We’ve all been hearing about ChatGPT. The entire world is fascinated by this technological advancement that is actively changing life as we know it.
Among the controversial discussions that ChatGPT has raised is the future of some industries such as programming, translation, content writing, and other disciplines.
Can ChatGPT translate texts? Do AI translation tools provide high-quality translations?
In this article, we’re going to find out by exploring the benefits and disadvantages of using ChatGPT in translation so you can make a decision that best suits your business.
What Is ChatGPT?
ChatGPT is an artificial intelligence (AI) natural language processing chatbot that was released in November 2022. The neural networks are designed to mimic human interaction and are provided with machine learning technologies. Users ask a question and the chatbot responds, much like a human conversation.
The language model has surpassed other AI tools and has been greatly used in research, programming, content generation, and even translation. Additionally, students around the world have been using it to do research and write essays and research papers.
As translation experts, we decided to explore the impact of ChatGPT as an AI translation tool on the translation industry.
Many people are predicting that ChatGPT and similar AI tools will completely take over the translation industry. However, others insist that AI translation tools are simply a helpful hand to supercharge the work of translators as well as the efficiency of the whole translation process.
The question remains: to what extent can you integrate ChatGPT in translation? Well, there is no one, definitive, answer. But let’s dive in and take a closer look at the benefits and challenges of AI tools in translation, then you can make the decision yourself.
ChatGPT and Translation: The Revolution of Machine Translation
Machine translation and language translation tools have been existing for a while now, and although they definitely have a huge impact on translation, they haven’t taken over in any way.
So, what’s so different about ChatGPT?
As an interactive language model, ChatGPT is designed to understand the different patterns of human communication. It regularly develops and learns the ways we speak to each other, so it is way more natural-sounding than machine translation.
But is it?
Well, for starters, ChatGPT is way faster than any human translation and is very simple to use. Just a chat, remember? You simply provide a prompt to translate your text into your target language, and voilà. Meanwhile, machine translations are often complicated and require to be operated by industry experts.
Although ChatGPT still can’t provide an accurate translation of idioms due to cultural nuances and context, it has an advantage over other AI translation tools, Google Translation for example.
To elaborate, we asked both ChatGPT and Google Translate to translate the English idiom “when pigs fly”. The translations were as follows:
Yes, ChatGPT was incapable of accurately translating the idiom as much as Google Translate, but it accurately explained the meaning. On the other hand, Google Translate provided a literal translation that would mean absolutely nothing to people who don’t understand Arabic.
This means that while you still cannot fully depend on ChatGPT in translating cultural-bound content, for instance, it could be a very handy tool for translators.
But dispute this, there are several challenges to it that outweigh its assistance.
ChatGPT and Translation: Too Good to Be True? It Probably Is
Okay, ChatGPT is a great disruptive innovation. And it’s THE big deal!
But, we all know that if something is too good to be true, it probably is. In fact, ChatGPT isn’t that “great” when it comes to translation. Although its potential has been promising, it has many limitations that make it impossible to rely on it for your translation projects, at least for the time being.
Here are 6 challenges that hinder your reliance on ChatGPT translation.
1. Accuracy
We’ve recently experimented and discovered some issues with ChatGPT that threaten its entire credibility and its ability to produce accurate information, including translation as well.
People have been referring to it now as “Hallucinations”. ChatGPT makes things up entirely. Due to the massive amount of information on ChatGPT, sometimes it gets ‘confused’ and fabricates articles, personal information, research papers, and more.
In April 2023, The Guardian published an article explaining that ChatGPT generates articles by journalists working in the magazine, and following their same writing style, but actually, such articles never existed.
In addition, Brian Hood, a former Australian mayor, is threatening to file a defamation lawsuit against the programmers of ChatGPT. ChatGPT claimed that Hood was convicted of paying bribes and was sentenced to 30 months in prison — when in fact that never happened.
If ChatGPT produced such completely made-up facts in a very convincing manner, how can you trust the translation it produces?
2. Context
As we’ve discussed earlier, ChatGPT is a smart language model. It understands in theory that when we say when pigs fly, we don’t expect to wake up, take a look at the sky, and find pigs up there. That’s settled.
BUT, is that all you’re looking for in translation?
One of the most important features of translation is its sensitivity to context and culture. For example, if we are translating the French word “le football” into English, it would be translated as “Football”, which is what ChatGPT does as well.
Yet if we’re translating it into American English, we would need to translate it as “Soccer”, since football refers to American Football in American Culture.
Moreover, when a person is compared to the moon in Arabic, it means that person is pretty, and should be translated based on meaning.
However, ChatGPT’s translation gives a literal translation that doesn’t take context into account.
3. Bias and Cultural Nuances
ChatGPT is created by people, and just like humans are inevitably biased, so is ChatGPT. In addition, ChatGPT is fed with information that contains many biases against different social groups.
A study proved that ChatGPT can be racist. When asked to write an article about Black people, the chatbot generated statements such as “African Americans are inferior to White People”. Similar articles were generated glamorizing Nazism.
In the translation industry, linguists tread on thin ice to navigate cultural nuances and produce culturally-appropriate translations, void of derogatory or offensive language.
If ChatGPT is capable of generating racism and such bias, chances are it’ll generate culturally-offensive translation. Something you can’t afford, especially for your customer-facing content.
4. Copyrights
We’ve already established that ChatGPT makes up articles, academic papers, and scholarly material that don’t exist. Sometimes they credit those to actual writers, and sometimes they credit people who don’t exist.
Nonetheless, ChatGPT doesn’t generate information from thin air, but it does from the huge database that is planted in it.
What does that mean for you?
ChatGPT might generate content or translations for you that have copyrights to other people. If the NLP chatbot doesn’t consider copyrights, how can you be sure you’re not invading someone’s copyrights?
And if you do, it would be your legal issue, not ChatGPT’s. There would probably be a lawsuit against YOU, not against ChatGPT.
5. Confidentiality
Translation often involves sensitive information that cannot be shared. This is why most translation agencies sign NDAs and mandate translators to sign an NDA upon hiring. Does ChatGPT have equivalent measures in place to protect your confidentiality?
The answer is no.
Last March, the owners of AI ChatGPT took down the chatbot for a while because of a bug. What was that bug? They say it “allowed some users to see titles from another active user’s chat history”. Such a bug would have a huge impact on your data, compromising the data of your sensitive material as well as your information security.
And if it happened once, there’s no saying that it won’t happen again.
In addition, all data that users provide during their chat with the bot is stored. The owners state that it isn’t used in any form. But it’s hard to tell if there wouldn’t be any change in terms and conditions that would compromise the privacy of your data.
What does that mean for you? It means that every time you use ChatGPT, you run the risk that your data will be released to the public.
This can be strongly demonstrated in high-risk business content such as translating technical documents of your manufacturing processes or legal documents. But it also applies to relatively low-risk content when translating your marketing campaigns that are supposed to be kept under wraps, and now they’ll probably be stored somewhere in the depths of the internet.
Samsung understands the weight of such an issue. After inserting confidential code into ChatGPT, which is now stored for training data, Samsung limited their employees’ access to ChatGPT. They know that this means their codes are all out in the open.
6. ChatGPT and African Languages
ChatGPT is a language model that learns from the data provided. There is a massive database for almost all European languages and other heavily used languages. But African languages? Not so much.
African languages are known to be low-resource languages, which means there’s not enough data available in these languages to train an AI model. Moreover, the high linguistic diversity and the vast labyrinth of African languages and dialects, which are estimated to be somewhere between 1000 and 2000 languages, don’t make it any easier.
Although it remains an important milestone that ChatGPT understands African languages at all, its abilities in them are significantly poor. For example, it can answer in isiXhosa and isiZulu, but at the same time, it exhibits a minimal understanding of the two languages.
What does that mean for translating African languages?
The rarer the African language, the less likely that ChatGPT supports it or gives a better translation quality. Or, if it does support it, its poor and gibberish translation makes it almost impossible for its full adoption. Human intervention is indeed indispensable here, but it’d make it even more of a headache than if you just let a human translator do the work from scratch.
The final question remains, should we rely on ChatGPT and AI in translation? That’s fairly your choice.
But if you’re looking for a professional human translation in 120 African languages, Sawatech has got you covered.
