On October 20th, 2026, Sotaro Takeshita successfully defended his doctoral thesis “Scholarly Document Summarization: from Foundation to Application” at the University of Mannheim, with the highest distinction: summa cum laude. Congratulations, Sotaro!
The rapidly growing number of scholarly publications makes it ever harder for researchers to find the papers relevant to them — a burden that weighs heaviest on those with less experience. Sotaro’s thesis approaches this problem through automatic summarization and contributes to four challenges that still hold the field back:
- Rigorousness: reliable evaluation, with high-quality summarization datasets built for research purposes, such as ACLSum, and a new evaluation metric that gives finer-grained information on summary quality than existing ones, ROUGE-K.
- Robustness: benchmark studies that systematically compare approaches for different summarization scenarios, for instance cross-lingual summarization of English papers into other languages, along with new techniques and model architectures that outperform existing models.
- Efficiency: distillation to shrink summarization models without sacrificing much quality, and multi-task models that produce more than just the summary, such as IRSum, which combines summarization and retrieval in one model.
- Accessibility: a development suite for cross-lingual summarization into four languages, and two web-based systems, among them the paper explorer GenGO, that make powerful summarization models usable for researchers who cannot set them up themselves.
The thesis was co-supervised by Kai Eckert and Simone Paolo Ponzetto (University of Mannheim). The committee was completed by Michael Strube (Heidelberg Institute for Theoretical Studies), Heiko Paulheim and Margret Keuper (both University of Mannheim). Many thanks to all of them!