hardly any businesses.

The businesses such platforms can entice are typically new to translation and the translation industry, usually, companies entering international markets for the first time. Such companies generally go along a long and painful track of word of mouth and web search to find a vendor that suits their needs. Then inquiries and quotes follow, and leave the business managers puzzled and hesitant with their heads spinning and aching. The many quotes collected differ substantially from one another, and all look invariably too costly, mainly because the service offered is essentially the same.

Therefore, if the ideal recipient for a self-service platform is the consumer (e.g., Translated.net, One-Hour Translations, Lingo24, Gengo, tolingo, etc.), SDL’s offering, with its SLA-like model, is aiming at SME’s while saving on sales and account management costs. Probably because SMEs would typically not approach a large LSP since they presume that they would not find the same responsiveness, flexibility, and speed.

And this only happens if everything goes well because those SME managers described above might easily bump into an LSP salesperson who tries to educate the prospective customer about the intrinsic value of translation and the wonders of CATs and TMSs. Unfortunately, there is no inherent value in service offerings, only a perceived one, and while the prospect customer knows this, maybe the salesman does not. And the selling effort is thus burnt.

Therefore, self-service translation platforms might target the consumer market, where SMEs with occasional translation jobs can also be found. However, to reach the consumer market, substantial investments are required to be always on top of SERPs and get the necessary conversions. The future effect of these platforms may thus further accelerate commoditization of translation service businesses.

This unintended impact should be feared by self-service translation platforms in particular, as it would require that they will need to sell more and more translations just to stay even in revenue terms. The situation is similar to the dilemma of vending machine suppliers. They need to continuously sell more and more vending machines and find cheaper and cheaper products to include in them.

Anyway, all of these DIY instant translation platforms look like their designers know little about how the need for translation arises in business, and how translations are performed and delivered. More importantly, they look as if they don’t know - or even care - about customer satisfaction and how this is expressed and assessed.

It is my observation, that these allegedly “new offerings” are usually just a response to the same offering from competitors. They should not be equated to disintermediation and they often backfire, both in terms of business impact and brand image deterioration. They all seem to look like dubious, unsound initiatives instigated by Dilbert’s pointy-haired boss. And the Peter principle rules again here and should be considered together with Cipolla’s laws of stupidity, which state that a stupid person is more dangerous than a pillager and often does more damage to the general welfare of others.




Luigi Muzii's profile photo


Luigi Muzii has been in the "translation business" since 1982 and has been a business consultant since 2002, in the translation and localization industry through his firm . He focuses on helping customers choose and implement best-suited technologies and redesign their business processes for the greatest effectiveness of translation and localization related work.

This link provides access to his other blog posts.