Front Matter

Abstract

Large Language Models (LLMs) continue to be the mainstay of Biomedical Language Processing, while the scope of BioNLP research continues to expand across foundational tasks, applications, languages and modalities.In 2025, we see increasing efforts to integrate textual features with visual and sequencing data; new approaches to named entity recognition and linking; work in several languages other than English; and applications ranging from drug discovery and gene editing to veterinary and clinical studies.Complex language technology tasks, such as question answering and summarization, as well as data generation and text mining are also strongly represented.Concerns about potential harms and irresponsible use of AI applications are being addressed through growing research into evaluation, debiasing, and understanding of models' behavior.The submissions to the BioNLP 2025 workshop and the Shared Tasks demonstrated once again that the workshop sponsored by the ACL Special Interest Group on Biomedical Natural Language Processing (SI-GBIOMED) is the preferred venue for the groundbreaking research and applications in Biomedical Language Processing, which encompasses biological, clinical and non-professional medical sub-languages, among others.BioNLP remains the flagship and the generalist in biomedical language processing, accepting all noteworthy work independently of the tasks and languages studied.The quality of submissions continues to impress the program committee and the organizers.BioNLP 2025 received 61 submissions, of which eight were accepted for oral presentation and 22 as poster presentations.The selected works span foundational research, biomedical language processing, clinical applications, and generation of new datasets and benchmarks.Four Shared Tasks were collocated with BioNLP 2025: SMAFIRA: annotating the literature for finding methods alternative to animal experiments.ClinIQLink 2025: LLM Lie Detector Test: evaluating the effectiveness of generative models in producing factually accurate information, using a benchmark dataset specifically curated to align with the knowledge level of a General Practitioner (GP) .

Ähnliche Arbeiten