MTS AI CEO Alexander Khanin and NLP engineer Nikita Martynov presented their products and solutions in the field of AI at NVIDIA GTC.
Natural language processing in theory and practice
The conference was held online from March 21 to 24. Among its guests and speakers were leading experts in artificial intelligence, visionaries and representatives of large corporations who shared their solutions and discussed the possibilities of AI to meet the world’s most complex challenges.
On March 23, MTS AI CEO Alexander Khanin took part in a panel discussion titled “Natural Language Understanding in practice: Lessons Learned from Successful Enterprise Deployments. It was also attended by other speakers: Kari Briski, vice-president of software product management for HPC/AI NVIDIA, Jiang Bian, associate professor department of health outcomes and biomedical informatics College of Medicine, University of Florida and Magnus Sahlgren, head of research for Natural Language Understanding, AI Sweden. They discussed the adoption and successful deployment of customized natural language processing (NLP) to an enterprise’s unique vocabulary, customer relationships, and the datasets on which their business runs. Alexander Khanin spoke about the MTS AI’s expertise and best practices in developing chat and voice bots for business customers in Russia.
The company’s developments in the field of natural language processing are used in various spheres, including telecommunications, banking, and so on. Chat and voice bots created by MTS AI help to serve tens of millions of smartphone app users, call center customers and home device owners daily within the MTS ecosystem and its partners.
”Our main idea is to provide the best customer service and instant access to all ecosystem functions using chatbots and voice bots,” said Alexander Khanin, CEO of MTS AI, speaking at the NVIDIA GTC conference.
Twice a year MTS AI seriously modernizes its language models, while continuing to work on improving its NLP solutions. Now MTS AI is actively exploring methods of content detoxification which can be used in chatbots and voice assistants training in the future.
On March 24, MTS AI team members presented their individual solutions and developments. NLP engineer Nikita Martynov spoke about the creation of a paraphrase generator using reinforced deep learning for the Russian language. This solution is used as part of the NLP platform built by MTS AI. It helps to expand the chatbots and voice bots vocabulary, and allows them to expand it when communicating with users. According to Martynov, several datasets containing more than 1 million pairs of paraphrases in Russian were used to create the generator. They were collected, among other things, from media texts. Thanks to this, the generator can reformulate a short query, for example, “Moscow-Toronto plane”, into a full-fledged sentence “Moscow-Toronto plane took off from Moscow”.
Later in the day, MTS AI CEO Alexander Khanin made a presentation on new generation chatbots for customer service based on generative and transformative neutral networks. He spoke about the main trends in the creation of virtual assistants based on the Transformer type language models. Modern chatbots and voice bots can be trained by communicating with customers, without involving additional data sets. In addition, the new generation of virtual assistants are able to support different conversation scenarios, which allows them to conduct a dialogue with users without losing context.
The presentations of the MTS AI team and other speakers are available by registering on the NVIDIA GTC conference website.