AI-based Methods for the Humanities

Advancing AI Research in the Humanities: A Workshop for Scholars at the Intersection of Digital Humanities and Machine Learning

The intersection of artificial intelligence (AI) and the humanities is an emerging frontier for research, promising transformative approaches to understanding and modeling historical, cultural, and linguistic phenomena. This workshop invites scholars and practitioners from diverse disciplines to explore innovative AI-based methodologies that address core challenges in the humanities. The goal is to foster dialogue, share methodological developments, insights, and propose new paradigms for using AI in humanistic research, emphasizing both interpretability and application.

The workshop will be held in Berlin from September 23 to 24, 2025. It is organized around four thematic sections, each focusing on specific methodological approaches and their potential applications to evolving processes. Particular emphasis is therefore placed on the subject of time and temporality. Temporal aspects—whether in modeling historical change, dynamic cultural systems, or the evolution of language—are especially relevant and are prioritized across all four areas of interest. By centering time as a key axis, the workshop seeks to encourage submissions that delve into not only static patterns but also processes, transformations, and trajectories, as well as the challenges associated with evaluating and verifying AI-based insights.

Confirmed Keynote Speakers

Section 1: Explainable Al and lnsights from lnterpretability

Anders Søgaard

University of Copenhagen

Section 2: Historical Networks and Cultural Dynamics

Ingo Scholtes

Universität Würzburg

Section 3: Modeling Language and Low-Resource Humanities Data

Seid Muhie Yimam

University of Hamburg

Section 4: Foundation Models for the Humanities

John Pavlopoulos

Athens University of Economics and Business

Call for Papers

We encourage submissions that address theoretical advancements, applied case studies, or critical perspectives across these themes. This workshop aims to provide a platform for interdisciplinary collaboration, fostering connections between AI researchers and humanities scholars to chart the future of this exciting domain. Submissions that emphasize temporal aspects in any of the above areas are especially valued.

Submissions will be evaluated based on their originality, relevance to the workshop themes, and potential to advance the field. We look forward to receiving your innovative contributions and engaging in meaningful discussions at the workshop.

Submission Details

Deadline: April 1, 2025 (rolling evaluations until slots are filled)
Format: Proposal (maximum 1 page)
Submission Email: valleriani@mpiwg-berlin.mpg.de
Travel & Accommodation: Reimbursed for selected speakers

Atomization-recomposition framework for model learning under sparse annotation settings developed for the analysis of a large corpus of unknown early modern astronomical computational tables extracted from the Sphaera corpus. DOI: 10.1126/sciadv.adj1719 — © BIFOLD

About the Workshop

This workshop is organized around four thematic sections, each focusing on specific methodological approaches and their potential applications. A key focus is on time and temporality, emphasizing the modeling of historical change, cultural dynamics, and linguistic evolution. We encourage submissions that explore dynamic processes, transformations, and trajectories in AI-based humanities research.

CorDeep. A classifier for identifying and labeling visual elements in early modern scientific treatises. DOI: 10.3390/jimaging8100285 — © Österreichische Nationalbibliothek & Library of the MPI for the History of Science

1. Explainable AI (XAI) and Insights from Interpretability
Understanding the “why” behind AI decisions is as critical as the “what,” especially in the humanities, where interpretability is paramount. This section explores the role of XAI techniques in providing insights into machine learning (ML) models applied to humanistic data. How can we adapt XAI methods to ensure that AI-driven findings align with the interpretive needs of historians, linguists, and cultural scholars? Topics may include interpretability techniques for complex models, cultural considerations in AI-driven interpretations, and case studies showcasing the integration of XAI in humanities research, with an emphasis on temporal trends and insights.

2. Historical Networks and Cultural Dynamics
The analysis of historical networks and cultural systems demands methods that can capture the interplay of diverse, temporally dynamic factors. We invite contributions that leverage graph-based approaches such as Graph Neural Networks (GNNs) or temporal extensions (TGNNs) to model and explore these systems. Inspired by emerging research on agentic systems—where ML models act as agents collaborating and communicating—this section considers novel applications of these paradigms. Could historical knowledge production be reframed as a collaborative process between human historians and AI agents, akin to modern collaborative software production? Submissions that address the representation and evolution of temporal relationships in historical and cultural dynamics are particularly encouraged.

Multiplex network of the editions of the Sphaera corpus organized in three different layers. In each layer, books are connected on the basis of a specific semantic relation. Visualization realized by means of muxViz (muxViz.net). DOI: 10.1038/s41598-020-76916-3 — © BIFOLD

3. Modeling Language and Low-Resource Humanities Data
Humanities data is often sparse, incomplete, or embedded in low-resource languages, posing unique challenges for AI modeling. Significant phenomena frequently occur in the long tail of data distributions, making it difficult to detect, model, and generalize rare patterns that traditional methods often overlook. This section highlights innovative, sample-efficient approaches for processing, analyzing, and interpreting such data, with a focus on NLP and machine learning strategies. Contributions may include novel techniques for encoding historical language structures, domain adaptation for low-resource scenarios, or methods for constructing meaningful models from fragmentary datasets. Approaches that incorporate temporal dimensions—such as language change over time or diachronic analysis of texts and records—are particularly welcome.

4. Foundation Models for the Humanities
Foundation models, with their unprecedented scale and generalization capabilities, are transforming AI research. Yet, their applicability to the humanities remains underexplored. How can these models be adapted to handle domain-specific data such as historical texts or cultural artifacts? Can they provide meaningful insights into questions of humanistic significance, or do their limitations outweigh their potential? This section invites contributions that assess the utility of foundation models in humanities research, with a focus on temporal challenges such as the handling of chronologically distributed data or understanding long-term historical trends.