With all the technological breakthroughs that have been witnessed during the last decade in what has been called a breathtaking era in human history, it is now commonly believed that Artificial Intelligence (AI) can pretty much replace translators… outside of the translation industry, that is.
Easy to use, fast and usually free, Machine Translations are often seen as the perfect solution by students and used in school settings. They are a great learning tool and help making other languages much more accessible, however they become a considerable issue when they start spilling into the workplace. Despite being useful and sometimes seamless for short sentences, they are not flawless and cannot be fully integrated into the work environment without a language service provider involved in the process. Inviting MTs into a company by using them for the translation of a website or of a marketing campaign guarantees a clumsy wording, or, in the worst cases, an offensive one.
Artificial Intelligence in translation is in a way a technology based on fixed rules struggling to grasp the ambiguity of language and the subjectivity of the field. Thanks to technological innovations, automatic translations have gotten better at dealing with context and nuances, however here the real downsides mainly lie in the details, or rather in the social issues MTs cannot grasp. Not only can the training data used by MTs include widely imbalanced representation regarding genders (for instance by mentioning five times more male ministers than female ministers) but they also get a third of one percent of their phrases from pornographic sources which can lead their automated translations to contain slangs. At the end of the day, MTs only achieve 25% reliability, and the risk of finding a slang in the automated translation is a risk too high for companies to take.
(more data and charts on this issue can be found in this post)
All of these points represent proofs as to why Machine Translations must be used with caution, they are not to be handled carelessly and especially not in a working environment. In spite of all the innovations, they have not yet managed to replicate human ability, and especially not a translator’s one. The only way they can be safely integrated into the workplace is by letting a language service provider revise them, which is the exact reason why nowadays, Artificial Intelligence cannot replace translators, but it sure can become their new right-hand man.
Post-edition
In the translation industry, one of the most commonly used technique involving AI remains post-edition. It even seems to be the strategy that helped gradually invite MTs in our workplace when they used to be seen as the enemy in their early days. It helps achieve the perfect product by combining the speed of AI with the linguistic skills and wisdom of a trained translator. With MTPE (Machine Translation Post Editing), a source text is first ran through a high-quality Machine Translation before being thoroughly analyzed and edited by a language service provider as a way to avoid any inaccurate grammar, misinterpreted context or misused terminology. It is a great way to handle projects involving a large volume of text, however like any other great trick, it must be used with caution.
If automatic translations cannot replace language service providers, there are even moments where they cannot even help them. Even though MTPE is a useful technique that can save translators a lot of time, it’s not a method suited for all projects. Whilst it works well with economic texts where the sentences are short and the wording plain, MTs often struggle to provide a clear translation of meteorological texts. Running texts too complicated through a MT with the intention of editing it will result in the translator having to re-think almost every sentence and waste more time revising it than they would’ve translating it firsthand.
Subtitling
In the same way as Machine Translations, automatic subtitling is proof of a great technological innovation, but a breakthrough that cannot be used on its own yet. Using ASR software (Automatic Speech Recognition), automatic subtitling is a fast and cheap way to enable wider accessibility, but it still cannot replicate human ability and struggles with aspects such as speech recognition accuracy, segmentation of two-line subtitles and reading speed. Because of these hindrances, automatic subtitling only reach between 60% and 90% accuracy, which is why creators often prefer to hire language service providers — accessibility should not mean sacrificing quality.
If the issue is the same as with MTs, then so is the solution. With editing, the accuracy rate of automatic subtitles easily go up to 98%. Even with automatic subtitles, hiring language service providers is necessary for creators if they want to avoid any clumsy subtitling thanks to editing. ASR software are useful, but they are not always a solution in themselves.
Time-saving software
Working with AI isn’t always about revising its flawed work, it can also be about translating more efficiently thanks to its software. Tools such as Smartling are translation software involving AI that allow its users to considerably optimize their process. Not only does the AI provide constant quality checks, but it also surfaces relevant data to the source text, compiles an automated quality score for the final product, and even sometimes includes an automated writing collaboration tool which can suggest interesting equivalents to your translation.
In many ways, AI is translators’ future best colleague. We just need to live with it and consider it as one of the many tools we have at our disposal.
If you wish to know more about post-edition or our working process, reach out to us now!
