Hi folks!

As Summer addressed the machine translation issue in her recent post, I couldn’t resist adding a few titbits from my “blunders and howlers” collection. Because one thing that keeps me on my toes most of the time I review machine-translated texts is how common lexical phrases or grammatical constructions are mistranslated. Take a look at this sentence, for example:

She made your aunt promise that she would…

Straightforward enough, you might say. But it became

Sie hat deiner Tante versprochen, dass wir uns …

which means:

She promised your aunt …

Similarly:

He was up again within seconds to throw up into a bush to the left, hucking up all the turkey he’d eaten earlier.

Er war innerhalb von Sekunden wieder auf den Beinen und warf sich in einen Busch auf der linken Seite, wobei er den ganzen Truthahn, den er zuvor gegessen hatte, in sich hineinschlang.

In the German sentence, our hero is not “throwing up” into a bush; he is “throwing himself into a bush”. This action seems to have made him hungry: he is gorging himself (!) on some more turkey rather than spitting it out. You can safely say that this would make for some very unusual reading …

Okay, so dialectal variants and sublanguages are perhaps not every machine’s forte. Still, it’s amazing how often you will find very simple constructions mangled.

To top off today’s selection, here’s my current favourite:  

He spied a young kid chewing on a piece of bread, leaning on a shovel.
Er entdeckte einen jungen Mann, der auf einer Schaufel kaute und sich auf ein Stück Brot stützte.

The translation says: „He spied a young man chewing a shovel and leaning on a piece of bread.” Cheers to that!

A few reasons why AI and translation software will NEVER top human-produced work
by Summer Worsley

With advances in modern technology, it's no surprise that machine translation has made its way into our working lives. There is now an app for your phone, a program to run on your laptop, and even an earpiece device that translates foreign languages in real-time.
 
The availability of machine translation technologies has changed the landscape of the translation industry. Although it's a great asset to be able to speak with someone from across the world in a different language (albeit with the grammar skills of a three-year-old), there are many reasons why you would want to use human translators instead of software.
 
Human translators understand context 

Context is the surrounding circumstances of a word or phrase, and it's critical to understanding what you're saying exactly, be it in text or speech. For example, if I say "I love ice cream," the three-word sentence can have different meanings depending on who I'm talking to and why.
 
If I say this to someone else who also loves ice cream and has just expressed a similar sentiment, then there's no confusion. But what if I say this to my partner while we're fighting over bills? She will likely interpret my words differently than if we were talking about dessert together.
 
In this case, context dictates that being left alone with a tub of Ben & Jerry's can be seen as anything but sweet.
 
If you've ever tried using machine translation software like Google Translate or Microsoft Translator (and let's be real: how many of us haven't?), then you've seen first-hand how context isn't taken into consideration when computers process language data into another language form.
 
In fact, it can often result in nonsensical translations due to a lack of proper context, like that time the Australian Defense force tweeted utter nonsense in Arabic. Or when Clairol, a company that should have known better, ended up with its ‘Mist Stick’ curling iron promoted as a ‘Manure Stick’ in Germany.
 
Machine translation can't yet respond to literary and figurative language 

Machine translation is generally only good at taking a specific word, searching its database of translations, and returning the corresponding result. This means that it cannot understand context or connotation, nor literary or figurative language.
 
In part, this is because machines do not process language in the same way humans do—they don’t have an innate understanding of meaning or nuance. For example:
 
“The sky was yellow with birdsong all day long; it was impossible to sleep.” A human reader understands that “yellow” doesn't mean colour here but rather describes an overwhelming abundance of something. But a machine would simply return a literal “yellow with birdsong all day long; it was impossible to sleep” as its final product without any explanation for what exactly makes this sentence so strange and beautiful at once.
 
If we look closer at our previous example, we can see where the problem really begins: the sentence contains idioms (a phrase with a meaning that cannot be determined from its individual words). Idioms depend on context and convey meaning not through literal definitions but through common usage by people within a specific culture over time, which machines simply don't have access to.
 
Literary translation is not just a literal word-for-word process. The translator is required to interpret, adapt, and transform the literary work in order to make it culturally accessible to the readers of their target language.
 
For example, if you’re translating a novel from English into German, your first step is to understand what kind of story you’re dealing with.
 
Is it fiction or nonfiction?
What kind of language use does the author use?
Does he/she use slang words or highfalutin expressions?
What are his/her cultural references?
 
Once you've answered these questions, then you can begin translating segments at a time without making any major mistakes that would affect the integrity of your translation effort overall. A machine cannot do this, nor ask the questions needed to do a good job in the first place.
 
Mistranslated documents carry the risk of legal liability 

Mistranslations can be dangerous. Mistakes in medical documents, for instance, could result in incorrect treatments, putting patients at risk of serious complications.
 
Mistranslated contracts could lead businesses to lose money or to inadvertently fall foul of regulatory requirements. And some mistranslations may not be immediately obvious—they might not even be noticed until they've had an impact on large numbers of people over a long period of time. And the more people the mistranslation affects, the likelier the chance of a stratospheric fee.
 
Mistakes carry risks for individuals as well as companies: if you simply want your message understood by its intended audience, you could end up looking silly. Like when ‘Got Milk?’ was translated as ‘Are You Lactacting’.
 
But if you're hoping for legal protection, then it's essential that everything is translated correctly so that all parties involved know exactly what is being said in written communications.
 
As translation technology improves, human translators will still be needed 

At a time when AI is improving so rapidly, it’s tempting to wonder whether we’ll need human translators in the future, but while AI can help with some things, it cannot translate anywhere near as well as a human expert, nor will it ever be able to.
 
A machine can’t tell if your text is meant to be funny or sarcastic, or if it's written in a formal tone or a casual one. And even if an algorithm was able to make those delicate distinctions, it would still have trouble writing for humans because it lacks our ability for nuance and creativity.
 
As AI gets better at translating, it will become a more attractive option for companies who want something fast and cheap. But savvy businesses know that words drive profit and growth and that words are only effective when wielded well; when they tap into the consumer’s psyche and cause them to take action.
 
If you need help with your words, either from English to German or vice versa, please get in touch.

A New Study Warns We Might Lose 1,500 Endangered Languages
by Summer Worsley

A groundbreaking study led by The Australian National University (ANU) warns that we could lose 1,500 endangered languages by the end of this century.
 
Lindell Bromham, an ANU Professor and co-author of the paper, noted that around half of the world’s 7,000 recognised languages are currently endangered and warned that without “immediate intervention” language loss could triple within the next 40 years.
 
Published in Nature Ecology and Evolution, the study focuses on the predictors that put endangered and vulnerable languages at higher risk.
 
A total of 6,511 languages, more than 90% of the languages spoken worldwide, were analysed with 51 predictor variables that included aspects such as the number of speakers (population), documentation of the language, legal recognition, landscape and climate (both associated with language diversity), and socioeconomic factors among others.
 
The study found that contact with other languages “per se is not a driver of language loss”, a finding that runs counter to commonly held beliefs.
 
Instead, the paper notes several other drivers including a few surprises such as greater road density, which encourages population mobility and movement. The study also found evidence that higher levels of education can negatively affect language diversity.
 
Professor Bromham noted that “Contact with other local languages is not the problem – in fact languages in contact with many other Indigenous languages tend to be less endangered.” Whereas with greater road density, “it’s as if roads are helping dominant languages ‘steam roll’ over other smaller languages.”
 
It’s telling that this major study comes from Australia, a nation with one of the world’s highest rates of language loss. The nation is home to more than 250 Indigenous languages, but according to a 2018-2019 government report, only 123 Aboriginal and Torres Strait Islander languages are still used.
 
Meanwhile, the use of foreign languages spoken by expatriates in Australia is also falling. For example, census data shows that around 354,000 Australians spoke Italian at home in 2001, but by 2016, that figure had dropped to around 272,000.
 
As Ingrid Piller, professor of applied linguistics at Macquarie University puts it: "Australia is in many ways a graveyard of languages."
 
Of course, language loss is not only an Australian problem; the falling level of language diversity is a global issue. In 2011, for instance, the Guardian reported that the last two remaining speakers of Ayapaneco, a language spoken in Mexico, were not speaking to one another. Then aged 75 and 69, the two speakers both worked with linguists to help produce a dictionary of the tongue they called Nuumte Oote ‘the true voice.’
 
Codifying threatened languages is one way to help prevent their demise, as the ANU study notes: “To avoid the loss of over 1,500 languages by the end of the century, urgent investment is needed in language documentation, bilingual education programmes and other community-based programmes.”
 
UNESCO runs an interactive map of the world’s most in-danger languages, which you can find here. The online version is complementary to the Atlas, a print version detailing the same information. Both also cover revitalised languages, those that have been brought back from the brink of extinction.
 
At-risk languages worldwide 

Currently, UNESCO lists 577 languages as critically endangered, a designation that comes just one categorisation behind extinction. Of these, 196 languages had just 1 to 50 speakers and 116 had 1 to 5 speakers. Here are just a few examples of critically endangered languages from around the world:
 
Dampelas 

Also known as Dian, Dampal, and Dampelasa, this language is spoken in several villages in the western coastal area of Sulawesi’s central province in Indonesia. According to the Endangered Languages Project, “Dampelas speaking parents tend to speak Indonesian to their children in hopes of improving their opportunities for the future.
 
Luri 

Not to be confused with the Iranian language of the same name, Luri, also known as Lúr and Lur, is a West Chadic language that was spoken by just two remaining people in Nigeria, according to information gathered in 2002. Of the region’s dominant languages, Hausa and Langas, the latter is also considered endangered with just a few hundred speakers remaining.
 
Oneida 

A native American language of the Iroquoian family, Oneida is now spoken by just a handful of people, with estimates ranging as low as 47 and as high as 500. According to the University of Wisconsin, the language “has only been written down systematically in the last couple of generations.”
 
 
Why language diversity matters 

Linguistic diversity matters for both ethical and aesthetic reasons. Because language is so closely tied to culture, the loss of a language means the loss of elements of some people’s identities. Unfortunately, this often comes about because of policies that marginalise both the language and its speakers, such as mandated schooling in a dominant language.
 
Plus, diversity, in and of itself, is a thing of beauty. Every language is special and unique, and each carries its own cultural knowledge. Losing these means losing a little piece of our humanity.
 
No matter how prominent or ‘minor’, every language can express infinite ideas and concepts, and every language holds generations' worth of knowledge that has been built and shaped by its speakers.
 
The world is a richer place with linguistic diversity.
​