Machine solves examinations: benchmarking natural language processing technology
- 講者Yoshinobu Kano 教授 (Informatics at Shizuoka University, Japan)
邀請人：蘇克毅 - 時間2016-11-29 (Tue.) 10:00 ~ 12:00
- 地點資訊所新館106演講廳
摘要
Answering examination questions has been attracting more researchers for these days. Examinations are good benchmarks because scoring is well defined. On the other hand, solving examinations include broad range of different intellectual processes that are still very challenging for machines as examinations assume "common" language ability of human applicants.
I will introduce a couple of examination question answering tasks which our laboratory has been trying to solve. The Todai Robot project aims to solve Japanese university entrance examinations (multiple choice selection and descriptive) for almost all of the required subjects, where we focus on the Social subjects introducing other subjects briefly.
We have also been challenging the National Examination for Medical Practitioners where doctors' medical experts knowledge is required. I also introduce our challenge of Legal Bar Exam for lawyers, which are much more difficult than others because Legal Bar Exam requires abstract thinking and logical inferences, not just legal technical terms.
BIO
Prof. Yoshinobu Kano (kano@inf.shizuoka.ac.jp http://kanolab.net/kano/index.en.html) is an associate professor of Faculty of Informatics at Shizuoka University, Japan. He got his BS in physics (2001), then MSc (2003) and PhD (2011) in information science and technology from the University of Tokyo. He has been a research associate in University of Tokyo (2009), JST PRESTO researcher (2011), and an associate professor (PI) in Shizuoka University (2014-).
His current research theme includes more human-like dialog system (AI werewolf, copy generation), examination question answering (social, medical, and legal), medical NLP (EHRs, MedNLP), text mining for neuroscience papers. He has served as organizers for these different projects.