Languages are a significant part of human evolution. Different communities use different ways to communicate with each other through distinct languages. Today, there are around 7,117 languages in the world, but languages uttered thousands of years ago are lost in current times, and their words, syntax grammar, and the significance is a puzzle. These ancient languages carry a lot of greatness because we yet don't know a huge part of information about a few human civilizations. However, Massachusetts Institute of Technology (MIT) is now attempting to evaluate and decode these ancient languages with Artificial Intelligence (AI).
These dead languages are not just for literary records, but they open a window into the new civilizations that spoke these languages. But because of fewer records and lack of related languages, even machine algorithms such as Google Translate are not capable to decode these dead languages. However, an innovation by the researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) support to understand these languages. The project was favoured by the Intelligence Advanced Research Projects Activity (IARPA). They have created an AI system that has been indicating a favourable response in automatically analyzing these lost languages without any previous records or relative languages. This system establishes a connection between languages, a recent instance was when the researchers used the system to verify recent scholarship suggesting that the language of Liberian is not related to Basque.
The team of researchers at MIT is led by Professor Regina Barzilay, and she aims to create an algorithm that can clarify ancient dead languages utilizing barely a few thousand words. Till now, they have translated Ugaritic languages and Linear B - the letter language was decoded by humans in decades. These languages were related to the early forms of Hebrew and Greek, respectively.
The AI System developed by MIT team utilizes laws that have origins in regulations and knowledge of ancient linguistics, that define reliable manners that a language evolves. For instance, a language rarely deletes or add another sound, however, specific sound shifts may be involved. A word with a ‘P’ in the parent language may shift into a ‘b’ in the descendant language, but shifting to a ‘K’ is less inclined to happen due to the massive pronunciation gap. The algorithm created by the team roots language sounds into a multidimensional opening were varieties in pronunciation are evaluated in the distance between adapting vectors. This method supports the algorithm capture patterns of language evolution and sets limitations. And the concluding model can segment words in an ancient language. The algorithm can also evaluate the nearness between two languages, and it has been recognized to accurately point out the exact language classes.
The new approach of the MIT team would involve identifying the exact meaning of the words, even if they don't know how to pronounce them. In future, the team aims for ‘cogent-based decipherment’, a method to broaden their work beyond the act of connecting phrases to related words in a known language. This method can distinguish all the connections to people or settings in the paper which can also be further analyzed with the known historical pieces of evidence. This new way of approaching and decoding the lost ancient languages can change several understandings about the process of human evolution and it can become a huge part to understand how our ancestors communicated with each other and evolved with time.
___________________________________________________________________
Reference:
- www.openculture.com
- www.news.mit.edu
- www.interestingengineering.com
- www.analyticsinsight.net