Intelligent Robot Dialogue (RIKEN)
Research Staff
-
Professor
Koichiro YOSHINO -
Assistant Professor
Seiya KAWANO
Research Areas
We are working to realize an intelligent conversational system for a robot toward a society where robots live with humans together. In order to realize conversational intelligence, we are working on three issues: understanding, management (control), and generation. We use multimodal features; speeches, images, and gestures for understanding and generation with focusing on their relation to language. Dialogue control (management) is also an important research topic, including reasoning, inference, and knowledge structure.
Language understanding and situation understanding
Language understanding means converting natural language expressions into machine-readable expressions. Our research focuses on understanding the contents of text and speech and uses various non-verbal information as input to understand the user's situation in dialogue correctly. We also work on fundamental natural language processing problems in spoken languages, such as morphological analysis, syntactic parsing, semantic parsing, and discourse parsing.
Investigating mechanisms of knowledge acquisition and memory
Knowledge acquired through dialogue and daily observation is important for a proper understanding of people and their surroundings. We study the knowledge acquisition process through dialogue and construct the knowledge system of robots to build an intelligent robot that can cooperate with human users. Another goal is to investigate the mechanisms of human memory through such constructive research.
Intention, dialogue models, reasoning, inference, control and decision making
Based on the results of understanding and knowledge acquisition, we work on the decision-making system that determines the responses and actions of a dialogue system/robot. We are particularly interested in using reinforcement learning to exploit the long-term effects of decision-making toward certain objectives. We also work on the reinforcement learning algorithm itself. We will also investigate how dialogue and decision models can be affected by various dialogue phenomena, such as politeness and entrainment.
Language, speech, and motion generation
With the advent of neural networks, we can generate a natural sequence of natural language, speech, or others. In order to make the generation system operate based on the intentions of the decision-making system, we study the controllable generation system and the effects of the control results. According to the intention, the system generates not only the language but also other robots' behaviors.
Key Features
Research topics in our laboratory are in a multidisciplinary area; thus, we welcome candidates from outside the field. What you need most is an interest in and curiosity about human intelligence, a passion for investigating what you are interested in, and the determination to accomplish it at all costs.
In the first semester of the first year, you will repeatedly meet with faculty members to determine your research topic according to your interests and research focus of the laboratory. We emphasize the development of your ability to work independently on your problems (research topics). You will be able to broaden your choice of research topic if you study mathematics, basic informatics (algorithms, data structures, information theory, etc.), computer programming (especially Python) after you pass the entrance exam.