Artificial Polyglots: AI and Multilingual Mastery
AI is a burgeoning field and continues to grow in many directions, one of which is Natural Language Processing. Obviously, this is one of the more crucial elements given that AI must be able to comprehend and communicate in human languages, in order to be fed instructions and output intelligible data. Millions of dollars have already been invested in this technology, and this is only the beginning, as we are coming to notice.
This is how natural language processing works: An automated system records human speech acoustically and transcribes it into text. After semantic processing, the textual material is translated into another spoken language. Obviously, this basic system opens up pathways not only for communication between human language and machines, but also human languages with other languages. AI is quickly becoming an artificial polyglot, able to use inputs from one language and after processing, output in another language.
Late last year in 2023, a video went viral on X (Twitter) of a man speaking in German. The video ended and the same man standing in the same angle, lighting and position spoke the exact same words in French and then another language. Sure, it may have been an actual polyglot who could speak numerous languages. But the video was very clearly being translated - it was the same video, but AI had converted the man’s language, accent and lip movement into another language completely seamlessly. The video ended with an advertisement for HeyGen, a company that claims to be able to translate in real time.
This is an enormous change in the industry. For centuries, humans have relied on human translators, who would accompany the speaker of one language to contexts where a different language was spoken and translate for them after every sentence, and this would go back and forth. But in the same way that the term ‘computer’ once received people who would manually compute data and calculations, the term translator might very quickly come to mean an automated system that translates in real time so that you never even have to hear the original sounds.
We have not yet seen such technology, and it will certainly be some time before it is available for free access by anything with internet access. What is holding back the progress so far? It mostly comes down to financial resources and expertise. While we can already imagine how this technology would work, there are some missing parts in the calculations and a lot of data yet to be collected. This will take time and possibly be used by national governments before it trickles down to the rest of us!
Sources:
https://www.lmu.de/en/newsroom/news-overview/news/polyglot-machines.html
https://www.quickchat.ai/post/problems-with-multilingualism-of-llms