Teacher-neural network "Olga Stanislavovna" was taught Russian slang and sarcasm

Developers from LiveDune, a social network analytics company, have created a neural network that can

assess the tone of comments on social networks. The program, which the company director named “Olga Stanislavovna” in honor of the teacher, will work together with philologists.

Artificial intelligence owns twolanguages: to a lesser extent Russian literary and to perfection the slang of the Russian language, which was formed in 2020–2022. To train the neural network, the developers used dictionaries of evaluative words and swear words. But they did not focus on literary collections, but on the language spoken on social networks.

For initial training, developers manuallymarked up 10,000 Russian-language comments from the social networks of the company's clients. After that, the service was put into operation, but, as the creators say, almost immediately there were complaints about the incorrect work of AI. The program did not recognize sarcasm well, and, for example, the comment: “this is a damned post” was considered positive.

Service example. Image: Live Dune

To further train the system, the creators providedcustomers the opportunity to adjust the assessment of "Olga Stanislavovna". The corrected data is sent to the system and used to refine the algorithms. In fact, the neural network used more than 500 thousand records for training.

Programmers say that the hardest thing wasteach "Olga Stanislavovna" to identify irony and sarcasm. For this, training materials were additionally loaded into the neural network - dictionaries of jargon and obscenities. And, of course, practice on real reviews with feedback from AI users.

"Olga Stanislavovna" has the most relevantthe vocabulary of modern Russian slang - the language used in social networks. About 20 million comments “passed” through the neural network on VKontakte alone. The developers plan to use the data accumulated by the system to study the Russian language together with philologists.

Borrowings, abbreviations andcolloquial-colloquial elements are massively used by us when communicating on the network. Dictionaries fixing the language norm have always lagged behind live speech, but the appearance of such a neural network can help philologists more accurately track the development of the language.

Maria Rogozhina, philologist

Read more:

Soon a solar storm will hit the Earth: the material flies at a speed of 800 km / s

Scientists filmed a strange creature with tentacles, which they mistook for a flower

Russia leaves the ISS: what will happen now and why the maintenance of the station is under threat