Amazingly, there are approximately 7,000 languages spoken throughout our world on a daily basis. However, this is actually just a fraction of the total number of languages that have been used throughout the history of mankind. In fact, today’s language total is actually less than one-fourth of that historical total.
In total, there have been approximately 31,000 languages that have been used throughout the history of the world. Just think about whenever one of these languages is lost, we have also lost a different line of thought and a different view of the world around us.
The Value of Languages
Each unique language brings with it a set of unique relationships, so when it is lost, we lose all the poetry that it brought to the world as well. So what if we were able to translate and figure out how to understand these dead languages? Research teams from Google Brain and MIT have recently developed an AI-based program that will be able to accomplish that very thing.
As our languages evolve and change, we have seen that a lot of the symbols associated with these languages, as well as the way that their characters and words are used, will stay relatively the same over time. Due to this, attempts can be made to decode a long-lost language whenever its relationship to a familiar progenitor language is known.
Achievements of the AI
This very insight is the thing that this research team employs machine learning to decode the ancient Greek language known as Linear B that was used around 1400 BC, and also a cuneiform Ugaritic that is over 3000 years old. The team was comprised of Regina Barzilay and Jiaming Luo from MIT, and Yuan Cao who works with the AI lab from Google.
The infamous Linear B was cracked previously in 1953, by a man named Michael Ventris. But this incident was the very first time that a machine decoded an actual language.
The methods employed by this research team decided to focus on 4 unique key properties that were linked to both the context and the character alignment – monotonic character mapping, distributional similarity, significant cognate overlap, and structural sparsity.
The team trained this AI network to find these characteristics and traits, accomplishing the right translation of 67% of Linear B cognates into the proper Greek equivalents.
The Future of AI that Decodes
The biggest value that AI brings into achieving these kinds of tasks, as reported in an MIT Technology Review piece, is that it’s able to use a bulldozer approach to accomplish tasks that are simply too exhausting for people to complete. These machines have the precision and the energy to undertake one tedious task after another, as they compare and test one symbol against another until they have finished the project.
So what is up next for these researchers? The obvious answer to that question is decoding Linear A, which is an Ancient Greek language that has baffled scientists for decades.
