Lost in translation: How Hausa, other languages lose meaning on Facebook

Since its inception, Facebook has provided a platform for people around the world to socialise, exchange ideas, as well as other activities.   This made it a platform of choice for many people with more number of users compared to similar social media platforms.   Its large followership transcends borders, languages and other socio-cultural differences […]

Since its inception, Facebook has provided a platform for people around the world to socialise, exchange ideas, as well as other activities.  

This made it a platform of choice for many people with more number of users compared to similar social media platforms.  

Its large followership transcends borders, languages and other socio-cultural differences existing in the world by providing a link through which people and nations seen as a valuable development.  

The attempt by Facebook founder, Mark Zuckerberg, to provide a translation of text on the social media platform from the original language to another language, or vice versa, is seen as commendable.  

The service utilises artificial intelligence costs a lot to the company, while it satisfies the needs of its followers.  

However, while they made efforts to translate posts for millions of people around the world, in over 100 languages spoken by different people and understood by many, meanings are being completely lost.  

Some of the translations suffer from contextual deficiencies as words and phrases are translated directly, leaving out the context in which they are used.

As the result, readers, especially those learning the languages, get wrong meanings and native speakers find it difficult to comprehend how their languages are being bastardised.  

They are being subjected to do an extra effort of explaining what was posted and what the translation really means in their languages, so as to educate others.  

Some of the translations appear ridiculous as they convey entirely different meanings of words, sentences and phrases translated from their languages to others, or from other languages to theirs.  

Angela Fan, a research assistant with Facebook, while writing on artificial intelligence as being used to translate languages, stated that researchers had been working for years to build a single universal model that could understand all languages across different tasks. 

“A single model that supports all languages, dialects and modalities will help us better serve more people, keep translations up to date and create new experiences for billions of people equally. This work brings us closer to this goal. 

“As part of this effort, we have seen incredibly fast-paced progress in pre-trained language models, fine-tuning and self-supervision techniques,” she stated.

She, however, stated that the machine translation system abilities were limited due to many factors when it comes to translating many languages.

She wrote, “Typical machine translation systems require building separate artificial intelligence models for each language and task. But this approach doesn’t scale effectively on Facebook, where people post content in more than 160 languages.

“Advanced multilingual systems can process multiple languages at once, but compromise accuracy by relying on English data to bridge the gap between the source and target languages.”

The limitation of the machine translation using artificial intelligence as hinted by the Facebook research assistant may be the cause of loss in the meaning of Hausa and other languages in words or sentences as a result of complexities.  

There are millions of mistranslations of words in Hausa-English or English-Hausa translations and other languages on Facebook.

These mistranslations are being reeled out every day with impressive accuracies in some instances and outright off point in many others.   

Unlike humans, the technology deployed may make attempts to provide a direct translation of words, which may be a right or wrong translation in the language.  

Second speakers of Hausa and other languages, as well as others who want to get meanings from the languages, are most times misled as to what a word or a phrase means if it was directly translated.  

A careful look at some of these translations exposed the need for a more rigorous human intervention, especially by languages and linguistic experts, which Facebook needs to deploy to streamline the use of artificial intelligence and human expertise to provide a near perfect translation of words and phrases across languages.  

Although almost all other languages, apart from English, French and some few European languages, suffer the same fate as Hausa, a language spoken by millions across Africa, the way its meanings are distorted might be second to none.

An instance of such translation, which was completely off point, was when Facebook translated “miyar kuka” as “soup of crying”.

Miyar kuka is a popular soup in Hausa land, which is prepared using dried leaves from the baobab tree (kuka).

‘Miya’ is the Hausa word for soup, and when making a sentence or a compound word to tell what is used to prepare it, the possessive marker ‘r’ is added to appear as ‘miyar’

So, in this instance, while ‘miyar’ was translated well to mean soup, the word ‘kuka’ (the dried baobab leaves) was translated as ‘crying,’ which is also written as ‘kuka’ in Hausa, leading to the wrong translation of one word for another.

Another example was a news headline posted by ‘Idon Mikiya,’ which reads, “Aikin shataletalen da aka bayyana ya lakume zunzurutun kudi har kimanin naira miliyan 16 a kasar Tchad.”  

It was wrongly translated as, “The Project of Shataletalen that was described as the one who stole a total of 16million in Chad.”

This should have been a straightforward headline speaking about a roundabout (shataletale) project which gulped N16 million in Chad.

Another BBC Hausa post about the accusations and counteraccusation on what led to the return of the Taliban, which reads, “Cacar baki ta kaure ne akan batun wanda yafi taka rawa wajen samun nasarar dawo da gwamnatin Taliban da yin nasara akan Amurka kuma raba mukamai.”

This was translated as, “Gambler’s mouth is a major contributor to the Taliban’s return and victory over the US and divisions”.

Here, we can see that “cacar baki,” which should have been translated as “disagreements or quarrel,” was translated as “gambler’s mouth.”

The reason for this is not farfetched as the two words were translated separately with their Hausa meaning. 

Gambling is called ‘caca’ while mouth is called ‘baki’ in Hausa. So Facebook translated ‘cacar baki’ as ‘gambler’s mouth.”

Another example of wrong translation was another BBC Hausa news headline, which reads, “Daular Larabawa Tana Neman Wasu Yan Najeriya Ruwa A Jallo”.

This means that the “United Arab Emirate (UAE) has declared some Nigerians wanted.” However, Facebook translated it as “The Arab emperor is looking for some Nigerians in Jallo.”

A look at how the mistranslation comes about can clearly be deduced.

The United Arab Emirate was translated as “The Arab Emperor”, “Ruwa a jallo,” which figuratively shows the seriousness the men were wanted, which shouldn’t have even appeared in the translated sentence in English because it would have no meaning at all.

Another example was where Facebook translated a headline by ‘Alfijir Hausa’ about a girl that was mistakenly buried alive in Katsina State.

It reads, “Anyi Kuskuren Binne Wata Yarinya Da Ranta A Katsina.”

This headline should have read, “Girl Mistakenly Buried Alive in Katsina,” but Facebook translated it as “There is a mistake of following a girl with her swearing in Katsina.”

Speaking on the issue, a Professor of Media and Cultural Communication and former vice chancellor of the National Open University of Nigeria (NOUN), Professor Abdallah Uba Adamu, said the way Facebook translates Hausa was not anything near what it should be.

“What you said is true. It is almost always the case. That is why a genuine translation needs humans, not machines. Facebook’s use of artificial intelligence in translation has not been perfected to function as it should be,” he said.

Yakubu Sarauniya, a Hausa native speaker said he gets amused whenever he sees a direct translation of Hausa phrase that conveys a different and sometimes ridiculous meaning.

“I know it happens to other languages too, but for Hausa, which is my language, I always notice the mistakes. Sometimes they get it perfectly well, but they fail whenever they translate a complex word or phrase,” he said.

Ibrahim Isa, a BBC Hausa journalists, whose work involves, among other things, translating and voicing over reports in English for his Hausa listeners, said the way Facebbok translates Hausa to English or English to Hausa portrays the fallacy of not using more of expertise but relying on the services of people who only have their way without a good grasp of Hausa language.

“Facebook is doing something that will have a long term impact on the language. The danger is that the language is being translated in a wrong way and is being seen by the world.  

“If this continued the ways it is, there will be a time the people reading the messages over and over again will believe and use the wrong words, thinking they are using the right words.  

“It is just like what is being done by advertising agencies in Nigeria on Hausa adverts. You always see Hausa being spoken to advertise products in a very wrong way. 

“It is all over on radio, billboards and television. These people use persons who are not native Hausa speakers to advertise in Hausa.

“They don’t consult experts on the language, or use native speakers to do the job, for reasons best known to them,” he said.

While explaining the problems being encountered, Fan, however, admitted that their artificial intelligence was contending with many obstacles as they attempt their language-to-language translation, which involves translation of millions of words and phrases.

“One of the biggest hurdles of building a many-to-many model is curating large volumes of quality sentence pairs (parallel sentences) for arbitrary translation directions not involving English. 

“It is a lot easier to find translations for Chinese to English and English to French, than, say, French to Chinese. What’s more, the volume of data required for training grows quadratically with the number of languages we support,” he added.

As it appears, unless Facebook perfects its artificial intelligence language translation mechanisms, the service of professional translators, linguists, as well as native speakers of Hausa and other world languages, meaning will continue to be lost in the course of translations.