Engineers at Facebook have announced a breakthrough in machine translation: a hack for translating spoken Hokkien, a dialect of Chinese with no written script, into English in real time. Oral tongues are more slippery than written text, and even the tiniest step towards translating one is a giant leap towards the Babelfish—the “small, yellow and leech-like” creature that is slipped into the ear in The Hitchhiker’s Guide to the Galaxy for instant comprehension across languages and cultures.
Facebook has open-sourced its work in the belief that other researchers will help take it forward. Altruism has nothing to do with it. Instant, bi-directional oral translation would be the killer feature of the Metaverse, since the others are, individually, old hat. Mark Zuckerberg is selling his virtual space as the future, and has rebranded his enterprise to reflect the name. But to old-timers on the net, it looks like bits and pieces of past worlds cobbled together, shined up and made commerce-ready.
In 1992, Neal Stephenson’s novel Snow Crash suggested that the grey internet of the time, written in flat HTML—typically viewed in the Mosaic browser, or from a UNIX command line—would one day give way to a visually rich metaverse in which avatars interact with bots in a digital geography featuring gatherings like in the real world. The digital gathering is even older than the visual, browser-based internet, and dates back to ‘Multi-User Dungeons’ (a nod to Dungeons and Dragons), role-playing games which ran on command line clients. These were worlds built with words, like fantasy literature, in which bots hosted meets for people dialling in from the world over. The most hospitable bots were bartenders serving virtual drinks. Scarily, they doubled as bouncers.
Fast-forward to Linden Lab’s Second Life, a world built by its inhabitants two decades ago. Its roots lie in role-playing games and like them, it was conceived as an escape mechanism for people tired of the limitations of brick and mortar. But it also ventured into virtual commerce with its own currency, the Linden dollar, which was used by entrepreneurs in Second Life to sell each other digital artefacts like clothes, accessories and homes. Second Life was hot because it had ventured into commerce, but interest flagged because the products were purely digital. You could only buy a digital shirt there, not an Arrow shirt.
Now, Mark Zuckerberg’s Metaverse is adding the secret sauce of mass market commerce. Using virtual and augmented reality technology developed over the last decade, and taking advantage of ever-cheaper storage and bandwidth, he is promoting a world which is exactly the opposite of earlier role-playing realities, including Second Life. His Metaverse is not a portal to escape into an alternate reality. It is a world where we do digitally the very things we do physically, including meeting in a boardroom, watching a movie with friends, or going to a store with the family to try out and buy a real shirt, while erasing distance. Under any impetus which disincentivises urbanisation—pollution, climate change, contagious diseases, the soaring cost of urban living—Metaverse will boom as an alternative place in which to transact our lives.
However, the scope of virtual communities is limited by the language of communication. The internet, which was distinct enough to count as a community in the 20th century (that explains the now-quaint term ‘netizen’) was originally American, because the phone networks on which it ran were almost exclusively American. Europe, which is almost as multilingual as South Asia, emerged as a distinct language space much later, followed by China, the nations speaking the Middle Eastern languages, and our own region. Though the internet has exploded, such speech bubbles remain, and Google Translate has its delightful absurdities. The mode of communication is another crucial barrier: you type on
the classical internet, while you speak and use body language in a metaverse. AI-based translating machines are trained by feeding them scads of text, original and translated. The quality of their work is gauged by comparing their output with that of human translators. But in the case of an oral language like Hokkien, there are no texts to compare.
Facebook’s engineers have used a cheat, first converting oral Hokkien to Mandarin script. A similar strategy could be used on Indo-European languages, which can be represented in Roman script with diacritics. But it’s still a cheat which will not make the dream of science fiction, the instant translation of alien tongues, come true. If you meet an Aldebaran chicken on a dark and stormy night and it says, “Take me to your leader”, you won’t understand a thing, even with your Babelfish in your ear.
Facebook has a rough history with AI. A few years ago, the press had sensationally reported that its experimental bots, which were learning to bargain online, had gone rogue and developed a secret language, and had to be shut down for fear of a machine apocalypse. The reality is less scary: the bots were speaking an English pidgin. AI uses incentives to nudge in the direction of desired behaviour, and FB’s programmers had neglected to incentivise the use of formal English. The bots were free to develop a dialect more comfortable for machines, and humans thought it was bizarre.
But whether it’s about communicating with pidgin-speaking bots or each other in a multiverse, humans may be better off not understanding everything. Grey areas keep the peace. In this, The Hitchhiker’s Guide to the Galaxy is reliable and cautionary: “Meanwhile, the poor Babelfish, by effectively removing all barriers to communication between different races and cultures, has caused more and bloodier wars than anything else in the history of creation.” As the poet sang in Four Quartets, “Humankind cannot bear very much reality.”
Pratik Kanjilal
Editor of The India Cable
(Tweets @pratik_k)