Improvement of Google Translate service depends only on the Kyrgyz people themselves. Tattu Mambetalieva, Director of Civil Internet Policy Initiative Public Foundation, said at the round table «Systems and Technologies of Artificial Intelligence: Risks, Opportunities and Approaches to Regulation in Kyrgyzstan».
Mirbek Okenov, the author of a project to translate 400,000 Kyrgyz phrases and terms for the online computer translator Google Translate, explained why the web service still does not translate texts into Kyrgyz perfectly.
«The system, of course, does not work perfectly. Google Translate has a neural grid. If the system gives the wrong translation, then you can offer the correct version. This means that you can help the neural network translate correctly. Prior to that, it worked on certain modules. We gave certain phrases, and it predicted what the translation might be based on the information taken from Wikipedia and, in general, from the Internet. There are few texts in the Kyrgyz language on the Web. This is bad for artificial intelligence. We have many enthusiasts who help this neural network develop, and every day the program starts to translate better and better,» said Mirbek Okenov.
Tattu Mambetalieva added that the improvement of Google’s translation functions depends on the activity of the Kyrgyz people themselves.
«It depends on the native speakers themselves, who write the maximum number of comments. They have a special window where you write that this or that expression was incorrectly translated. And the system improves every time,» Tattu Mambetalieva concluded.
