The Alexa Prize Socialbot Grand Challenge is an undergraduate and postgraduate competition, first launched in 2016 and dedicated to the advancement of spoken artificial intelligence technologies. The teams are tasked with developing a social bot that can communicate fluently and coherently in English with people on news events and popular topics such as entertainment, sports, politics, technology, and fashion through Alexa-enabled devices. Another goal is an ambitious goal - to achieve a score of 4 out of 5 points and an average duration of more than 20 minutes in dialogues with the judges at the final stage. Prototyping will begin in November with a winner announced in August 2021.
The competition is being held for the fourth year and this time there are nine teams participating, including the winners of the past The Alexa Prize Socialbot Grand Challenge 3: Emory University, Stanford and Czech Technical University. Each team will receive a $ 250,000 research grant, access to cloud computing in Amazon Web Services, and support from Alexa developers.
The team of the Moscow Institute of Physics and Technology is called DREAM and is participating for the second year in a row under the leadership of Mikhail Burtsev . The team includes graduate students of Phystech and employees of the DeepPavlov project from the laboratory of neural systems and deep learning at MIPT: Denis Kuznetsov , Dmitry Evseev , Anton Peganov , Alsu Sagirova ,Dmitry Karpov and team captain, Dilyara Baimurzina .
To learn more about the architecture of the DREAM socialbot created for the previous competition, see the technical report of the DREAM team for the Alexa Prize 2019.
After the The Alexa Prize Socialbot Grand Challenge 3, the DeepPavlov team transformed the original bot and released the first version of the AI assistant - DREAM AI Assistant Demo . This bot is currently available on the demo site and in the Telegram messenger, and is ready to communicate only in English. DREAM Assistant combines about 40 different communication skills to allow open dialogue on a variety of topics. The helper is based on modern NLP models and components, including 14 annotators, 4 post-annotators, and WikiData's own knowledge graph integration. This work allowed to lay down support for working with knowledge graphs and expand the capabilities of the original bot, making it independent of Amazon services. But new ideas are required for a new round of the famous competition. If you have ideas and suggestions for improving the system, then we are waiting for feedback .
The DeepPavlov team worked hard to analyze the work of teams competing with its DREAM Team from other universities and form new directions for the development of the bot for The Alexa Prize Grand Socialbot Challenge 4.
We have a lot of work to do this year. Over the past year, my team and I prepared the engineering infrastructure, including interaction with Amazon services, learned how to test a bot and, most importantly, work in a team, '' says graduate student and team captain Dilyara Baimurzina. - Therefore, now we have everything to immediately start implementing ideas and, of course, reach the final!And if you are a student or postgraduate student of MIPT and have always dreamed of participating in a competition of this magnitude, then you have a chance to join the team and make DREAM a reality!