The virtual K-centre Large Language Models in the Social Sciences and Humanities (LLMs4SSH) provides researchers from various disciplines with expertise and technical support facilitating the effective use of LLMs within the SS&H environment. Its main focus lies in supplying accurate guidance on the application of LLMs as well as assisting in the selection of models suited to the specific needs of individual research goals. The centre also enables the access to other computational resources within its available capacities at the Wrocław Centre for Networking and Supercomputing infrastructure (WCSS). 

Researchers from various disciplines may benefit from the resources of the K-centre. The most prominent audiences served are as followed (this is however an incomplete list):

• Computational linguists
• Digital humanities scholars
• Lecturers
• Psychologists
• Sociologists
• State institutions
• Students

Swiss contribution

LLMs4SSH is in part hosted by CLARIN-CH. Members from Swiss Institutions lend their expertise to the centre:

From UZH, Prof. Rico Sennrich and Dr. Jannis Vamvas take part in the endeavours of the K-centre assisting researchers in effectively making use of LLMs. The former supports tasks concerning multilinguality (such as cross-lingual transfer and translation), multimodality (for text, speech and vision), low-resource adaptation of LLMs and LLM interpretability, whereas the latter covers the subjects of text generation algorithms (for example sampling, self-consistency/minimum bayesian risk, or constrained text generation), the Hugging Face transformers framework (contributing new models, model architectures, using or contributing to the code base, or parameter-efficient fine-tuning (PEFT)), using LLMs for translation or translation evaluation. Additionally, Prof. Noah Bubenhofer supports the team with expertise on AI literacy (research and analysis), language theory (in the context of generative AI) as well as the application of AI and LLMs, specifically for research methods in corpus linguistics and the adaptation of said methods to individual needs. Dr. Simon Clematide contributes proficiency in the application of LLMs to large-scale classification problems (such as mapping problems with large ontologies), support for low-resource languages, historical texts, and digitised texts that contain OCR errors and domain adaptation methods for LLMs and text embedding models.

From USI, Prof. Lonneke van der Plas provides expertise to researchers working with the foundations of LLMs, multilingual LLMs, as well as testing and improving cognitive abilities of LLMs (for example creative story writing, or analogical reasoning) and also ethical considerations. Prof. Andrea Rocci supplies expertise concerning the application of LLMs to pragmatic analysis of text and speech transcripts in financial communication, argumentation mining, and analysis of causality in online climate change discourse.

For a more detailed explanation of the K-centre visit its website: