Marieke van Erp
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Marieke van Erp has a total of 56 co-authors. Here are the top co-authors by number of shared publications:
- Valentina Pasqual
Department of Classical Philology and Italian Studies, Digital Humanities Advanced Research Center (/DH.arc), University of Bologna
4 shared publications - Aldo Gangemi
University of Bologna
3 shared publications - Bruno Sartini
Ludwig-Maximilians-Universität München, Munich, Germany
3 shared publications - Francesca Tomasi
Alma Mater Studiorum University of Bologna
3 shared publications Niels Dekker
Department of Computer Science, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands
3 shared publications- Tobias Kuhn
Vrije Universiteit Amsterdam
3 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 20 March 2026
- Published by Emerald Publishing in Journal of Documentation
- Vol. 82 (7) , 206-250
Knowledge graphs generation from cultural heritage texts: combining LLMs and ontological engineering for scholarly debates
Purpose: Cultural heritage (CH) texts contain rich knowledge that is difficult to query systematically due to the challenges of converting unstructured discourse into structured knowledge graphs (KGs). This paper introduces ATR4CH (Adaptive Text-to-RDF for Cultural Heritage), a systematic five-step methodology for Large Language Model (LLM)-based knowledge extraction from CH documents. We validate the methodology through a case study on authenticity assessment debates. Design/methodology/approach: ATR4CH combines annotation models, ontological frameworks and LLM-based extraction through iterative development: foundational analysis, annotation schema development, pipeline architecture, integration refinement and comprehensive evaluation. We demonstrate the approach using Wikipedia articles about disputed items (documents, artifacts, etc.), implementing a sequential pipeline with three LLMs (Claude Sonnet 3.7, Llama 3.3 70B and GPT-4o-mini). Findings: The methodology successfully extracts complex CH knowledge: 0.96–0.99 F1 for metadata extraction, 0.7–0.8 F1 for entity recognition, 0.65–0.75 F1 for hypothesis extraction, 0.95–0.97 for evidence extraction and 0.62 G-EVAL for discourse representation. Smaller models performed competitively, enabling cost-effective deployment. Research limitations/implications: The produced KG is limited to Wikipedia articles. While the results are encouraging, human oversight is necessary during post-processing. Originality/value: To the best of the authors’ knowledge, this is the first systematic methodology for coordinating LLM-based extraction with CH ontologies. ATR4CH provides a replicable framework adaptable across CH domains and institutional resources. ATR4CH enables CH institutions to systematically convert textual knowledge into queryable KGs, supporting automated metadata enrichment and knowledge discovery.
- 20 March 2026
- Published by Emerald Publishing in Journal of Documentation
- Vol. 82 (7) , 206-250
Top Cited PapersOpen AccessKnowledge graphs generation from cultural heritage texts: combining LLMs and ontological engineering for scholarly debates
Purpose: Cultural heritage (CH) texts contain rich knowledge that is difficult to query systematically due to the challenges of converting unstructured discourse into structured knowledge graphs (KGs). This paper introduces ATR4CH (Adaptive Text-to-RDF for Cultural Heritage), a systematic five-step methodology for Large Language Model (LLM)-based knowledge extraction from CH documents. We validate the methodology through a case study on authenticity assessment debates. Design/methodology/approach: ATR4CH combines annotation models, ontological frameworks and LLM-based extraction through iterative development: foundational analysis, annotation schema development, pipeline architecture, integration refinement and comprehensive evaluation. We demonstrate the approach using Wikipedia articles about disputed items (documents, artifacts, etc.), implementing a sequential pipeline with three LLMs (Claude Sonnet 3.7, Llama 3.3 70B and GPT-4o-mini). Findings: The methodology successfully extracts complex CH knowledge: 0.96–0.99 F1 for metadata extraction, 0.7–0.8 F1 for entity recognition, 0.65–0.75 F1 for hypothesis extraction, 0.95–0.97 for evidence extraction and 0.62 G-EVAL for discourse representation. Smaller models performed competitively, enabling cost-effective deployment. Research limitations/implications: The produced KG is limited to Wikipedia articles. While the results are encouraging, human oversight is necessary during post-processing. Originality/value: To the best of the authors’ knowledge, this is the first systematic methodology for coordinating LLM-based extraction with CH ontologies. ATR4CH provides a replicable framework adaptable across CH domains and institutional resources. ATR4CH enables CH institutions to systematically convert textual knowledge into queryable KGs, supporting automated metadata enrichment and knowledge discovery.
- Open Access
It's All About the Confidence: An Unsupervised Approach for Multilingual Historical Entity Linking using Large Language Models
Despite the recent advancements in NLP with the advent of Large Language Models (LLMs), Entity Linking (EL) for historical texts remains challenging due to linguistic variation, noisy inputs, and evolving semantic conventions. Existing solutions either require substantial training data or rely on domain-specific rules that limit scalability. In this paper, we present MHEL-LLaMo (Multilingual Historical Entity Linking with Large Language MOdels), an unsupervised ensemble approach combining a Small Language Model (SLM) and an LLM. MHEL-LLaMo leverages a multilingual bi-encoder (BELA) for candidate retrieval and an instruction-tuned LLM for NIL prediction and candidate selection via prompt chaining. Our system uses SLM's confidence scores to discriminate between easy and hard samples, applying an LLM only for hard cases. This strategy reduces computational costs while preventing hallucinations on straightforward cases. We evaluate MHEL-LLaMo on four established benchmarks in six European languages (English, Finnish, French, German, Italian and Swedish) from the 19th and 20th centuries. Results demonstrate that MHEL-LLaMo outperforms state-of-the-art models without requiring fine-tuning, offering a scalable solution for low-resource historical EL. The implementation of MHEL-LLaMo is available on Github.
Open AccessIt's All About the Confidence: An Unsupervised Approach for Multilingual Historical Entity Linking using Large Language Models
Despite the recent advancements in NLP with the advent of Large Language Models (LLMs), Entity Linking (EL) for historical texts remains challenging due to linguistic variation, noisy inputs, and evolving semantic conventions. Existing solutions either require substantial training data or rely on domain-specific rules that limit scalability. In this paper, we present MHEL-LLaMo (Multilingual Historical Entity Linking with Large Language MOdels), an unsupervised ensemble approach combining a Small Language Model (SLM) and an LLM. MHEL-LLaMo leverages a multilingual bi-encoder (BELA) for candidate retrieval and an instruction-tuned LLM for NIL prediction and candidate selection via prompt chaining. Our system uses SLM's confidence scores to discriminate between easy and hard samples, applying an LLM only for hard cases. This strategy reduces computational costs while preventing hallucinations on straightforward cases. We evaluate MHEL-LLaMo on four established benchmarks in six European languages (English, Finnish, French, German, Italian and Swedish) from the 19th and 20th centuries. Results demonstrate that MHEL-LLaMo outperforms state-of-the-art models without requiring fine-tuning, offering a scalable solution for low-resource historical EL. The implementation of MHEL-LLaMo is available on Github.
- Open Access
- 21 November 2025
- Published by The Association for Computers and the Humanities
- Vol. 3, 1310-1322
Tracing Colonial Discourse in Dutch Historical Newspapers
The expansion of colonialism often involves categorisations of colonised people based on their race, language, social status, and perceived level of “civilisation”. While such categorisations may seem fixed, the language used to describe colonised populations is dynamic, shifting in meaning and connotation as colonial relationships and power structures change over time. We investigate diachronic semantic changes in Dutch colonial terminology through newspaper articles from 1860 to 1960 using word embeddings. This period witnessed major colonial expansion, the implementation of “ethical imperialism” (the Dutch “Ethical Policy” promoting education and welfare while maintaining colonial control), and eventual decolonisation. We combine computational semantic change detection with a more fine-grained analysis of the associated adjectives, demonstrating how the combination of distant and closer techniques can reveal patterns of linguistic transformation that reflect broader societal and political changes in Dutch colonial discourse.
- 21 November 2025
- Published by The Association for Computers and the Humanities
- Vol. 3, 1310-1322
Open AccessTracing Colonial Discourse in Dutch Historical Newspapers
The expansion of colonialism often involves categorisations of colonised people based on their race, language, social status, and perceived level of “civilisation”. While such categorisations may seem fixed, the language used to describe colonised populations is dynamic, shifting in meaning and connotation as colonial relationships and power structures change over time. We investigate diachronic semantic changes in Dutch colonial terminology through newspaper articles from 1860 to 1960 using word embeddings. This period witnessed major colonial expansion, the implementation of “ethical imperialism” (the Dutch “Ethical Policy” promoting education and welfare while maintaining colonial control), and eventual decolonisation. We combine computational semantic change detection with a more fine-grained analysis of the associated adjectives, demonstrating how the combination of distant and closer techniques can reveal patterns of linguistic transformation that reflect broader societal and political changes in Dutch colonial discourse.
- Open Access
Knowledge Graphs Generation from Cultural Heritage Texts: Combining LLMs and Ontological Engineering for Scholarly Debates
Cultural Heritage texts contain rich knowledge that is difficult to query systematically due to the challenges of converting unstructured discourse into structured Knowledge Graphs (KGs). This paper introduces ATR4CH (Adaptive Text-to-RDF for Cultural Heritage), a systematic five-step methodology for Large Language Model-based Knowledge Extraction from Cultural Heritage documents. We validate the methodology through a case study on authenticity assessment debates. Methodology - ATR4CH combines annotation models, ontological frameworks, and LLM-based extraction through iterative development: foundational analysis, annotation schema development, pipeline architecture, integration refinement, and comprehensive evaluation. We demonstrate the approach using Wikipedia articles about disputed items (documents, artifacts...), implementing a sequential pipeline with three LLMs (Claude Sonnet 3.7, Llama 3.3 70B, GPT-4o-mini). Findings - The methodology successfully extracts complex Cultural Heritage knowledge: 0.96-0.99 F1 for metadata extraction, 0.7-0.8 F1 for entity recognition, 0.65-0.75 F1 for hypothesis extraction, 0.95-0.97 for evidence extraction, and 0.62 G-EVAL for discourse representation. Smaller models performed competitively, enabling cost-effective deployment. Originality - This is the first systematic methodology for coordinating LLM-based extraction with Cultural Heritage ontologies. ATR4CH provides a replicable framework adaptable across CH domains and institutional resources. Research Limitations - The produced KG is limited to Wikipedia articles. While the results are encouraging, human oversight is necessary during post-processing. Practical Implications - ATR4CH enables Cultural Heritage institutions to systematically convert textual knowledge into queryable KGs, supporting automated metadata enrichment and knowledge discovery.
Open AccessKnowledge Graphs Generation from Cultural Heritage Texts: Combining LLMs and Ontological Engineering for Scholarly Debates
Cultural Heritage texts contain rich knowledge that is difficult to query systematically due to the challenges of converting unstructured discourse into structured Knowledge Graphs (KGs). This paper introduces ATR4CH (Adaptive Text-to-RDF for Cultural Heritage), a systematic five-step methodology for Large Language Model-based Knowledge Extraction from Cultural Heritage documents. We validate the methodology through a case study on authenticity assessment debates. Methodology - ATR4CH combines annotation models, ontological frameworks, and LLM-based extraction through iterative development: foundational analysis, annotation schema development, pipeline architecture, integration refinement, and comprehensive evaluation. We demonstrate the approach using Wikipedia articles about disputed items (documents, artifacts...), implementing a sequential pipeline with three LLMs (Claude Sonnet 3.7, Llama 3.3 70B, GPT-4o-mini). Findings - The methodology successfully extracts complex Cultural Heritage knowledge: 0.96-0.99 F1 for metadata extraction, 0.7-0.8 F1 for entity recognition, 0.65-0.75 F1 for hypothesis extraction, 0.95-0.97 for evidence extraction, and 0.62 G-EVAL for discourse representation. Smaller models performed competitively, enabling cost-effective deployment. Originality - This is the first systematic methodology for coordinating LLM-based extraction with Cultural Heritage ontologies. ATR4CH provides a replicable framework adaptable across CH domains and institutional resources. Research Limitations - The produced KG is limited to Wikipedia articles. While the results are encouraging, human oversight is necessary during post-processing. Practical Implications - ATR4CH enables Cultural Heritage institutions to systematically convert textual knowledge into queryable KGs, supporting automated metadata enrichment and knowledge discovery.
- Open Access
Explicit vs. Implicit Biographies: Evaluating and Adapting LLM Information Extraction on Wikidata-Derived Texts
Text Implicitness has always been challenging in Natural Language Processing (NLP), with traditional methods relying on explicit statements to identify entities and their relationships. From the sentence "Zuhdi attends church every Sunday", the relationship between Zuhdi and Christianity is evident for a human reader, but it presents a challenge when it must be inferred automatically. Large language models (LLMs) have proven effective in NLP downstream tasks such as text comprehension and information extraction (IE). This study examines how textual implicitness affects IE tasks in pre-trained LLMs: LLaMA 2.3, DeepSeekV1, and Phi1.5. We generate two synthetic datasets of 10k implicit and explicit verbalization of biographic information to measure the impact on LLM performance and analyze whether fine-tuning implicit data improves their ability to generalize in implicit reasoning tasks. This research presents an experiment on the internal reasoning processes of LLMs in IE, particularly in dealing with implicit and explicit contexts. The results demonstrate that fine-tuning LLM models with LoRA (low-rank adaptation) improves their performance in extracting information from implicit texts, contributing to better model interpretability and reliability.
Open AccessExplicit vs. Implicit Biographies: Evaluating and Adapting LLM Information Extraction on Wikidata-Derived Texts
Text Implicitness has always been challenging in Natural Language Processing (NLP), with traditional methods relying on explicit statements to identify entities and their relationships. From the sentence "Zuhdi attends church every Sunday", the relationship between Zuhdi and Christianity is evident for a human reader, but it presents a challenge when it must be inferred automatically. Large language models (LLMs) have proven effective in NLP downstream tasks such as text comprehension and information extraction (IE). This study examines how textual implicitness affects IE tasks in pre-trained LLMs: LLaMA 2.3, DeepSeekV1, and Phi1.5. We generate two synthetic datasets of 10k implicit and explicit verbalization of biographic information to measure the impact on LLM performance and analyze whether fine-tuning implicit data improves their ability to generalize in implicit reasoning tasks. This research presents an experiment on the internal reasoning processes of LLMs in IE, particularly in dealing with implicit and explicit contexts. The results demonstrate that fine-tuning LLM models with LoRA (low-rank adaptation) improves their performance in extracting information from implicit texts, contributing to better model interpretability and reliability.
- Open Access
Structuring Authenticity Assessments on Historical Documents using LLMs
Given the wide use of forgery throughout history, scholars have and are continuously engaged in assessing the authenticity of historical documents. However, online catalogues merely offer descriptive metadata for these documents, relegating discussions about their authenticity to free-text formats, making it difficult to study these assessments at scale. This study explores the generation of structured data about documents' authenticity assessment from natural language texts. Our pipeline exploits Large Language Models (LLMs) to select, extract and classify relevant claims about the topic without the need for training, and Semantic Web technologies to structure and type-validate the LLM's results. The final output is a catalogue of documents whose authenticity has been debated, along with scholars' opinions on their authenticity. This process can serve as a valuable resource for integration into catalogues, allowing room for more intricate queries and analyses on the evolution of these debates over centuries.
Open AccessStructuring Authenticity Assessments on Historical Documents using LLMs
Given the wide use of forgery throughout history, scholars have and are continuously engaged in assessing the authenticity of historical documents. However, online catalogues merely offer descriptive metadata for these documents, relegating discussions about their authenticity to free-text formats, making it difficult to study these assessments at scale. This study explores the generation of structured data about documents' authenticity assessment from natural language texts. Our pipeline exploits Large Language Models (LLMs) to select, extract and classify relevant claims about the topic without the need for training, and Semantic Web technologies to structure and type-validate the LLM's results. The final output is a catalogue of documents whose authenticity has been debated, along with scholars' opinions on their authenticity. This process can serve as a valuable resource for integration into catalogues, allowing room for more intricate queries and analyses on the evolution of these debates over centuries.
- Open Access
- 5 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 223-224
Unflattening Knowledge Graphs
Large general-purpose knowledge graphs (KGs) are a critical component for knowledge-driven applications. However, most KGs represent only a limited view of the entities and concepts they describe. The concept coffee can, for example, refer to the plant that yields coffee seeds, the beverage ‘coffee’, and the activity of drinking the beverage. Moreover, it has a long history that is deeply connected to colonialism and status. All of these notions are an intricate part of national identities, have changed dramatically over time, and connect to many different narratives with different opinions on them. This complexity is not captured in current KGs. In this vision paper, I present the three crucial challenges for unflattening knowledge graphs and directions for future work.
- 5 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 223-224
Open AccessUnflattening Knowledge Graphs
Large general-purpose knowledge graphs (KGs) are a critical component for knowledge-driven applications. However, most KGs represent only a limited view of the entities and concepts they describe. The concept coffee can, for example, refer to the plant that yields coffee seeds, the beverage ‘coffee’, and the activity of drinking the beverage. Moreover, it has a long history that is deeply connected to colonialism and status. All of these notions are an intricate part of national identities, have changed dramatically over time, and connect to many different narratives with different opinions on them. This complexity is not captured in current KGs. In this vision paper, I present the three crucial challenges for unflattening knowledge graphs and directions for future work.
- Top Cited PapersOpen Access
- 9 August 2023
- Published by Association for Computing Machinery (ACM) in Journal on Computing and Cultural Heritage
- Vol. 16 (3) , 1-38
ICON: An Ontology for Comprehensive Artistic Interpretations
In this work, we introduce ICON, an ontology that models artistic interpretations of artworks’ subject matter (i.e., iconographies) and meanings (i.e., symbols, iconological aspects). Developed by conceptualizing authoritative knowledge and notions taken from Panofsky’s levels of interpretation theory, ICON ontology focuses on the granularity of interpretations. It can be used to describe an interpretation of an artwork from the pre-iconographical, icongraphical, and iconological levels. Its main classes have been aligned to ontologies that come from the domains of cultural descriptions (ArCo, CIDOC-CRM, VIR), semiotics (DOLCE), bibliometrics (CITO), and symbolism (Simulation Ontology), to grant a robust schema that can be extendable using additional classes and properties coming from these ontologies. The ontology was evaluated through competency questions that range from simple recognition on a specific level of interpretation to complex scenarios. Data written using this model was compared to state-of-the-art ontologies and schemas to both highlight the current lack of a domain-specific ontology on art interpretation and show how our work fills some of the current gaps. The ontology is openly available and compliant with FAIR principles. With our ontology, we hope to encourage digital art historians working for cultural institutions in making more detailed linked open data about the content of their artifacts, to exploit the full potential of Semantic Web in linking artworks through not only subjects and common metadata but also specific symbolic interpretations, intrinsic meanings, and the motifs through which their subjects are represented. Additionally, by basing our work on theories made by different art history scholars in the last century, we make sure that their knowledge and studies will not be lost in the transition to the digital, linked open data era.
- 9 August 2023
- Published by Association for Computing Machinery (ACM) in Journal on Computing and Cultural Heritage
- Vol. 16 (3) , 1-38
Top Cited PapersOpen AccessICON: An Ontology for Comprehensive Artistic Interpretations
In this work, we introduce ICON, an ontology that models artistic interpretations of artworks’ subject matter (i.e., iconographies) and meanings (i.e., symbols, iconological aspects). Developed by conceptualizing authoritative knowledge and notions taken from Panofsky’s levels of interpretation theory, ICON ontology focuses on the granularity of interpretations. It can be used to describe an interpretation of an artwork from the pre-iconographical, icongraphical, and iconological levels. Its main classes have been aligned to ontologies that come from the domains of cultural descriptions (ArCo, CIDOC-CRM, VIR), semiotics (DOLCE), bibliometrics (CITO), and symbolism (Simulation Ontology), to grant a robust schema that can be extendable using additional classes and properties coming from these ontologies. The ontology was evaluated through competency questions that range from simple recognition on a specific level of interpretation to complex scenarios. Data written using this model was compared to state-of-the-art ontologies and schemas to both highlight the current lack of a domain-specific ontology on art interpretation and show how our work fills some of the current gaps. The ontology is openly available and compliant with FAIR principles. With our ontology, we hope to encourage digital art historians working for cultural institutions in making more detailed linked open data about the content of their artifacts, to exploit the full potential of Semantic Web in linking artworks through not only subjects and common metadata but also specific symbolic interpretations, intrinsic meanings, and the motifs through which their subjects are represented. Additionally, by basing our work on theories made by different art history scholars in the last century, we make sure that their knowledge and studies will not be lost in the transition to the digital, linked open data era.
- 22 May 2023
- Published by Springer Nature
- p. 502-519
A Knowledge Graph of Contentious Terminology for Inclusive Representation of Cultural Heritage
- 22 May 2023
- Published by Springer Nature
- p. 502-519
A Knowledge Graph of Contentious Terminology for Inclusive Representation of Cultural Heritage
- Open Access
- 30 April 2023
- Published by Association for Computing Machinery (ACM)
- p. 660
2nd International Workshop on Multisensory Data and Knowledge (MDK)
- 30 April 2023
- Published by Association for Computing Machinery (ACM)
- p. 660
Open Access2nd International Workshop on Multisensory Data and Knowledge (MDK)