Presentation 2021-09-18
[Keynote Address] Neural Network as an Explainable Human
Yugo Murawaki,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this talk, I argue that techniques developed in the field of explainable AI (XAI) have potential applications in computational human sciences. In typical XAI scenarios, artificial intelligence is seen as a technological Other that is obliged to win human trust by explaining the black box. In computational human sciences, however, it is humans that are the black box, and artificial intelligence serves as an approximation of human functions. This observation motivates us to use explanation methods to explain humans. As a concrete example, I show that neural network-based classifiers can be applied to contrastive studies. We begin by training a classifier to discriminate texts written by two groups of humans and then apply an explanation method to analyze how it performs classification. A major advantage of this approach is that the high expressive power of modern neural networks allows us to investigate context-sensitive words and long expressions in an explorative manner.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) explainable AI / neural networks / classifiers / contrastive studies
Paper # TL2021-16
Date of Issue 2021-09-11 (TL)

Conference Information
Committee TL
Conference Date 2021/9/18(1days)
Place (in Japanese) (See Japanese page)
Place (in English) Online
Topics (in Japanese) (See Japanese page)
Topics (in English) Language Processing and Language Learning
Chair Yasushi Tsubota(Kyoto Inst. of Tech.)
Vice Chair Tadahisa Kondo(Kogakuin Univ.) / Kazuhiro Takeuchi(Osaka Electro-Comm. Univ.)
Secretary Tadahisa Kondo(Kobe Gakuin Univ.) / Kazuhiro Takeuchi(Ferris Univ.)
Assistant Nobuyuki Jincho(Miidas) / Akio Shimogori(Hakodate-ct)

Paper Information
Registration To Technical Committee on Thought and Language
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) [Keynote Address] Neural Network as an Explainable Human
Sub Title (in English) A New Approach to Contrastive Studies
Keyword(1) explainable AI
Keyword(2) neural networks
Keyword(3) classifiers
Keyword(4) contrastive studies
1st Author's Name Yugo Murawaki
1st Author's Affiliation Kyoto University(Kyoto Univ.)
Date 2021-09-18
Paper # TL2021-16
Volume (vol) vol.121
Number (no) TL-180
Page pp.pp.23-27(TL),
#Pages 5
Date of Issue 2021-09-11 (TL)