Presentation 2020-01-26
Modeling for Infant Vocabulary Acquisition System using Deep Reinforcement Learning
Masaki Taguchi, Yasuhiro Minami,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We propose an infant vocabulary acquisition model that identifies psychological infant vocabulary development findings (joint attention, learning bias, and understanding intention) associated with symbol grounding using deep reinforcement learning. In deep reinforcement learning, we use DDQN and LSTM, which treats the time sequence data of long-term dependency. We use the features obtained from real objects to ground words to those objects. Simulation experiments investigated the symbol-grounding abilities of the model and the appearances of psychological findings in the process of infant word acquisition. We confirmed that our proposed model can ground words to the objects and achieved joint attention and understanding intention. We also confirmed that it acquires (by learning) noun bias, which is thought to innately exist by many psychologists. These results confirm that the multiple psychological phenomena of language acquisition can be modeled using the latest neural network.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) double deep q-network / long short-term memory / image recognition / feature extraction
Paper # HCS2019-73
Date of Issue 2020-01-18 (HCS)

Conference Information
Committee HCS
Conference Date 2020/1/25(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Room407, J:COM HorutoHall OITA (Oita)
Topics (in Japanese) (See Japanese page)
Topics (in English) Psychology and Life-stage of Communication, etc.
Chair Masafumi Matsuda(NTT)
Vice Chair Tomoo Inoue(Univ. of Tsukuba) / Yugo Hayashi(Ritsumeikan Univ.)
Secretary Tomoo Inoue(Kanazawa Inst. of Tech.) / Yugo Hayashi(Osaka Electro-Comm. Univ.)
Assistant Tomoko Kanda(Osaka Inst. of Tech.) / Kazuki Takashima(Tohoku Univ.) / Ken Fujiwara(Osaka Univ. of Economic) / Kazunori Terada(Gifu Univ.) / Atsushi Kimura(Nihon Univ.) / HUANG HUNGHSUAN(Riken)

Paper Information
Registration To Technical Committee on Human Communication Science
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Modeling for Infant Vocabulary Acquisition System using Deep Reinforcement Learning
Sub Title (in English)
Keyword(1) double deep q-network
Keyword(2) long short-term memory
Keyword(3) image recognition
Keyword(4) feature extraction
1st Author's Name Masaki Taguchi
1st Author's Affiliation The University of Electro-Communications(UEC)
2nd Author's Name Yasuhiro Minami
2nd Author's Affiliation The University of Electro-Communications(UEC)
Date 2020-01-26
Paper # HCS2019-73
Volume (vol) vol.119
Number (no) HCS-394
Page pp.pp.111-116(HCS),
#Pages 6
Date of Issue 2020-01-18 (HCS)