trumpery in trashlation: MT & AI
trigger warning: truth hurts!
A spectre is haunting Translation World — the spectre of algorithmic trashlastion. All the powers of Corporatism and Cowardice have entered into an unholy alliance to summon this spectre and loose it upon our societies: the Linguistic Sausage Producers (LSPs) of the bulk market bog, also known as Big Transla (or Pig Transla as I prefer, in a nod to one of its worst actors, the UK-based company known as thepigturd), craven “professional” associations who recommend submission to the worst abusive conditions of work and provide a platform for corporate members or the vicarious agents of corporate control, and desperate slavelancers who surrender their agency to bulk market brokers and celebrate the rites of their poverty cult.
“Another leg to stand on,” these wordworkers call the engagement with deskilled and underpaid tasks as they crouch in a tattered web with arachnid aspect over empty bank accounts, trembling at the thought of those scary paginae vacuae and bring skills and self-respect as sacrifice to the loas of spewing algorithmic wordstreams.
Some 43 years ago, Nigel Bevan1 noted concerns for the psychological aspects of post-editing activities and possible effects on language skills:
One fear is that too much post-editing will distort an individual's perception of the language.
Sharon O’Brien noted twelve years ago that
For many, editing is seen as a less creative task than translation… and job satisfaction is further diminished by having to correct machine-generated mistakes or human mistakes propagated by the machine. This is exacerbated by a concern that the more one proofreads and edits, the less well one can translate…2
Translators also fear that prolonged exposure to ‘good enough’ (or, perhaps more appropriately, ‘fit-for-purpose’) quality will reduce their ability to produce high quality, when it is required…3
Dr. O’Brien also stated repeatedly that “no empirical evidence has yet been presented to support such claims” but fails to acknowledge the reluctance of researchers like herself to endanger their funding sources by documenting the lived experience of so many wordworkers in ways that would call into question the subsidized brief to make the case for faster and cheaper without much regard to occupational health concerns.
I first became concerned about the psycholinguistic implications of post-editing work when translation technologist Jost Zetsche began to dabble in it and remarked with some apparent surprise that he found the quality of his translations declined after a lot of post-editing and that it took some time to find his groove again.
And at memoQ Fest 2015 in the Q&A session after Dion Wiggins’ presentation for Omniscien, an LSP principal whose company made heavy use of machine translation with internal resources for post-editing admitted that after doing a lot of machine translation post-editing, staff needed long breaks with other tasks to recover their impeded linguistic skills.
Many of us have probably had the experience of how exposures to certain kinds of text influence our writing, particularly at an impressionable age. After I read J. Stuart Mill’s On Liberty at age 15, I submitted an essay in my English class where one sentence extended over three full, handwritten pages. The teacher gave me an A+ because he spent over two hours looking for some grammatical error and found none, but threatened to fail me in the class if I ever repeated that performance.
For more than decade now, I have been deeply concerned that the inclusion of significant work with machine translation and post-editing in university courses runs the risk of damaging the developing skills of young writers, perhaps capping their skills at a level well below their native intellectual capacity.
And now in the time of an AI “gold rush” with chatbot-fueled writing, translation and editing, there are additional concerns of how the mediocrity of LLM output may also degrade or inhibit the development of writing and translation skills, leaving us with a jar full of half-dead lightning bugs, with no lightning on the horizon. Mark Twain would be appalled at the educational malpractice that encourages such modes of work.
A backlash against this has begun as the sizzle sold to the masses has proven to have not a steak but just some chewy singed shoe leather behind it. Goldman-Sachs issued a report recently in which they busted myths and hopes for return-on-investment, noting that even when the automation processes sped things up, the cost of human labor was much lower. And then, of course, there are the environmental concerns, with data centers used for LLMs sucking up unbelievable and unsustainable quantities of energy while people die for lack of electricity to run their air conditioning.
Is it ethical to have ChatGPT prepare your terminology candidate lists when that may contribute to the death of an elderly neighbor when the power available for communities is insufficient? Or what about that energy-intensive AI summary of your Google search? Is my time so important that I should dismiss concerns about environmental impact if I can save an hour?
Another of my concerns is my lived experience of the value of boredom, and the discipline of trivial tasks. It might be tedious to spend an hour or two rephrasing an unwieldy, overly complex paragraph in an essay, and perhaps ChatGPT might offer time-saving suggestions, but relying on these over time will result in an intellectual slackness and a loss of the sharp edge I can sometimes find in writing and find more as I take the time to consider, slowly, possible rewrites and their implications. Such an exercise builds the mind in ways that algorithm-fed spew is unlikely to accomplish.
My greatest breakthroughs in so many endeavors have come at times when I am engaged in boring, time-consuming and often messy considerations of a problem. But inevitably the result achieved is superior to anything those “gods” in the machine could give me for the sacrifice of my mind.
And as so many wordsmiths with decades of experience in rendering the linguistic and cultural treasures of text from one social, political, business or scientific context in the source language to appropriate equivalents in other languages call it quits to escape the intellectual and economic degradation of their machine-haunted former profession, perhaps we should reconsider the way before it leads us all over the cliff.
Nigel Bevan. 1981. Psychological and ergonomic factors in machine translation, p. 78, in Translating and the Computer: Practical experience of machine translation, London, UK.
Sharon O’Brien, 2012. Translation as Human-Computer Interaction, p.13, School of Applied Language and Intercultural Studies Centre for Translation and Textual Studies, Dublin City University, Ireland.
Sharon O’Brien, ibid, p.15.