Dangers of translation ever present in the digital era

From literal translations that miss the mark entirely, to clever interpretations that create entirely new meanings, the danger of miscommunication is always lurking

From the early days of the Bible to our modern technological age, translation has often stirred controversy — even resulting in executions
Jamie Wignall
From the early days of the Bible to our modern technological age, translation has often stirred controversy — even resulting in executions

Dangers of translation ever present in the digital era

Sacred texts have played a significant role in the emergence and evolution of the translation field, paving the way for its establishment as a distinguished academic field replete with techniques, principles and customs.

The translation of the Bible from Greek to Latin stands out as a prime example. Since those times, modern concepts such as literal, interpretive, and semantic translations, as well as “disloyalty” and “faithfulness,” have been circulating and gaining importance. What better context to apply the terms “disloyalty” and “faithfulness” than when conveying the word of God?

In certain cases — such as the execution of Englishman William Tyndale in 1536 for his translation of the Bible into English — the appropriate term was rather “heresy.” This vague term, which was responsible for the crucifixion of thousands, equated translation with blasphemy, making them one and the same “crime.”

A decade ago, researchers Kevin Windle and Anthony Pym highlighted in an article an incident that occurred around the same time as Tyndale's execution. The incident involved the execution of Frenchman Etienne Dolet in 1545. Dolet, who believed that those who merely translated texts word for word lacked true insight, was accused of “mistranslation” of Plato's dialogues.

In one instance, Dolet reworked a passage from Plato's book into French and added the words “anything at all” — a phrase not found in the original text — writing: “After death, you will no longer be anything at all.”

This alteration was deemed heretical by religious authorities of that time, as it was interpreted as a rejection of the afterlife, resulting in Dolet's execution.

Dark Ages live on in modernity

Let's take a 550-year leap forward. As the 20th century came to a close, a tragic event unfolded. In July 1991, Hitoshi Igarashi, a Japanese researcher and translator, was brutally stabbed to death in his university office. This followed a previous attempt on his life the year before, shortly after he announced his translation of ‘The Satanic Verses’ by Salman Rushdie.

Ettore Capriolo, an Italian translator, was also targeted and stabbed in July of the same year for the same reason. These incidents draw a clear link between the dark ages of Europe and the close-mindedness of the 20th century, which had nearly claimed the life of Salman Rushdie himself not long before.

AFP
In this file photo taken on February 17, 1989, Iranian women are seen holding banners which read 'Holly Koran' and 'Kill Salman Rushdie' during a demonstration against British writer Salman Rushdie in Tehran.

Rushdie's ‘offense’ was daring to interpret and imaginatively explore the sacred text, while Igarashi and Capriolo's offense was simply translating it. In many parts of the world, the translator is still seen as a heretic, even though they are an essential part of the process of delivering the book in the translated language without contributing to its original creation.

In many parts of the world, the translator is still seen as a heretic, even though they are an essential part of the process of delivering the book in the translated language without contributing to its original creation.

Emerging debate

This brings us to the debate at hand: machine translation v artificial intelligence (AI) translation. Is it better to have a surefire literal translation or opt for interpretation which comes with the possibility of compromising the integrity of the original work?

The choice between the two methods is especially critical when considering the translation of creative and poetic works because of their subjective nature.

Considering the abovementioned incidents, it seems that people would be wary of taking on such a high stakes job? Afterall, deeming something as 'compromising the integrity of the original work' is so subjective, and the punishment can be so severe (being killed) that wouldn't it be better to outsource the job to a machine?    

This question takes on greater significance whenever interpretation and imagination are required. A billboard, traffic sign, or instructions for a vacuum cleaner do not require interpretation, whereas a news report may require a degree of imagination and interpretation, depending on the recipient's culture and relevance to its content.

Read more: Al Majalla quizzes ChatGPT on the crises of the Arab world. Here is what we learned.

Similarly, translating a political article would require even more interpretation, but the lack of interpretation may not necessarily hinder reception, although it could deteriorate the quality.

Where interpretation is indispensable

When it comes to literature, specifically literary criticism, interpretation and imagination are indispensable elements of translation. The question is no longer whether to interpret the text, but rather the degree of freedom allowed in the process. Among all literary forms, poetry presents the greatest challenge in terms of translation and interpretation.

I will expand on this topic later on but, for now I will delve into machine translation and recount my recent experience using Google Translate and the much-discussed ChatGPT to translate the opening lines of Mahmoud Darwish's In Jerusalem (Don't apologise for what you did, 2004) from Arabic into English. According to Google Translate, the poem reads as follows:

In Jerusalem, I mean inside the old wall,

I walk from time to time without memory

aim me

Meanwhile, ChatGPT generated the following translation:

In Jerusalem, inside the old walls,

I wander from time to time

without a memory to guide me.

Regarding the human translation, I opted for the renowned Darwish translator, Fadi Joudeh, and his rendition reads as follows:

In Jerusalem, and I mean within the 

ancient walls,

I walk from one epoch to another

Without a memory

To guide me.

These are the three different versions I managed to extract.

Observations

Google Translate, which is the most popular machine translation service worldwide, has shown similarities to the translations generated by ChatGPT and human translation. It shares with the former the translation of "from time to time" and with Fadi Joudeh the translation of "I walk," while ChatGPT uses "I wander" instead.

Additionally, both ChatGPT and human translation used the phrase "to guide me," while Google Translate opted for "aim me," an incorrect translation and erroneous grammatical choice.

Although it is difficult to determine whether ChatGPT's use of "guide me" was influenced by Fadi Joudeh's choice, given that his text is publicly available online, ChatGPT's version does indicate a familiarity with him. When prompted, ChatGPT was able to provide a wealth of information about Joudeh's experience in translating Mahmoud Darwish's work.

Although it is difficult to determine whether ChatGPT's use of "guide me" was influenced by Fadi Joudeh's choice, given that his text is publicly available online, ChatGPT's version does indicate a familiarity with him. When prompted, ChatGPT was able to provide a wealth of information about Joudeh's experience in translating Mahmoud Darwish's work. 

This leads to the question of why Google Translate failed to produce a similarly nuanced translation. The likely explanation is that machine translation depends heavily on dictionary definitions and approved modern Arabic dictionaries, while AI takes into account contextual factors.

Despite minor mistakes or divergent interpretations, the real issue at hand is the very cadence of the poem. With Google Translate and ChatGPT, we are presented with words arranged linearly, attempting to uncover a linguistic match for the original text.

While Google may overlook the nuances of significance, the contextual intricacies of language, and the artistic elements that underlie it such as similes and metaphors, ChatGPT places greater emphasis on translating the meaning, ensuring that the phrase is structurally and semantically sound.

The poet's essence

However, the meaning conveyed by a poem goes further than structure and semantics and extends into the realm of rhythm (which is reflected in the musical aspects of classical Arabic and Western poetry through meter and rhyme) and what I refer to as the poet's voice or essence.

Read more: Writing a poem doesn't make one a poet

Fadi Joudeh's translation not only captures Mahmoud Darwish's text but also his soul, conveying the poet's essence in a far superior manner than any machine. 

On 8 March, coinciding with International Women's Day, Omani poet Aisha al-Saifi made history by becoming the first woman to be awarded the prestigious title of Princess of Poets on the renowned Arab TV show named after the late Egyptian poet Ahmed Shawqi from the 1930s.

Despite her adherence to the classical Arabic style, with its strict meter and rhyme scheme, al-Saifi's victory has triggered backlash from traditionalists who have resorted to cyberbullying over a section of one of her poems that has gone viral.

I utilised this particular section filled with 'nonsense' to test the accuracy of Arabic-to-English translation. Google Translate produced the following output:

A tender hand rubs the clouds

Never fly with two legs

Bahar on a hill squatting

The Arabic word "qot" was mistakenly translated as "never" by the programme due to the missing shaddah (emphasis mark). Similarly, the word "bahar" was initially considered a proper noun because of the absence of diacritics, specifically the dhammah (a sort of vowel mark) on the last letter.

Yet, once the shaddah and dhammah were added, the programme correctly translated "qot" as "cat" and "Bahar" as "sea."

A tender hand rubs the clouds

A cat that flies with two legs

A sea on a hill sits squatting

ChatGPT initially provided me with an unexpected response in Arabic, yet it failed to weave the words into a poetic text. It generated what translates as the following:

The first line-a clenched hand rubs the clouds, allowing the sun to shine through and brighten the earth.

The second line-a cat flies with two legs, an unfamiliar and astonishing image, much like the ones found in dreams.

The third line portrays a surprising image where a bug is sitting on a hill that appears to be a sea.

As I analysed ChatGPT's Arabic output, I realised that I had erred in instructing it to translate the English passage into Arabic instead of vice versa. Given that the text was in Arabic, the programme likely inferred from my use of the term "translation" that my intention was to seek an interpretation or explanation of the poem rather than a straightforward translation.

Upon rectifying the directives, ChatGPT generated the following translation:

A gentle hand rubs the clouds

A cat flies with two legs

A sea on a hill where a beetle sits

So why did ChatGPT use "bug" in its initial response and "beetle" in its second output?

Beetle and squat

It didn't take me long to realise that the programme struggled to distinguish between "beetle" and "squat." Despite ChatGPT's supposed superior intelligence, I found it odd that Google Translate was able to accurately translate the word while ChatGPT couldn't grasp its intended meaning.

The issue lies in the interpretation of the phrase. Google treated "the squatting sea" as two separate words and therefore translated them correctly, whereas ChatGPT utilised its imaginative and interpretive abilities. It processed that the poet could not have meant that the sea was squatting, so ChatGPT deduced that the intended word was "beetle," as it made more sense for a beetle to be perched on a hill.

The issue lies in the interpretation of the phrase. Google treated "the squatting sea" as two separate words and therefore translated them correctly, whereas ChatGPT utilised its interpretive abilities. It processed that the poet could not have meant that the sea was squatting, so it deduced that the intended word was "beetle," as it made more sense for a beetle to be perched on a hill. 

The similarity in structure and shape between the Arabic words aided ChatGPT to make this interpretation. But what about the sea in this context? The programme's logic and reasoning became apparent in the following statement: "The third line portrays a surprising image, where a bug is sitting on a hill that appears to be a sea."

Imagine being in a classroom with two siblings — Google and ChatGPT. The first sibling is a stickler for rules and relies on memorisation and repetition to get by, while the second, having failed his class more than once, enjoys playing with words and tends to interpret things in his own unique way, sometimes to the point of delusion.

The journey ahead is long and the destination unknown. The confusion we see in Google Translate and ChatGPT is not all that different from the challenges we face in the world of human translation.

From literal translations that miss the mark entirely, to clever interpretations that create entirely new meanings, the danger of miscommunication is always lurking.

font change

Related Articles