Conversations with language

Translation is an act of hope. The translator tries to carry across not just words, but the rich and varied worlds they depict, steadfast in their resolution that little must slip through their grasp. They go beyond the mechanical; translating with the necessary artistic and creative fervour to illustrate a world only glimpsed through the words on a page. What’s lost is nowhere near as important as the new perspective on reality that is recovered.

However, what happens when AI takes over? A recent Oxford Academic report on the use of machine learning in translating one of the world’s oldest languages, Akkadian, from its cuneiform clay-tablets into modern English, posed such a question.

Heralding a significant step in translating the hundreds of thousands of texts which survive from the Ancient Mesopotamian world, this machine translation may greatly illuminate the literature, learnings, and lives of people who thrived in the Babylonian golden cradle of humanity. Nonetheless, Trinity College’s Al Maktoum Associate Professor Dr Martin Worthington stresses the need to be cautious about artificial intelligence’s apparent breakthroughs. “It wasn’t the computer who looked at the tablet and decided which signs are on the tablet,” he explains: ‘that was done by a human. Which in itself is very revealing, because one of the hardest things with cuneiform is deciding what the signs are on the tablet. And that’s the step that they skipped.”

Upon inspection, these human-aided translations were also not entirely promising. Worthington stated that while impressed by the capabilities of the computer, he found some of the results “absolutely dreadful.”

Artificial intelligence is lauded for its ability to reproduce human-sounding dialogue, but this dialogue is more of a bad imitation than a lifelike depiction of human thought. The machine doesn’t know or understand the language as well as a scholar or even a student. Rather, they produce a probabilistic model of what English form of words is most likely to match the Babylonian form of words. While according to Worthington “you could argue that humans also have some kind of a probabilistic model […] so far humans seem to do it a lot better.”

“And it’s a bit like ChatGPT,” Worthington continues: “when you look at it under a microscope, the results are less impressive.” Intimately dissecting the output of ChatGPT leaves much to be desired—subtle patterning and “waffling” sentences reign supreme. It’s no wonder it can be detected so accurately on college assessments. Here, the mind in all its irrationality and hectic interconnectivity is unmistakably absent.

Worthington is cognisant of where the times are taking us. A translation model capable of standing its ground amongst the work of academics will certainly enhance the accessibility of these ancient texts, but poses danger not only to the academics but to the field itself. Translating the surviving documents of the ancient languages isn’t a lone act, but rather an intimately interpersonal one founded on a shared body of knowledge. “You have all these different human experiences which converge towards something like a consensus. And it’s a well informed and richly generated consensus. Whereas if we put all of this into the arms of one so-called intelligence, then the decisions are going to be made no longer by committee and by consensus, but by a central diktat.”

However, when the human isn’t lost in the cogs of the machine, AI-based translation presents itself as an effective tool—in particular when the translator is faced with a corpus of surviving documents so immense as that as Akkadian, or, likewise, Old English from over a thousand years ago. “The great opportunity of machine learning is scale,” says palaeographer and medievalist Dr Mark Faulkner, who is currently involved in building a new open-access corpus of Old English using machine learning which has produced translations with 98% accuracy. “And I don’t think that disqualifies the human being part of much more detailed work as well.” Nowhere is the human touch absent, it seems.

What role do we play? One that’s not without risk, Faulkner emphasises, discussing the importance of being careful about any assumptions that our ancestors understood languages exactly as we do: “How did contemporary speakers of Old English or any ancient language perceive their language, what was their linguistic self-awareness of language, and how did that inform the way that they spoke and wrote?” Ignoring these questions leaves the translator blinded, for example, to the brilliant moment of linguistic invention where King Alfred actually calls the English language ‘Ænglisc’ for the first time.

Ultimately, translation is an act of communication, and it’s important to consider that it is an ‘act of humility’, as Faulkner relates, to acknowledge that we are shaped by what’s gone before us—and conveying what’s gone before us as it was, not how we wish it to be, is of vital importance. Now, that’s all well and good, but how can the translator do this? Translation is, inevitably, a task rooted in failure. Theorist Ortega y Gasset called translation a ‘utopian task’, a logical impossibility, whilst Lawrence Venuti argued there is never any hope of achieving a literal one-to-one translation. So-called perfect translations don’t exist. Instead, translation is “a game of losses and gains and what you want to lose for the sake of gaining what”, as Worthington notes.

Vladimir Nabokov found a solution for this, writing in an essay on his translation of Pushkin’s Onegin that his version will either appear with a thousand footnotes, or not at all. It’s difficult to know if he was being sincere or sarcastic—the latter seems most plausible. In truth, no reader wants to be faced with a single word and a daunting cascade of footnotes, nor on the other hand to be left without a note when it’s needed. Both extremes betray an unwillingness to compromise for the reader. Instead, the humble translator is the best translator, conscious of their limitations while still aiming to impart comprehension.

When AI takes the back seat, translations can shine, astounding and invigorating the reader. Admittedly, Akkadian administrative documents dominate the surviving corpus, and the language received little attention in academia—until tablets depicting the Epic of Gilgamesh were discovered amongst the proverbial rubbish-heap. One tablet told the story of a Great Flood—dated over a thousand years before its Book of Genesis counterpart. Upon translating these words, George Smith is reported to have said “I am the first man to read that after more than two thousand years of oblivion.” The discovery garnered worldwide fame and attention for the Gilgamesh, one of the oldest known pieces of literature and a catalyst for changing how we see the Abrahamic religions. “The understanding of the Hebrew Bible and its intellect has been revolutionised by Mesopotamian sources,” Worthington explains. “It’s now become apparent that a lot of important parts of the Hebrew Bible seem to be responses to the Mesopotamian worldview”—a worldview which proves revelatory when the translator takes their individualised plunge into the language.

In conversation with Dr Worthington, I ask him if his work has offered him an alternative lens on how he thinks about the world: “When you study ancient societies, you’re always confronted with the power and dominance of the elites. […] I think anybody who is confronted with these inequalities day in, day out in their professional life is bound to have a heightened awareness of inequalities in the present world.” 

Documents from Ancient Mesopotamia can poignantly reveal the underlying, uneasy structures on which our societies depend. “One of the things that people talk a lot about in the ancient world is administration—as a means of social control, organisation, inclusion, and assignation of resources and entitlements,” Dr Worthington recalls, “and you always get this sense of something like a clockwork machine governing the lives of ordinary people, like a sort of sinister unseen force ticking on in the background that affects different people to different extents. My frequent meetings with the works of ancient administrators remind me with a shudder that, if I look down the street and I see a thousand people, there’s probably some of them for whom the machine isn’t doing what it should be doing.”

In the long view of history granted to keen scholars of these texts and their ancient languages, neither is it a surprise that wars and conflicts remain with us even in the new millennium. Recent events in many places show that stable, peaceful societies ruled by democracies are not guaranteed: they depend, among other things, on a social consensus which supports and values them. “The ancient world where human life was not particularly precious is a stark reminder that for thousands of years, it was perfectly possible for successful societies to organise ourselves in a very different way. And if we want to use our way, we have to argue for it, we have to protect it.”