Uncertainty is an integral part of the scientific research process and is inherent in the construction of new knowledge. In this article, we examine the way in which uncertainty is expressed in scientific articles, and propose an annotation framework that takes into account the different dimensions of this notion. Scientific uncertainty is defined here as the expression of a lack of knowledge or a lack of precision in the information on an identified subject or concept. We propose a gold standard dataset composed of 1,839 sentences of manually annotated scientific articles from several disciplines. We also propose a linguistic knowledge-based approach for the automatic annotation of articles and for the detection and categorisation of scientific uncertainty. We compare the effectiveness of our approach in terms of Precision, Recall and F1 scores to the few-shot prompting methods performed via the Phi-3.5 and Llama 3 Large Language Models for the same annotation task. This comparative evaluation shows similar scores between the different approaches, with F1 scores up to 0.858 for our approach. - 122