Perfecting the Babel Fish
• Photo by Eddi Aguirre
Words by Hollen Singleton
Photos sourced from Unsplash
This story was originally published in Issue 3
How close are we to seamless universal communication? We consider the complexities of language and the limitations of Google Translate as a medium for human connection.
“If you stick a Babel fish in your ear you can instantly understand anything said to you in any form of language.”
When Douglas Adams invented the Babel fish for his 1979 five-book ‘trilogy’, 'The Hitchhiker’s Guide to the Galaxy', he created a wish-fulfilment fish for every traveller. In my own travels, I have wished for this thing – an unseen and instantaneous translator – especially when I’m trying to say something important like, “I’m sorry I stepped on your foot with my enormous hiking boot,” or, “What boat should I be on right now?” The Babel fish has such potential, could be “so mindbogglingly useful” to both the individual and the world, that the idea of a yellow, slippery helper has stuck with me.
But Adams did not invent the desire for universal understanding; he named the Babel fish after a Genesis story with similar themes. According to biblical myth, Noah's post-flood descendants lived in Babylon and spoke the same language. Humans were cooperating so well that they began work on the famous Tower of Babel, trying to reach Heaven and impress their greatness upon the world. This offended God, who struck down the tower and “confounded” everyone with different languages so they could no longer understand one another. And so, perhaps due to this great confounding, we mortals are left to wrestle open the doors of our separate languages. So last year, when I tried to tell a hotel clerk in Bogotá I was leaving the city early the next morning, I unknowingly asked – or worse, demanded – her to go with me. “Vamos mañana…?” She declined.
As it stands, there are more than 6,500 languages in use worldwide. And yet, as early as the 1940s, pioneer computer scientists considered a universal translation machine an easy matter – a realistic development of the near future. Languages are rules-based systems, it was thought, and could be codified successfully with some patience.
Problem is, language does not behave in the way grammar sticklers and scientists might imagine. Language is not a defined set of words that behave according to rules; it is fluid and full of anomalies. Grammar rules exist to serve communication, a thing that occurs between fickle human beings. The exceptions to the rules – the bending and breaking of rules to maintain communication, as well as the rapid advance of new words into common usage – have stymied universal translation machines. As an example, 2018 additions to the Oxford English Dictionary include ‘mansplain’, ‘Kegel’, ‘hangry’, ‘binge-watch’ and, a full 12 years after its invention and airing on The Simpsons, ‘cromulent’. For universal translation to be possible, a machine would need an unthinkably huge pool of translated textual information, or parallel corpora, to draw from – like a million Rosetta Stones in different languages – as well as constant human feedback, relaying the news about the changes in language. Enter the internet.
The translator at the forefront of its field, Google Translate, relies on a technology called neural networks. These are used in a variety of modern-day online interactions. Neural networks work after being fed information (parallel corpus), which they’re taught to recognise and respond to before being graded on their level of success. Google Translate uses statistical analysis of translated texts found online to rate the likelihood of a word or phrase meaning one thing or another. There is much that can and does go wrong. Neural networks need to be tweaked and, as their processes are not completely understood, this must be done by trial and error.
In the last five years, there has a been a range of products released, kick-started and otherwise launched into the world to facilitate translation between a handful of languages. Even the US Army released a limited device with a bank of Arabic-English phrases to its personnel. But none came close to the Google Neural Machine Translation system. Google Translate has become the go-to solution for language dilemmas all over the world. In my own travels, I have used it as both a medium for conversation and a last resort, exchanging my phone for every sentence. Every day, Google Translate runs 143 billion words through its system.
• Photo by Ilya Ilford
• Photo by David Clarke
Yet the release of Google Pixel Buds met with resounding disappointment. Part of the reason for this might be that the earbuds, supposed to take in speech and relay it back in your language, are merely expensive, wearable supplements to what the app can do for free. Secondly, for all its uncanny abilities and deep learning and artificial intelligence, machine translation is not good enough yet.
Perhaps we have become accustomed to perfection and ease with technology. Still, there’s a reason why Google Translate hasn’t replaced human translators in international industries, at a governmental level or within the military. In 2016, Google fixed a bug that translated ‘Russian Federation’ as ‘Mordor.’ In its response, Google stated that this bug did not originate from a prank, but was automatically generated through the same neural process of looking for patterns in countless passages of text. This process concluded that, statistically, Russia was most likely to be Tolkien’s fictional “barren wasteland, riddled with fire and ash and dust”.
Google Translate, while often coming up with a sensible, legible translation, can also generate critical errors that, unlike a human interpreter or even novice speaker, cannot be queried. Google translates one way; it does not rephrase or equivocate. As an example, I ran the Australian anthem through Google Translate into Japanese and back into English:
Australians all let us rejoice
For we are young and free
We've golden soil and wealth for toil
Our home is girt by sea
Our land abounds in nature's gifts
Of beauty rich and rare
In history's page, let every stage
Advance Australia Fair
In joyful strains then let us sing
Advance Australia Fair
We are young and free
We have golden soil and wealth
Our house is garbage in the sea
Our land is full of natural gifts
Rich and beautiful beauty
On the history page, every step
Advance Australia Fair
Let's sing in a pleasant sorrow
Advance Australia Fair
All the Australians make us happy
Despite the widely enjoyed failings of machine translation, there remains a timeless yearning for universal translation and an expectation of what it may bring to the world. That is, purportedly, world peace. Linguists have created entire languages reaching for this goal: an interlingua, or universal language for humanity. The most famous and time-tested remains Esperanto, a supposedly simple language invented specifically to become a vehicle for uniting the globe. Esperantists remain passionate about this project, maintaining a community since the language’s 1887 conception. The excitement around universal translation, from ancient texts and Victorian treatises on invented languages to today’s buzz around translation machines, often touches on the potential benefits for world togetherness, co-operation and globalisation, without examining what this would look like. In 1905, Nikola Tesla argued that while “[m]utual understanding would be immensely facilitated by the use of one universal tongue”, an invented language would never be accepted, “however time-saving it may be”.
“That would be contrary to human nature,” he wrote. "Languages have grown into our hearts.”
A language cannot be sustained if it is literally devalued, if its speakers can’t earn and survive in a society that uplifts English to the detriment of all other languages.
• Photo by
Tesla touches on something beautiful and unsettled in the debate around language: how much does language shape us? Are our hearts and minds subject to language, or vice versa? Languages, particularly those of First Nations people in colonised countries like Australia, are still being lost – about one every fortnight. Some estimates show that 90 percent of languages currently in use may be extinct by the end of the century. Census data shows that the earning potential and employment prospects for culturally and linguistically diverse people are well below those of English speakers in Australia and other Western, English-dominant countries. A language cannot be sustained if it is literally devalued, if its speakers can’t earn and survive in a society that uplifts English to the detriment of all other languages.
The most radical possibility of universal translation is that it could preserve vanishing languages. While the anecdotes in this story reflect the perspective from which translation devices are often sold – that of people with the privilege to travel –language barriers have an enormous financial and social impact on the day-to-day lives of many. The cultural – and financial – capital of English and other major languages, coupled with a pristine, middle-class accent, put other ways of communicating at risk. But perhaps seamless universal translation in workplaces and schools could strip away the obstacles faced by non-dominant language speakers. Rather than speed up globalisation, universal machine translators might enable human speakers to preserve language diversity and the uniqueness of particular languages.
While walking back to my hostel in Bolivia, a woman in a uniform and hairnet ran out of the restaurant to my left and threw up in the street. She was alone and seemed distressed. Not feeling it possible to walk away, I approached her as carefully as I could, trying not to be intrusive. I could not gauge how ill she was in that moment, and she was alone. I approached her and said, in my limited, careful Spanish, “Excuse me, are you okay? Do you need help? Necesita ayuda?”
“No, thank you,” she said. “Estoy bien.”
“I’m there,” I said, pointing to the porch of my hostel, “if you need help.”
That moment has stayed with me because, as with many moments after I have forced myself to approach a stranger, I still ask myself, did I do the right thing? Without an alien fish or Google device in my ear, I was left confounded; she and I came down from opposite sides of the Tower of Babel. I had to make do with what measly Spanish I’d picked up along the way. But, right or wrong, I could not have approached her wielding a smartphone. If she had needed something, how could I make her ask it of a screen?
There may come a time where machine translation is rapid, streamlined and perfect. That world seems to be fast approaching. We have come to expect technology to not only mimic or assist us, but to surpass us altogether. But it is hard to imagine the normalisation of machine translation, especially outside the wealthiest nations. Machines are still no substitute for human sensitivity: the ability to modulate your voice in line with your expression, which kicks in when translation is needed most.
