Artificial intelligence has seen significant advances in many areas, including machine translation. However, we cannot blindly trust artificial intelligence without human oversight for certain sensitive applications. In this article, we will discuss the importance of translation companies in the context of artificial intelligence, highlighting the quality of the work of human translators when compared to artificial intelligence systems like chatGPT.
Artificial Intelligence in Translation: A Technological Marvel with Limitations
Artificial intelligence has demonstrated incredible capabilities in many domains, including text translation. Artificial intelligence language models, such as chatGPT, are capable of performing various tasks, from translating texts to generating computer code. However, despite their advanced abilities, they are not infallible.
Artificial intelligence does not have human common sense and does not understand complex context to the same degree that humans can. Artificial intelligence has difficulties in generating complex structured content, in handling multiple tasks at the same time and can even produce potentially biased responses due to biases in training datasets. In addition, artificial intelligence is known to have accuracy and grammar issues, and it can be expensive to use due to computational costs.
The Matter of Sensibility: Why Human Oversight is Key
Would you completely entrust your health or safety to a machine, for example, during a flight or a surgical procedure, without any human oversight? That is what you are doing when you entrust your translation to artificial intelligence, without the oversight of a professional who knows how to correct potential errors made by artificial intelligence.
Health equipment, medications, safety equipment and flights all have user manuals, manufacturing and handling instructions. These manuals and instructions need to be translated into several languages so that everyone can understand them.
There are serious errors that it can easily detect, but could it recognise that ‘amygdala’ in English is not the same as ‘amígdala’ in Portuguese?
Amygdala (‘amígdala cerebelosa’ in Portuguese) is a brain structure located in the cerebellum.
‘Amígdalas’ (‘tonsils’ in English) are organs located in the throat.
Examples of Artificial Intelligence Errors
There are well-known historical examples, for example Uber’s self-driving car which killed a pedestrian in the USA. The car was unable to correctly identify the pedestrian, as the pedestrian was on the pavement as expected under normal circumstances. The existing artificial intelligence models and training were not correctly implemented by the system.
Another example is an AI-based recruitment tool used by Amazon to recruit top talent by analysing only their resumes. It so happens that the machine was trained using biased data from the last 10 years, during which the vast majority of candidates were men. Thus, the tool gave priority to male candidates.
But we can ask ChatGPT a few simple questions ourselves.
Error 1: Countries that start with O in alphabetical order
Not only does ChatGPT respond with only 1 country that starts with the letter O, but most countries listed start with the letter P.
If we then ask which countries start with the letter O, without requesting them to be ordered alphabetically, here ChatGPT is not accurate.
Of the 4 countries listed, 3 are repetitions, and another one seems to be in French (minus the acute accent => Ouzbékistan).
We ask the same question again, but with more specific wording, asking for countries that start with the letter O as their first letter:
It does apologize for the repetition earlier, but no mention to why Uzbekistan made it to the list.
Error 2: A simple calculation.
It starts by giving the wrong result, but it arrives at the right answer after showing the workings for the calculation.
The Role of Translation Companies in the Age of Artificial Intelligence
A translation company like M21Global plays a crucial role in ensuring the quality of your translations. Translation specialists check and correct the quality of the translation produced by artificial intelligence. In fact, we have been using machine translation, neural networks and artificial intelligence in the translation sector for over 10 years. Our job as translators is not to do what machines do faster and more efficiently than a human translator, but rather, our job is to check that the translation provided by the machine is accurate, and to correct or improve the sections which are not.
Our (proprietary) machine translation system incorporates AI translation into our workflow. The system tries to identify the best possible translation and presents it to the translator, who then reviews it and checks that it agrees with the original text (i.e., whether the terminology, semantic accuracy, lexical cohesion, style, localisation, the formatting and the translation’s target audience are in accordance with each client’s specific needs). The system is subsequently fed again with this new translation, so that future translations are increasingly reliable and accurate.
Conclusion: Artificial Intelligence and Translation – A Partnership, Not a Replacement
Although artificial intelligence is advancing by leaps and bounds, there is still a long way to go, that is, if the destination can even be reached, before it can completely replace the need for human intervention in sensitive applications such as translation. While artificial intelligence can help provide alternatives in decision making, ultimately humans still need to be involved in the decision-making process. At most, artificial intelligence is a form of augmented intelligence and not pure artificial intelligence which can replace the subtlety of human intelligence.
A translation company, with its experience and specialist knowledge, plays a crucial role in ensuring the quality and accuracy of your translations. By understanding the limitations of artificial intelligence and applying human oversight when necessary, you can ensure that artificial intelligence is used safely and effectively.
Article inspired by: