While AI and machine learning have enabled video language translators to support a wider range of voices, they are still far from being capable in all languages. Today, the most popular translation engines, such as Google Translate and Microsoft Translator, support more than 100 languages, which captures a sizable proportion of World languages but misses out on a large number of lesser or regional tongue. Such as Google translate which provides on the fly translation in 133 languages (mostly geared towards widely spoken languages due to the large datasets needed to train good models).
This is complex if you look back at several languages based on syntax, grammar, cultural nuances. Japanese and Korean, for example, are complicated languages in which certain words depend heavily on their context and can be easily mistranslated. Microsoft announced a 92 percent accuracy rate for European languages on its Translator platform in 2022, but the accuracy rate dropped sharply for less widely spoken languages like Xhosa or Wolof. This difference in accuracy illustrates the technical challenges that AI faces with languages that have low amounts of digital resources to work with.
Some platforms have tried to fill this gap incorporating neural machine translation, which means AI learns context, tone and native use of language. For example, as an effective language processing method, neural networks, the DeepL translator can offer a better translation between some languages such as Polish and Dutch compared to previous translation models, which shows the power of context-driven AI in providing better quality translation. Unfortunately, neural translation can be expensive to run, demanding state of the art GPUs and larges databases of languages which many platforms will not have access to or be ready to pay for all languages.
As the prominent linguist Noam Chomsky once said: “Language isn’t just words; it’s culture and history — and people. And that’s why translation is so tricky; it can never be a straightforward word-for-word conversion.” Automatic translators cannot correctly interpret cultural nuances which is key for a good translation, particularly for languages with few resources such as standard dictionaries or other digital information.
Now, are video language translators capable of translating all languages? No, but we are making headway. Over time, the accuracy and coverage of such models may improve as AI develops further and more data is collected for these underrepresented languages. While most of the video language translator software available works great for widely spoken languages, it is at best a mediocre alternative for less common dialects and culture-rich phrases.