While current large language models, such as ChatGPT and Gemini, can produce text that appears human-like, they often miss the underlying linguistic nuances that connect different phrasings of the same idea. A direct example of this limitation is the inability of such models to reliably explain why and how two text segments are similar, or the linguistic characteristics that distinguish similar texts. This project aims to give AI a deeper understanding of the text it generates and manipulates. 

The key innovation of the project lies in the concept of “paraphrase types.”  Instead of treating paraphrasing as a simple binary choice (either two sentences mean the same thing or they don’t) Dr. Ruas proposes shifting the paradigm in paraphrase generation and detection to a more granular one using specific linguistic changes. These can range from simple word swaps (lexicon) to altering sentence structure (syntax), to more complex changes such as switching from active to passive voice (diathesis).  “This approach has great practical relevance because a better understanding of AI generation methods could, among other things, revolutionize plagiarism detection,” says Dr. Jan Philip Wahle, who wrote his dissertation on the topic and now works as a project leader in the research group.

The project team will collaborate with a team of researchers in Germany and abroad, including researchers from the University of Göttingen, GWDG, University of Wuppertal, LMU Munich, the National Institute of Informatics in Japan, Federal University of ABC and University Center FEI in Brazil, and the National Research Council in Canada. 

Press release by Göttinger Tageblatt: https://www.goettinger-tageblatt.de/beruf-und-bildung/regional/wer-schrieb-den-text-wirklich-goettinger-projekt-sucht-versteckte-muster-von-ki-und-menschen-GBBROHMF7ZA5NPZ4CYMZ4LIUGQ.html

Der Beitrag DFG Funds Project on Paraphrase Types and Plagiarism Detection erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

AI-generated media has significantly expanded the scope of digital crime. Highly realistic deepfake videos, cloned voices, and automated disinformation campaigns are now frequently used in fraud, identity theft, impersonation, and digital extortion. As a result, police work is increasingly shifting into the digital domain. While traditional street crimes such as pickpocketing are declining, citizens are more often confronted with sophisticated scams in which they are contacted by seemingly trustworthy individuals, authorities, or superiors and urged to transfer large sums of money – only to later discover that these communications were entirely AI-generated. 

At the GippLab, the project is being carried out by incoming graduate researchers Adam Lehavi and Kia-Jüng Yang, under the supervision of Dr. Terry Lima Ruas and Dr. Jan Philip Wahle. Their work focuses on developing a state-of-the-art multimodal dataset and generation pipeline for AI-manipulated content, designed to keep pace with the rapidly evolving capabilities of generative models. 

The fake news project addresses a growing societal and scientific challenge: existing deepfake detection systems are increasingly strained by the realism, diversity, and scale of AI-generated manipulations. Over a structured, multi-phase research program, the team will analyze gaps in current datasets, collect and preprocess authentic source material, generate single-modal and synchronized multimodal deepfakes, and systematically annotate and standardize the resulting data. The outcome will be a robust benchmark dataset and an automated pipeline ready to handle the challenges of an evolving and changing landscape. 

By combining expertise in natural language processing, computer vision, and generative AI, the project aims to strengthen the scientific foundations of deepfake detection research. Beyond academic impact, the project contributes to broader efforts to combat misinformation, cybercrime, and digital fraud, supporting reliable authenticity systems for media, research, and society at large. 

Der Beitrag GippLab Joins International Research Project to Detect Fake News erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

In the presence of his family, friends, and colleagues, Jan Philip delivered an outstanding presentation of his work, earning him the title of Dr. rer. nat. The entire team extends its heartfelt congratulations on this significant achievement.

If you’d like to learn more about Dr. Wahle’s research on paraphrase generation and language modeling, his Dissertation is available here.

Der Beitrag Congratulations to Jan Philip Wahle on his successful PhD thesis defense erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

The goal of the project is to make museum collections accessible through AI. In this way, researchers will in the future be able to access relevant collection objects more efficiently and gain new insights—using intelligent models that can, for example, identify statues or sculptures with the help of metadata such as excavation texts. The AI solutions are based on the latest technologies of large language models.

“By using AI, we are creating completely new opportunities for the scientific research and presentation of museum collections,” explains Prof. Dr. Bela Gipp, head of the research group “Analysis of Scientific Information Resources” at the University of Göttingen. Jan Philip Wahle, project manager at the chair, adds: “With this project, we are developing advanced AI processes that combine machine learning and language processing to automatically analyze museum data, draw intelligent conclusions, and enable completely new scientific analyses.”

The project “AI in Museums” demonstrates how modern technologies can revolutionize access to our cultural heritage. Through intelligent networking of data, support for provenance research, and the development of practical tools for everyday museum work, collections are not only digitally accessible but also made tangible for research, education, and the public.

Press release by the University: https://www.uni-goettingen.de/de/54088.html?id=7932
Press release by our partners: https://newsroom.jade-hs.de/magazin/wie-ki-die-museen-von-morgen-gestaltet
Press release by Göttinger Tageblatt: https://www.goettinger-tageblatt.de/beruf-und-bildung/regional/wie-ki-und-goettinger-forscher-die-geheimen-schaetze-unserer-museen-enthuellen-UD3DDNHXVZAO7CBUJS72IIVXJM.html

Der Beitrag Lower Saxony Funds Project on AI in Museums with €2.25 Million erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

At the heart of the project are digital lexicographic resources, based on analyses and annotations created by highly specialized experts. With newly developed AI methods, this often highly detailed expert knowledge will now be applied to a much larger collection of historical documents, while enabling new “self-learning” extensions.

The research consortium, led by Prof. Dr. Anja Lobenstein-Reichmann at the Göttingen Academy of Sciences and Humanities, is collaborating closely with the University of Göttingen and the State and University Library of Lower Saxony (SUB) under the leadership of Prof. Dr. Thomas Kaufmann and Prof. Dr. Bela Gipp. The project is funded by the Lower Saxony Ministry of Science and Culture (MWK) with approximately €4.5 million over four years.

“EDIKILEX combines historical lexicography and text editing with cutting-edge AI language models,” explains Prof. Dr. Bela Gipp, Scientific Director at SUB Göttingen and Chair of Scientific Information Analytics. Jan Philip Wahle, project manager in Gipp’s research group, adds: “Our goal is an AI system that makes Early New High German texts much more accessible and comparable for research and teaching.”

The platform under development will make it possible to automatically classify Early New High German texts according to their historical linguistic meaning and usage. Researchers will be able to quickly identify, for example, how a term changed over time, in which contexts it appeared, or which social and historical nuances it carried.

For instance, the word “Abenteuer” (or “adventure” in English) had different meanings in Early New High German: in some texts, spelled “Abentheür” or “Abenteür”, it referred to a knightly trial or tournament, often linked with the idea of a risky undertaking. Elsewhere, as “Abenthewer”, it conveyed illegitimacy or immorality. Until now, recognizing and categorizing such variations has only been possible through painstaking manual work by experts.

“Analyses across large amounts of text were hardly possible in the past,” note Gipp and Wahle. “They would have required hundreds of experts to precisely annotate each word’s meaning.” The Göttingen-based AI is expected to significantly accelerate this work, making historical knowledge accessible that might otherwise have been lost.

Press release from the state library: https://www.sub.uni-goettingen.de/sub-aktuell/einzelansicht/niedersaechsisches-ministerium-fuer-wissenschaft-und-kultur-foerdert-goettinger-projekt-edikilex/

Der Beitrag Lower Saxony Ministry of Science and Culture Funds Göttingen Project “EDIKILEX” with €4.5 Million erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

This project was the most popular task ever hosted on Codabench, with over 700 participants, 3,600 submissions, and nearly 100 system papers. The task spanned 30+ languages across seven language families, spotlighting emotion classification, intensity scoring, and cross-lingual transfer, all with a strong focus on low-resource languages from regions like Africa, Asia, Latin America, and Eastern Europe.  

This effort was through collaboration with researchers and annotators from around the world, including teams from Cardiff University, Imperial College London, MBZUAI, NRC Canada, and many more.  

Paper: BRIGHTER: Bridging the Gap in Human-Annotated Textual Emotion Recognition Datasets for 28 Languages
https://aclanthology.org/2025.acl-long.436/

Project Page: https://brighter-dataset.github.io/

Der Beitrag GippLab wins ACL Best Paper Awards erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.