Project development and proposal support
We offer several forms of support to help digital humanities and social science scholars develop projects.
Methodological support. We assist project applicants with methodological services, including expert guidance on identifying and clearly formulating scientific challenges related to AI in DH. As AI components are increasingly common in DH projects, specialised expertise is crucial at the project development stage. This includes consultation on the realistic possibilities, limitations, and challenges of applying AI to specific tasks and datasets.
Proposal writing support. We can provide dedicated support for proposal preparation, covering both scientific and non-scientific components of applications through structured guidance and proposal reviews. This support may also include workshops and mentorship programmes.
We also welcome collaboration through our visiting research opportunities, which allow applicants to spend time with us in Ljubljana and develop their project proposals with our support (see our open call). We encourage you to review our current research topics and AI methodologies and to contact us with your proposed topic and specific needs.
Please note that the AI4DH CoE is expected to participate in the proposal as a funded partner or to host the applicant (e.g. for ERC or MSCA and ERA fellowships).
Current research topics
The Centre is initially focusing on seven interdependent infrastructure and research challenges:
- Natural language processing tools for historical newspaper analysis
- Comparative analysis of historical newspaper collections
- Producing semantics data and evaluation datasets from lexical resources
- Injecting knowledge into large language models
- AI infrastructure for data-driven folkloristics
- Explanation of generative large language models
- Tools for social science analyses of news media in less-resourced languages
Please note that we are open to other topics.
Methodological approaches
We offer our AI expertise to advise scholars in project proposals about how to use AI for digital humanities and social science research.
The technological underpinning of our methodological approaches are large language models (LLMs), vision language models (VLMs), and machine learning models, as well as application and fine-tuning of other multimodal foundation models. Our approaches also take into account ethics, bias and explainability. We can for instance help you with the following:
- VLMs: integrating both visual and textual data, enabling image-captioning, text-to-image generation, and multimodal analyses of cultural artefacts.
- Machine Learning Techniques: Supervised, unsupervised, and reinforcement learning approaches used to uncover patterns in (historical) data, enhance text classification, clustering, and predictive analysis in DH. Models such as Convolutional Neural Networks (CNNs) and Transformers to analyse images, manuscripts, and visual data. This includes image classification, object detection and segmentation, Optical Character Recognition (OCR) for digitising (historical) texts, analysing artworks, and studying multimedia sources.
- Advanced qualitative data analysis tools: integrated with LLMs and ML models for automatic text coding, theme extraction, sentiment analysis, topic modelling, statistical text analysis and multilingual support.
- Knowledge Graphs: creating interconnected networks of concepts extracted from texts for enhanced knowledge representation and for detecting and understanding abstract concepts such as conflict resolution, justice, rituals, and symbolic meanings within texts or visual media. This can be with custom LLM models trained on domain-specific data to identify these concepts accurately.
- Ongoing analysis of ethical issues in LLMs and AI applications: including fairness, transparency, and accountability, including research in identifying and mitigating gender and cultural biases in AI models, particularly LLMs.
- Development of explainable AI techniques: ensuring that model decisions are transparent and understandable for DH scholars.
- Temporal analysis methods: For comparing linguistic styles, cultural motifs, or historical narratives across different time spans. Techniques include diachronic language modelling, diachronic embeddings, trend analysis, and change detection in visual and textual datasets.
- Investigating philosophical implications of LLMs: Including their knowledge representation, reasoning capabilities, and the nature of their “understanding.” Critical analysis of LLMs in terms of knowledge ethics, epistemology, and their impact on humanistic research.