HIPE-2026 is a CLEF lab evaluating person-place relation extraction (at/isAt) from noisy multilingual historical texts.
HIPE-2026 is a CLEF evaluation lab focused on extracting person–place relations from noisy, multilingual historical texts, advancing research in semantic relation extraction for natural language processing (NLP) and artificial intelligence (AI). The task centers on identifying two types of temporal relations: at, which determines whether a person was ever at a given place prior to the document's publication date, and isAt, which assesses whether the person was present at the location around the time of publication. These distinctions require systems to perform temporal reasoning and geographical inference based on often sparse or indirect textual cues.
The evaluation framework includes a three-fold profile that jointly assesses accuracy, computational efficiency, and domain generalization, promoting the development of robust and scalable NLP methods applicable to real-world historical data. This makes HIPE-2026 particularly relevant to AI research, as it challenges both large language models (LLMs) and traditional classification systems to go beyond simple entity co-occurrence and engage in abductive interpretation—inferring plausible relationships from contextual evidence. For instance, if a text states that a person delivered a speech at an event held in a specific location, the system must infer presence (at = true, isAt = +) even without explicit mention of physical attendance.
Building on previous editions HIPE-2020 and HIPE-2022—which addressed named entity recognition and linking in historical newspapers—the 2026 edition shifts focus toward semantic relation extraction, supporting applications in digital humanities such as knowledge graph construction, biography reconstruction, and spatio-temporal analysis. Training and test data are drawn from historical newspapers in English, German, French, and Luxembourgish, with entity pairs pre-identified to allow participants to focus on relation classification.
Registration for the lab remains open until April 23, 2026, with the evaluation period scheduled for May 5–7, 2026, and results to be discussed at the CLEF conference in Jena, Germany, from September 21–24, 2026. The inclusion of a surprise test set from a different domain further evaluates model generalization, reinforcing the goal of building adaptable, real-world NLP systems.
The CLEF HIPE-2026 lab establishes a rigorous benchmark for the extraction of semantic relations, specifically identifying at and isAt links between persons and places, within noisy, multilingual historical documents. Moving beyond standard Named Entity Recognition (NER), this shared task targets the more complex challenge of Relation Extraction (RE) in a domain characterized by significant data quality issues, such as Optical Character Recognition (OCR) errors, archaic syntax, and linguistic variance across multiple languages. The lab provides a standardized evaluation framework and dataset, challenging participants to develop systems that can accurately map geographic mobility and residence despite the inherent noise and structural inconsistencies of digitized historical archives.
A key contribution of this work is the emphasis on evaluating both the accuracy and efficiency of proposed models. By focusing on historical texts, the initiative addresses the scarcity of benchmarks for low-resource and noisy domains, encouraging the development of robust architectures that do not rely solely on massive, clean datasets like Wikipedia. The insights generated from this competition are expected to drive advancements in handling linguistic drift and orthographic noise, providing a valuable testing ground for modern NLP techniques—ranging from large language models to more lightweight, efficient classifiers.
This research is significant because it bridges the gap between cutting-edge AI and the specific needs of the Digital Humanities. Accurate relation extraction from historical texts enables large-scale semantic analysis of cultural heritage, facilitating applications in automated genealogy, historical mapping, and prosopography. Furthermore, by pushing the boundaries of NLP performance on noisy, multilingual data, HIPE-2026 contributes to the broader goal of creating more generalizable and resilient AI systems capable of understanding and processing the long tail of human linguistic production.
The CLEF HIPE-2026 lab focuses on advancing person-place relation extraction (PPRE)—specifically identifying (at/isAt) relationships—from noisy, multilingual historical texts. Unlike modern NLP datasets, historical texts pose unique challenges due to linguistic variability, OCR errors, archaic terminology, and sparse annotations. The lab provides a benchmark for evaluating models on these tasks, encouraging research in robust relation extraction under adversarial conditions. Participating systems are evaluated on metrics like precision, recall, and efficiency, with a focus on scalability given the large-scale nature of historical corpora.
Key contributions of this work include: 1. A curated multilingual historical dataset with manually annotated person-place relations, enabling cross-lingual and noise-robust NLP research. 2. Evaluation protocols that account for OCR noise and linguistic diversity, pushing models to generalize beyond clean, contemporary text. 3. Insights into model performance under real-world constraints, highlighting gaps in current state-of-the-art (SOTA) systems for historical data.
This research matters because it addresses a critical gap in semantic NLP for digital humanities, where accurate PPRE is essential for historical event reconstruction, biographical analysis, and cultural heritage preservation. By benchmarking models on noisy, multilingual data, HIPE-2026 accelerates progress in low-resource and adversarial NLP, with implications for information extraction in legal, medical, and archival domains.
Source: [arXiv:2602.17663](https://arxiv.org/abs/2602.17663)