in All Things Localization, Language Technology

2016 was the year of Neural Machine Translation (NMT). After Facebook’s announcement in June 2016 and Google’s in September 2016, that the giants’ translation platforms were being powered by Neural Machine Translation, the tech and translation worlds went wild speculating on the future of cross-cultural communication and the relevance of human-powered translation in the coming years. (We have to admit, we did as well.)

But a year on from Facebook’s announcement, NMT hasn’t fully blossomed into the language-demystifying super-machine it was ordained to be.

Google continues to announce every few weeks or so that they’ve added another language pair to their NMT-compatible list. However, according to Facebook Engineering Manager Necip Fazil Ayan, as of April 2017, Facebook is still only halfway to their goal of powering all translations across Instagram, Facebook, and Workplace by NMT.

What’s the holdup? The fickle human mind, that’s what’s up.

Pop culture references have always evaded dictionaries for some time after their initial adoption, and now due to the voracity and speed at which the internet consumes and discards new slang and references, machines can’t quite keep up. Ayan has pointed to “odd spellings, hashtags, urban slang, dialects, hybrid words, and emoticons” as being the major hurdles for Neural Machine Translation.

And here we digress. It’s emoji time.

A long history, but still much confusion

Emoji were invented in Japan in 1997/98 by Shigetaku Kurita, who was working at cellphone provider NTT Docomo. They were widely-used in Japan for nearly ten years before crossing the Pacific to the West, but once they arrived, they took over texting by storm.

Related:  How LSPs Can Keep on Truckin’ with Green Tech

As their popularity has spread worldwide, linguists and laymen alike have questioned their impact on written language, and whether they even constitute a new language entirely.

A study by an Emory University undergrad has shown that not only were participants unable translate sentences to and from strings of emoji with any level of consistency, they could not even agree on the meaning behind a single emoji. Individual participants would even attach multiple meanings to a single emoji or change their concept of the emoji over the course of the study. These inconsistencies made it clear that these ideograms are nowhere near capable of precisely communicating what a real language can.

With this debate out of the way, what are emoji useful for? Conveying emotion, primarily.

Like the first smiley faces created by punctuation marks, emoji were invented to convey information faster and to give cold, black-and-white texts a human component.  Linguists have found that “[of] the 20 most frequently used emoji, nearly all are hearts, smilies, or hand gestures—the ones that emote”.

So, when character limits and tired thumbs constrain us to only sending tiny snippets of text, emoji can help convey a heart’s worth of prose in just the width of a letter.

How to handle emoji with Neural Machine Translation (NMT)

It’s clear that emoji aren’t a language, as it’s impossible to pin down their exact meanings. This fact has made them nearly impossible to translate, and is why Facebook’s NMT has chosen to simply ignore them and those other pesky language inputs like hashtags and slang.

So, when it comes to the original question of whether or not machines can translate emojis, the answer is a resounding “no”—but, neither can humans. Perhaps, this is why it’s best to leave translating Facebook posts up to Facebook and the serious documents up to the professionals.

[dqr_code size="120" bgcolor="#fff"]