Keynote speakers

Gerold SCHNEIDER, PhD

Titulary Professor of Computational Linguistics and co-coordinator of LiRI's service area "Natural Language Processing"

Department of Computational Linguistics

University of Zurich

Keynote Title

Frontiers of AI – and Applications to Language Learning

Bio

Gerold Schneider is Titulary Professor in Computational Linguistics at the University of Zurich. He studied English and Computational Linguistcs. His research interests include automated content analysis and large language models (LLMs), corpus linguistics, Digital Humanities, learner language, text mining, and how the latest AI technology can help research and society. He has over 180 publications on these topics.

He leads the NLP group in LiRI, a University platform which offers services in text processing for interdisciplinary research. He further conducts content analysis, hate speech detection and LLM bias detection in the URPP Digital Religion(s), all at Zurich University.

Abstract

The talk first discusses frontiers of current Large Language Models (LLMs), then I aim to assess the state of applications to language learning (ICALL). In the first part, we evaluate LLMs, generally and on medical and linguistic applications. I will give brief showcases from some of my recent research projects. For the case of pragmatics, the function of language in society, I show that they reach human performance on the task of everyday pragmatic inferences, and also that the gap between the frontier models like GPT-5 and the open weights models like Qwen 3.8 is closing. Increasingly, the societal dangers of AI are also put in the foreground, ranging from hallucinations, lacking confidence levels to fears of job loss and deskilling as we delegate mental tasks to AI. While hallucination risks are decreasing, they persist with very detailed questions, which is paradoxically exactly when we need them, and which can pose real dangers when medical expertise is needed. The discussion of AI replacing humans and human skills has only gained momentum recently, and remains critical. In its application to AI, the second part of the talk, the crucial question to ask is thus under which considitions AI can lead to upskilling, stimulate and support language learning. It can detect errors as we showcase, but pointing out errors needs to be emdebbed in a communicative and stimulating setting, ideally with engaging topics for society and/or matching users' interests. Importantly, the AI tutor constantly has to challenge rather than delivering answers, as chatbots typically do. Overstating this point a bit, AI should not generate answers, but only question, also in a language learning scenario. As humans are social beings, real people should also be involved in the interaction in some of the exercises. An obvious advantage of an AI tutor is endless patience, and the potential to create artificial immersion. Immersion, alongside intrinsic motivation and stimulation, are key factors according to many studies on learners' progress. I thus argue that the future of AI for language teaching is artificial immersion, and deliberation both in the virtual and real space.

Disclaimer

Site Map