Extraordinary Advance Could Help Decipher Lost Languages

It is an extraordinary advance that could help decipher lost languages. The approach built on setting up artificial intelligence limits beyond the available data. The algorithm does not even need to know what each word means

In 1886, the British archaeologist Arthur Evans found an old stone with an enquiring set of inscriptions in an unknown language. The nugget came from the Mediterranean island of Crete, so Evans traveled immediately there to search for more evidence. He soon found many stones and tombstones with similar writings and dated them about 1400 BC. vs.

They made an inscription one of the first forms of writing ever discovered. Evans argued that his linear form had its origin in the simple drawings drawn that represented the beginnings of art, emphasizing its importance in linguistics history. lost languages

Subsequent surveys determined lost languages

Subsequent surveys determined that stones and tombstones we inscribe in two different languages. The oldest, called Linear A, dates between 1800 and 1400 BC. C., When the island we dominate by the Minoan civilization of the Bronze Age. The other writing, linear B, is more recent; it appears after 1400 BC. C., when the Greek Myceannes of the continent have conquered the island.

Evans and other researchers have tried to decipher the old scripts for many years, but these dead languages ​​have resisted all attempts. The problem had not to answer before 1953 when an amateur linguist named Michael Ventris cracked the code of B.

The answer we create on two key findings. First, it assumed that many of the Linear B B vocabularies repeated words were placement names on the island of Crete. Turned out to be ok. His second discovery was the hypothesis that writing accounted for an early form of ancient Greek. This idea allowed him to decipher the rest of the tongue quickly. In this process, Ventris proved that the former Greek had appeared for the first time in written form many centuries sooner than expected.

The work of Ventris was an outstanding achievement. But the oldest, linear A script has remained one of the big linguistic problems so far.

It is easy to think that recent progress in automatic translation could help you. In just rare years, the study of linguistics has stood developed by the availability of enormous annotated databases and the development of machines techniques to learn from them. Therefore, the automatic translation of one language to another has become routine. And although it’s not perfect, it has provided a brand new way of thinking about the language.

Mit Jiaming Luo lost languages

Thus, researchers Mit Jiaming Luo and Regina Barzilay and the Google Artificial Intelligence Lab in California (USA) Yuan Cao have decided to use it to solve the problem. The team has developed a machine learning system capable of deciphering dead languages, and they have demonstrated it by decoding linear B. This demonstration has become the first automatic translation of this language. However, their approach is nothing like standard machine translation techniques.

The process begins by mapping these relationships for a specific language.

The process begins by mapping these relationships for a specific language. It requires enormous text databases. The algorithm then analyzes the text to see how often a word appears next to another. This model is a unique signature that defines a term in multidimensional parameter space. A comment can be considered a vector in this space. Vector acts powerful constraint on how this word may appear in any translation from the algorithm.

These vectors follow some simple mathematical rules. For example: king – man + woman = queen. And a sentence can be considered a set of vectors that follow one after the other to form a kind of trajectory space.

The main idea is that automatic translation can find words in different languages ​​that occupy the same points in their respective parameter spaces. In this way, the process of translation of sentences becomes the process of searching similar trajectories through these spaces. The process does not even need to “know” what the penalties are. Allows a complete language to be a map to another direct correspondence.

The approach to build on large datasets. But a few years ago, a team of German researchers has shown.

And also Read: Biometrics