The neural network deciphered the drafts of Pushkin

The neural network unravels the secrets of manuscripts. To look under a layer of ink and see crossed out words is now possible thanks to artificial intelligence. How to decipher the drafts of Pushkin – learned Julia Kundryukova. Dmitry Nikolaev and Nikolai Pertsov have been working together for only a couple of months. They came together like ice and flame – a programmer and a philologist – in order to understand the drafts of Pushkin’s Eugene Onegin. Nikolai Pertsov, an employee of the Russian Language Institute of the Russian Academy of Sciences, devoted about thirty years to studying the manuscripts of Russian writers and poets. He is sure that in the drafts – a whole parallel universe with many details, which not even all experts know. “It is very interesting to compare draft versions and final ones, since one can see the movement of thought,” says Nikolai Pertsov, a leading researcher at the V.V. Vinogradov Institute of Russian Language. On the pages of the first draft of Onegin, he recently found that Eugene was not born at all “on the banks of the Neva”, but in the village. And this is due to one epithet – “patriarchal”, which Nicholas read in manuscripts. “That's amazing. That is, at first he had the idea that perhaps Onegin should be placed first in the village and then in St. Petersburg. He refused the epithet “patriarchal”, and he is unknown to Pushkin studies, explains Nikolai Pertsov, a leading researcher at the V.V. Vinogradov Institute of Russian Language. But why did a linguist need the services of a programmer? The fact is that deciphering any drafts or manuscripts is not the easiest task. First of all, handwriting: many things are simply impossible to understand. But the main difficulty is the words crossed out by the author or censor. And here to help the tired eyes of the philologist comes artificial intelligence. Here we witnessed a scientific discovery: Dmitry Nikolaev with the help of a trained neural network was able to remove the strikeout, and the word appeared in its original form. This is the eighth chapter of Eugene Onegin. "You are constantly called by the serpent / To himself, to the mysterious tree ..." - the word "mysterious" is written above the crossed out "perishing". This is the epithet Pushkin invented initially, but then refused. In order to carry out such a cross-section, the neural network was long taught to read Pushkin manuscripts. “We needed a large set of crossed out and unexplained words from Pushkin’s autographs. Then one neural network tried to cross out words that were not crossed out. And the second tried to distinguish from the real crossed-out words those that the first crossed out, says the technical director of Smart Engineering Dmitry Nikolaev. How this system works, Dmitry explained with the help of this video. There is no doubt about the reliability of the results, because this program is arranged as a generative-adversarial network. Simply put, one part of it crosses out and suggests the option, the other – checks whether this option is plausible. The prospects for this system are great: it can be trained on the handwriting of any classic. And perhaps soon we will have many sensational discoveries and new readings.

Reagiere auf diesen Beitrag

👍 0 ❤️ 0 😂 0 😮 0 😢 0 😡 0

Подписаться на NX Россия

🔗 Share

Facebook X (Twitter) WhatsApp Telegram LinkedIn WeChat VK LINE KakaoTalk Reddit Pinterest E-Mail