In Focus: Opportunities of Perils
AI vs Human Interpreters. True Illusion or Phantom Fear?
In Focus: Opportunities of Perils
AI vs Human Interpreters, True Illusion or Phantom Fear?
“Or to take arms against a sea of troubles.”
· Much Ado About Things Which Matter Deeply
· Winning a Battle Is Not Winning a War
· Illusions and Phantoms
Much Ado About Things Which Matter Deeply
Living in the fast-changing world of today is pretty much the navigation in the sea of troubles.
According to Merriam-Webster Dictionary ‘peril is 1: exposure to the risk of being injured, destroyed or lost: DANGER, 2: something that imperils or endangers.’ Perils of the Sea is probably one of the best-known phrases containing the word.
Professional risks associated with the Pandemic-caused interpreting services markets and relevant response have been extensively explored in our previous blog article on Operational Feedback Analysis. In focus here will be the perils associated with the Artificial Intelligence.
So many important and bright people are discussing the subject of AV against Human Interpreters and relevant threats posed to us. Something one may definitely want to hear are the opinions of the legendary and inspirational Barry Slaughter Olsen, Associate Professor at Middlebury Institute of International Studies at Monterey:
and very engaging Ewandro Magalhaes, KUDO’s Chief Language Officer, former UN conference interpreter:
Opinions are many and versatile; hidden and open messages de re the issue in question range from those of extreme alarmists (we all gonna die), to stubborn negationists (the machine ain’t not gonna ever do that). Somewhere in between we find IT people whispering softly trying to be comforting, well, you still have time but voice recognition, machine translation and AI voice over technologies are developing fast, AI is being trained, results are shockingly good, it is not impossible that the time has come for you to begin considering learning new professions, maybe IT development.
However, as Deniz Sasal says in his How to save your job from AI and Automation
“and what if AI soon becomes the best IT developer able to do the work of millions in just few days”?
Probably, there is no much point to add your overtone to this chorus song unless you intend to try giving that some different touch.
Back in the end of 1990s I noticed an interesting phenomenon interpreting for the senior business administrators of leading world corporations operating in something that back then was referred to as the NIS region: except MBAs from very reputable institutions many of them also held diplomas or Ph. D. in purely humanitarian disciplines such as history, philosophy, psychology, etc. I took the liberty of asking one such gentleman: is there a trend now, as so many instances are too numerous to be incidental? His answer was, affirmative, the trend is to hire administrators with such academic background as making decisions in the new markets now it necessitates being able to effectively put things into much broader context than usual. It is always the past norm and new norm; you hardly envision or wish your personal reality to be THAT static so that there is no change at all.
Thus, the focus of this article is putting AI vs Human into the broader context in an attempt to add a bit of emergency management and philosophical dimensions.
This is an absolutely ascertained fact of life: AI is taking many jobs making many people redundant, thus, we are imperiled. However, we are continuously endangered one way or the other, so it is always the matter of your personal emergency preparedness and response. Thus in order to react properly, not to under or over react, it is essential to scope the problem to know the extent of your exposure, degree of actual danger, qualitative and quantitative assessment of threat to first define and then possibly improve your response and remediation capabilities insofar as practically achievable.
Adrian Probst has this insightful video “Will AI Steal Our Jobs? (Fear of Machine Translation)
There he mentions the amount of data generated currently in the whole world that needs to be translated or localized otherwise, stating that all available human translation force in the world would not cope with the problem.
There are reports that in average people spend 75-80 min a day watching social media videos; it is expected that it will be more than 100 min a day next year.
In addition to that it should be mentioned that the global market of language services over the last ten years has doubled in size, reaching 49.6 billion US dollars in 2019, and is substantially bigger now. (www.statista.com)
At the same time, anyone monitoring job postings in such translation marketplaces as ProZ, Translators’ Café, etc., can acknowledge some catastrophic-type decrease of number of translation job, increase of PEMT (Post Editing Machine Translation) assignments and growing demand for Audiovisual translation (which is a pretty different concept of translation which we put in focus in our article on Localization).
So, the market is B.I.G, and will grow. How one will fit into it and get that miniscule share of the billions available out there is another question that, we believe, can be answered only in the result of individual introspection.
Below are a few input elements into analysis to begin with:
· the market is sizable; theoretically, there should be enough work for all of us in different aspects of language services; the market is continually growing;
· AI and automation are absolutely necessary to cope with the amount of data to be rendered into various languages;
· AI-powered technology is getting more effective every day; humans are losing and will continue losing jobs to automation; this is not necessarily the end of the world; it is the new world that theoretically can be a quite comfortable place from the professional standpoint;
· Assumption is that at least in the foreseeable future human aspect of engaging localization services will still remain the essential driving factor.
The latter should be explored in greater detail.
What is Artificial Intelligence? It is not a monster ex machina, it is a set of algorithms developed by humans, not extraterrestrials. First of all, it is not a personality. Maybe for now, but philosophizing on gloom and doom of the future is neither desired nor required here for the purpose of this article. Anyway, it is not now an entity or a party to the process.
Some linguists are so patiently in love with their profession - which is very, very commendable - that they tend to forget about one principal actor: The Client.
Transaction involving provision of a linguistic service assumes interaction of two human entities: the Client and ‘Translator with Equipment’, henceforth referred to as ‘The Translator’. The client chooses to contract an individual translator performing the task using CAT tools like MemoQ, SDL Trados or just MS Word, or “translator as agency” that may use Machine Translation, engaging or not a human linguist for PEMT.
Figuratively speaking, the Client is made up of the localization job’s terms of reference, style sheets, project budget, client’s boss or client’s client directives, TAT (Turn Around Time) requirement. The Translator consists of cognitive and technical capabilities to perform the assignment in time and per budget. And it will always be the client to make contracting decisions. This sounds trivial, however, if this principle is not internalized it will be very difficult for us to develop a response strategy and relevant infrastructure to stay in the market. Focusing exactly on the client’s needs is the necessary pre-requisite.
As yet another legendary man, Peter Baker, the outstanding voice actor, said in one of his empowering videos: do not put much information at your site on how good you did the job for the world’s leading company. People like and need to know what you can do exactly for them, not something you have done so well for the others.
However, the client’s decision is one of the human elements of the equation when it comes to choosing between the AI and a human as ‘the Translator’.
People have different needs in Multimedia translation, sometimes also referred to as Localization, in order to transfer text into another language or culture. Another human element is target audience, or, to put it specifically, effectiveness degree of target audience engagement.
In a sense, the Client is the client to respective target audiences, it is such client that comes to other humans to be heard, convey message, engage. In order to be listened to, such message should be clear, not misleading, attractive and at times entertaining. In general, most people have these intrinsic needs to talk, socialize, get involved in intrigues, have benefit of doubt. So much meaning is transferred using body language, intonation, sarcasm. People want things to be interesting and appealing, they want to be amused, empathized with, and ‘play transactional games’.
There is also the concept of ‘connotation noise’. Oxford Languages definition of ‘connotation’ is an idea or feeling which a word invokes for a person in addition to its literal or primary meaning. People, at times, tend to ‘hear something different’; they say one thing thinking that they have said something different. Then there are cross-cultural problems. A good example is when a major pharmaceutical company wanted to launch a painkiller in the Middle East market and used three drawings for visualization: a man holds his head as if he is in pain, then takes medicine, and at the third picture he is happy, the pain has gone. But people of the target culture read things from right to left, not from left to right…
But localization is not the end of the media translation story, as there is Hyper Localization when texts should be localized even for specific regions of a country. Will AI do all that right, effectively addressing all these numerous nuances and intricacies in a decade to come? Nothing is impossible in the modern world. Saying no is as bad or good as saying yes; it means nothing as we know so little about the future.
Humans mostly do not need a big ‘bag of facts’ in logical wrapping paper. One lady, when feeling depressed, went to supermarkets to talk, have an occasional energizing small talk with personnel as they are trained to radiate vitality and positivism which can be healing, among other things. She shopped for food but was emotionally getting so much more gratis. You can MT your text, generate automatic subtitles and add audio tracks with next-to-perfect machine voices - and it will work perfectly if your goal is to effectively and cost-effectively provide information.
So it appears, at least for now that the issue of AI complete taking over Localization jobs is not about how amazingly good Artificial Intelligence becomes - as good as humans, or better than humans - it can only become almost human. The issue is whether people decide to become something different, choosing to get rid of all those human “things”, existential, psychological, cultural elements. Humans oftentimes are illogical, overly emotional and so forth, and these ‘imperfections’ are one of the fundamental elements of sustainable psychological stability. In a sense, living is a continual and on-going existential crisis, and you need somehow to vent off to occasionally regroup and regenerate. One will be bored, envious and belittled to always talk to someone who is perfect to the point of being humiliating. Without saying this, humans always have the choice to ‘want to want’ to be perfect in everything and gradually excel in something; the machine has no choice but only to become all-perfect for real and do it fast.
This was an attempt to render things in question somewhat philosophical, however, there is another ‘effectiveness and productivity’ issue with ‘tendering’ of AI at present.
Winning a Battle Is Not Winning a War
MT is amazingly effective, remaining remarkable generally, while leaving the omnipresent impression of being an unfinished semi-product specifically. It sounds good until you go to analyzing this in detail when the brightness of the sun of the fun gradually fades away as you go deeper into specifics.
Certain people simply need some sort of orientation - then Machine Translation or YouTube automatically-generated captions will perfectly suffice. AI-powered MT, voice recognition and voice over, especially domain-trained, deliver amazingly good results. With one huge BUT.
Winning a battle is not winning a war.
Suppose, you want to convey something serious in your promotional YouTube video and have decided to rely on automatically-generated subtitles. Here you are saying something important to your clients about your business, and an automatically-generated out-of-context word pops up in subtitles. You are lucky if this word is just incorrect or awkward, but, oftentimes the situation is funny or even ridiculous. Almost certain, you do not want your audience to abruptly laugh when you are so serious; the impression will now be irreparably ruined and will come with a connotation: you say that you give us a quality product but save money on human-written subtitles; if your QA/QC is such, how can we be sure it will not happen to us, then.
If you need to get the financial, tax report, scientific paper, medical examination documents translated, will you completely rely on machine translations, the product of which is so good until you analyze the detail? Most probably not.
This is all is about ‘losing the war’. Well, one can say that it will not be a complete failure, just few annoying facts. This is okay in terms of so-called ‘fuzzy logic’ which is a serious discipline, not something formalized for fun; compared with Aristotle Formal Logic, binary, where everything is either true or false every time, one can be 65% sober or 79% pregnant and this all would be true statements in that system; however, this logic is intended for very special applications.
As Interpreters with Equipment we extensively use AI-powered software and automation, however only to support our DI and AV activities. Those outside the profession probably do not know that for every interpreting assignment you have to prepare spending quite a lot of time reading the reference material, searching things in the Internet and so forth. At times you spend many hours and even days to get ready, notwithstanding the fact that you are experienced in the specific domain or a subject matter. ‘There is a million of things you should know being a translator’, my tutor told me in my sophomore year in the University. ‘Well, it's true, Watson so very true,’ so said Sherlock Holmes. And MT and AI voice recognition are a very, very big help in this regard.
In all our marketing channels we explicitly say that we strictly DO NOT USE MT for written translation. But of course, there is no translation job outside MemoQ - but MemoQ is a productivity tool as well as excellent mechanism for streamlining the work flow and interaction with clients. However, machine translation is a big no-no. Only because it can be the fast-tracked wide boulevard to loosing clients.
There is a remarkable team associated with the Translation Report.com; they assess and evaluate leading translation and localization companies checking them for value by ordering translations to see how information they declare about their quality and turn-around-time corresponds with the actual reality. You may want to visit this educational resource at:
Names will not be named in the interests of professional ethics and non-disclosure, but there was an interesting report. One of the ‘we will deliver excellent quality’ companies turned in the work in two weeks, not in two days as initially declared; ready translations were apparently split in two pieces: ‘traces of machine translation’ were found in one part of text, another was definitely human-produced but left quite a bit to be desired.
What is this all about? The role of human in the allegedly ‘AI-supremacy winning’ situation of today. Not tomorrow or day after tomorrow, but today.
Localization agencies now offer their clients three alternatives: 1) pure MT; 2) MT and human combined with humans doing the Post Editing Machine Translation; 3) pure human translation. The cost will differ dramatically.
Apparently, items 1 and 3 are self-explanatory, and these three alternatives are kind of alluding that number 2 is a cost effective compromise. It is until you go into specifics. One thing should definitely be borne in mind: you get what you pay for. The situation with number 2 is very much nuanced and intricate.
Who will be that human ‘normalizing’ text after the MT? The one meeting the requirement of ‘MA in Translation and minimum 3 years of experience’, or a domain-savvy professional specialist?
There are two elements in this equation.
1. Experienced specific domain free-lance translators have been around for a while to develop a robust client base; the rates for PEMT are two, three, four times lower than those for the ‘real work’; in most situations you would prefer doing something else than work for this kind of money. The Pandemic crisis did not hit translators hard; they have always been working remotely;
2. Translators hate re-doing somebody else’s texts. Most of them will say it is better to start anew, do everything from scratch again rather than correct someone’s product. It is still a lot, a lot of work which is more difficult than translating by yourself, especially because you have to convert other person’s logic, choice of words and style synergizing those with yours into something homogeneous and consistently new; it is not just to correct few things and get 1/3. It is especially difficult with texts to be translated and prepared for subtitling and voice over, as you have to creatively adjust texts, at times shorten them and still preserve the meaning, format, style and ‘intonation’.
Thus, the option number 2 may not be necessarily the cost-effective quality solution.
But it is always the client to make the decision. However, many do not understand that language rendition is rather an art than a craft, let alone the mechanical process. If the situation allows and clients agree to be guided, it is advisable to navigate them through and make them aware of the perils before the onset of this journey into the world of words and cultures. In essence, this is one of the purposes of this article as well. In a sense, making not-informed decision regarding the choice of ‘the Translator’ may eventualize in losing the war before even the first battle begins.
Illusions and Phantoms
It is like with that wonderful phrase with ‘known unknowns and unknown unknowns’; there are true phantoms and illusions and the false ones; it is always good to be able to distinguish between them. We all used to live in the ‘Yellow Submarine’, now we all live in this world of VUCA (Volatility, Uncertainty, Complexity and Ambiguity)
It is even more crucial to have a consultant on your side before setting your sails. But how to get the right one? It is a very personal question. Based on my experience, people like to listen to personal true stories rather than to read a formal instruction. I do not have a readily available personal story stashed away for this case, thus I will just make it up.
Let us make the situation very challenging by choosing a very peculiar culture and language.
Inputs. Assume, I need to do business in the country of Vietnam and I need linguistic support. I have a local friend speaking my language or my compatriot speaking the same. Assume further, I am a health physicist or a radiation safety expert; English is my native or preferred language of communication.
As far as I know, Vietnamese is a tonal language, like Chinese, except it has six tones not four as in Mandarin, thus, every same pronounced word may have at least six meanings depending on the tone - and we will not submerge in cultural things, connotation noise or fuzzy logic. Thus, I will never have a way of understanding of what will be said and how. Naturally, this puts fear in me and the illusion that I will never succeed as there is no mechanism to personally control things. I need a consultant.
This approach we use when in need of finding a partner in language pairs we do not work with. In our team we speak – or understand to a degree - English, Russian, Ukrainian, Belarusian, Polish, Slovak, German, Swedish, French, Spanish, Portuguese and Japanese (when we were young we loved to hop countries and learn languages). Thus we can assess in general the relevant overall linguistic capabilities with regards to these languages. But not Vietnamese.
First thing to assess will be the command of English, the ability to speak in the ‘rich’ manner, addressing nuances, intonate phrases - and only then comes the most important.
The person will be asked to elaborate about the subject in English. Like in any profession, there is the Minimum Minimorum that you should know, if you do not know this or cannot differentiate among basic concepts any further steps will probably be not worth the effort.
Do you differentiate between safety and security concepts (in Slavonic or Romance languages it is usually one word)? What is the difference? Do you differentiate among radiation, exposure and irradiation (e.g. in Russian it is an immediate express test ‘go/not go’ result, as there are words that sound pretty similar but are totally different in meaning)? What are the three basic rules of radiation safety? If you immediately do not come up with Time, Distance and Shielding, too bad. What is the word in your native language for Shielding?
On the peculiar side. Not many people know that fault is also a geological term, meaning fracture in the rocks of Earth's crust, and not ‘responsibility for an accident or misfortune’. Thus, misinterpreted ‘it is our fault’, or ‘here is the fault’ may definitely lead to a serious misfortune that, of course, will be the fault of the interpreter, but so much more of the person who contracted ‘the Translator’.
So, finally, what about the personal ‘verdict’ de re AI, MT and automation?
Well. We believe there is no point to either fear or panic. AI learns, fast, effectively, it will always be around. We consider AI a yet another linguist, just like other linguists competing with us for jobs. Though, it is not a human linguist and this gives AI both competitive advantages and disadvantages.
With or without AI, there have always been and always will be human competitors; some of them have become our friends, colleagues to team up with. With some elements of AI we are already partnering, and we are happy about that.
Perhaps, AI considers us as some kind of friendly algorithm sets as well.
Things are always the same: client will choose the ‘Translator’. In order for clients to show their continued interest in our services we should always learn, excel, improve ourselves in professional specialist domains, be cooperative, helpful and productive.
The last phrases of this article may seem unexpectedly illogical given the overall intonation of this narrative and might sound somewhat out of context at the first sight. But only at the first sight, only in the narrow context.
Now, with AI now playing in the market, nothing has really changed for us essentially and in general. The specifics will continually become different, and the latter trend has never changed as well.
I am not welcoming you into the world of VUCA. We are already there, and, perhaps have always been; it is only the current, the specifics that appears to be different than that of the past.
All the best to you all, indeed. As ProZ says: happy translating.
Written by Andreas (Andrey Parkalov), Z4T Integrated Text and Voice
Photograph by Andrey Parkalov