MIT's AI translator deciphers texts in long-lost languages

At the Massachusetts Institute of Technology, an algorithm has been developed that allows you to create an AI translator specializing in one of the lost languages. These are those adverbs and dialects that no one has spoken for a long time, from which only a few notes remain, but almost no information about syntax, grammar and vocabulary. A human linguist in such a situation does not have enough data to work with, but AI is able to generate the missing fragments and read ancient inscriptions.

The authors of the development were Professor Regina Barzilai and graduate student Jaming Luo, who last year published developments on deciphering the Ugaritic language and Linear B. They proceeded from the fact that the former is the evolutionary development of Hebrew, and the Greek language is considered the “ancestor” of the latter. The new AI will be able to independently select languages ​​that are related to what is to be deciphered, and learn it through this connection.

Historical linguistics clearly indicates that all languages ​​develop according to certain patterns. For example, due to the closeness of hearing the sounds "b" and "n", they will certainly be confused when inventing new words. But confusing "b" and "k" is unlikely to work, although "Baal" easily transforms into "Baal". Knowing such patterns and having a well-known language as a basis, AI will be able to simulate many options for the formation of some words from others. By analyzing these models, the system will build the most plausible "design" of the language that people could speak.

The first experiments with the algorithm proved the correctness of those scientists who argued that the Iberian language is not related to the Basque language. Now the development of a module for analyzing the semantic content of the language is underway so that AI can highlight keywords and search for them in context. For example, if an unreadable name of a commander is encountered, but the approximate place and time of the battle is indicated, the AI ​​will be able to find this event among the historical chronicles and thus find out the name of the warrior.