HIPE

Based on a pilot study that confirmed the feasibility of the task, HIPE-2026 targets a single but fundamental relation type (person–isAt–place) and additionally requires participants to 1) determine the temporal scope of this relation, and 2) assess the textual evidence that supports it. Working with challenging materials – i.e OCR-noisy, multilingual and domain-diverse newspaper articles – participants will contribute to the development of approaches that are key to constructing historical knowledge graphs, reconstructing biographies, enabling spatial analysis, and advancing text understanding of historical material. Given the energy costs of frontier models and the need to process large-scale cultural heritage collections, we identify efficiency as a critical challenge. HIPE-2026 will therefore offer two sub-tracks: one targeting maximum accuracy, the other prioritizing a trade-off between accuracy and computational efficiency. A surprise dataset will be included to evaluate generalization across domains. All datasets will be released to support transparency, reuse and further research.

Organizers

• Juri Opitz (UZH Zurich)
• Emanuela Boros (EPFL)
• Andrianos Michail (UZH Zurich)
• Matteo Romanello (UZH Zurich)
• Maud Ehrmann (EPFL)
• Simon Clematide (UZH Zurich)