Hwichan Kim / 金輝燦

I earned my Ph.D. in Information Science from Tokyo Metropolitan University on March 2025 ( supervised by Professor Mamoru Komachi). Since April 2025, I have been working as an NLP research engineer at SB Intuitions.

Email : hwichan.kim[at]sbintuitions.co.jp

Professional Experience

2025.04 – present

SB Intuitions

NLP Research Engineer
I have been working on post-training for LLMs. In particular, I focus on improving model performance by applying RL strategies (e.g., GRPO). To achieve this, I conduct multiple experiments such as hyperparameter tuning and carefully selecting training samples. I also implement and manage the experimental environments required for these experiments.

2025.11 – present

Hitotsubashi University

Visiting Researcher
At Hitotsubashi University, I collaborate with students on computational humor, including conducting detailed evaluations of current state-of-the-art LLMs on the Oogiri task, a Japanese humor competition. Recently, we have been working on improving humor understanding and generation capabilities using RL strategies.
Publications: AAAI 2025

2025.06 – present

Tohoku University

Visiting Researcher
At Tohoku University, I collaborate with students on multilingual NLP research, with a focus on enhancing multilingual model performance and analyzing the interpretability of multilinguality.
Publications: BlackboxNLP 2025

During my master’s and doctoral studies, I completed several internships and part-time positions, as listed below. I am deeply grateful to these institutions and collaborators for the valuable experiences they provided.

2025.03 – 2023.11: CyberAgent AI Lab
2024.09 – 2024.12: MBZUAI (supervised by Timothy Baldwin and Kentaro Inui)
2023.09 – 2024.01: Rimo Voice
2020.08 – 2024.09: Insight Tech

Publications

International Conference/Workshop (Refereed)

Ritsu Sakabe, Hwichan Kim, Tosho Hirasawa, Mamoru Komachi. Assessing the Capabilities of LLMs in Humor:A Multi-dimensional Analysis of Oogiri Generation and Evaluation. AAAI. 2025 [paper]

Suchun Xie, Hwichan Kim, Shota Sasaki, Kosuke Yamada, Jun Suzuki. Can Language Neuron Intervention Reduce Non-Target Language Output? BlackboxNLP. 2025. [paper]

Taisei Enomoto, Hwichan Kim, Zhousi Chen, Mamoru Komachi. A Fair Comparison without Translationese: English vs. Target-language Instructions for Multilingual LLMs. NAACL. 2025. [paper]
Hwichan Kim, Jun Suzuki, Tosho Hirasawa, Mamoru Komachi. Pruning Multilingual Large Language Models for Multilingual Inference. Findings of EMNLP. 2024. [paper] [poster] [slide]
Ayako Sato, Tosho Hirasawa, Hwichan Kim, Zhousi Chen, Teruaki Oka, Masato Mita, Mamoru Komachi. DejaVu: Disambiguation evaluation dataset for English-JApanese machine translation on VisUal information. PACLIC. 2024. [paper]
Kyotaro Nakajima, Hwichan Kim, Tosho Hirasawa, Taisei Enomoto, Zhousi Chen, Mamoru Komachi. A Survey for LLM Tuning Methods: Classifying Approaches Based on Model Internal Accessibility. PACLIC. 2024. [paper]
Hwichan Kim, Shota Sasaki, Sho Hoshino, Ukyo Honda. A Single Linear Layer Yields Task-Adapted Low-Rank Matrices. LREC-COLING. 2024. [paper] [poster]
Taisei Enomoto, Tosho Hirasawa, Hwichan Kim, Teruaki Oka, Mamoru Komachi. Simultaneous Domain Adaptation of Tokenization and Machine Translation. PACLIC. 2023. [paper]
Hwichan Kim, Mamoru Komachi. Enhancing Few-shot Cross-lingual Transfer with Target Language Peculiar Examples. Findings of ACL. 2023. [paper][poster]
Hiroto Tamura, Tosho Hirasawa, Hwichan Kim, Mamoru Komachi. Does Masked Language Model Pre-training with Artificial Data Improve Low-resource Neural Machine Translation? In Findings of EACL. 2023. [paper]
Hwichan Kim, Sangwhan Moon, Naoaki Okazaki, Mamoru Komachi. Learning How to Translate North Korean through South Korean. LREC. 2022. [paper][poster]
Hwichan Kim, Mamoru Komachi. Can Monolingual Pre-trained Encoder-Decoder Improve NMT for Distant Language Pairs? PACLIC. 2021. [paper][slide]
Hwichan Kim, Tosho Hirasawa, Mamoru Komachi. Zero-shot North Korean to English Neural Machine Translation by Character Tokenization and Phoneme Decomposition. ACL SRW. 2020. [paper] [slide]

International Conference/Workshop (Non-refereed)

Ayako Sato, Kyotaro Nakajima, Hwichan Kim, Zhousi Chen, Mamoru Komachi. TMUHIT’s Submission for the WMT24 Quality Estimation Shared Task: Is GPT-4 a Good Evaluator for Machine Translation? WMT24 Quality Estimation Shared Task. 2024. [paper]
Taisei Enomoto*, Hwichan Kim*, Tosho Hirasawa, Yoshinari Nagai, Ayako Sato, Kyotaro Nakajima, Mamoru Komachi. TMU-HIT at MLSP 2024: How Well Can GPT-4 Tackle Multilingual Lexical Simplification? MLSP Shared Task at BEA. 2024. [paper]
- Our system achieved the highest performance in 9 out of 10 languages at the MLSP shared task. * denotes equal contribution.
Hwichan Kim,Mamoru Komachi. TMU NMT System with Japanese BART for the Patent task of WAT 2021. Patent task of WAT. 2021. [paper][slide]
Hwichan Kim, Tosho Hirasawa, Mamoru Komachi. Korean-to-Japanese Neural Machine Translation System using Hanja Information. Patent task of WAT. 2020. [paper] [slide]

Journal

Hwichan Kim, Hirasawa Tosho, Sangwhan Moon, Naoaki Okazaki, Mamoru Komachi. North Korean Neural Machine Translation through South Korean Resources. ACM TALLIP. 2023. [paper]

Grants and scholarships

“Sotsui” Fellowship Program for Doctoral Students (April 2022-March 2025)

https://soutsui-f20.net/

The Korean Scholarship (April 2020-March 2022)

http://www.korean-s-f.or.jp/

Reviewer

Peer

2025: ACL Rolling Review
2024: ACL Rolling Review
2023: IJCNLP-AACL

Secondary

2022: AACL-IJCNLP, WMT

Page updated

Google Sites

Report abuse