"What sport do you like?" "What do you think of Frozen 2?" This apparent small talk was actually carried out among robots called “chatbots”. It happened during the Third "Social Robotics" Forum i.e. the First Robot Group Chat Competition held in Tianjin on Dec. 14th.
The robot designed by Dr. Peng Zhang’s team, from Tianjin University’s College of Intelligence and Computing proved to be the most eloquent robot and won the first prize. “Robot group chat is a very novel mode of machine-machine dialogue, and its form has rarely been studied and applied to-date. It is very interesting and challenging.” Dr. Zhang noted.
In the robot group chat scenario, the competition organizer gives a group chat theme, which involves sports, music, movies, etc., and requires the robots to conduct multiple rounds of conversation with each other in free group chat under each theme. The robots have to generate fluent responses that meet the group chat theme and context logic. The "dialogue" process also tests the robot's ability to understand context semantics and to select the best reply. The competition results were evaluated by human judges based on the robot's output response to the group's theme and the relevance, fluency, diversity, grammatical errors, context, and response speed of the robot in the preliminary and final round of the competition.
"Robot group chat can be used to evaluate robots’ conversational intelligence in group chats of multiple robots under the same dialogue scenario. It mainly compares the collection of its data sets and the mining of dialogue data, the effective modeling of dialogue history, and the design of corresponding dialogue strategies. etc. " Dr. Weinan Zhang of Harbin Institute of Technology said. As the main promoter of the competition and the evaluation expert, Zhang explained that the competition aimed to promote the development of human-machine dialogue technology in multi-party dialogue scenarios and provide a good exchange and communication platform for academic researchers and industry practitioners related to human-computer dialogue technology.
By the College of Intelligence and Computing
