Naoki Yoshinaga

Naoki Yoshinaga is a tenured Associate Professor at the Institute of Industrial Science, The University of Tokyo, with extensive experience in natural language processing and computational linguistics. He has held academic positions since 2008 and currently leads research on pragmatic NLP models and multilingual systems.

• PhD in Computer Science, The University of Tokyo (2005-2008)
• MSc in Information Science (2000-2002)
• BSc in Information Science (1996-2000)

His research focuses on mechanistic interpretability in NLP models, multilingual/multimodal NLP, and efficient model design using trie structures and conjunctive features. He also investigates knowledge acquisition from social data and evaluation metrics for language generation.

Recent publications include work on neuron empirical gradient analysis (ACL-25), multilingual knowledge representation (EACL-24), and compact embedding methods (CoNLL-24). His research has been funded by multiple grants, including the University of Tokyo Excellent Young Researcher program and JSPS fellowships.

• Committee Special Award, Association for NLP (2023)
• JSAI SIG Research Award (2022)
• Best Interactive Award, DEIM Forum (2019, 2016)

He developed widely-adopted NLP tools like pecco (fast classification library), RenTAL (LTAG-to-HPSG grammar converter), and J.DepP (Japanese dependency parser). His lab emphasizes strong equivalence in formalism comparisons and pragmatic model design.