Transformer-Based Architectures versus Large Language Models in Semantic Event Extraction: Evaluating Strengths and Limitations

[RESUBMIT of 3673-4887] Dear Reviewers, thank you for your constructive feedback which helped us to improve our article. Please see below for our detailed replies to your suggestions. In summary, these are the major changes to our revised article: - We have clarified definitions in Section 2 "Problem Statement". - We added the new Section 2.1 "Assumptions" which states our assumptions made to address our problem definitions with our methodologies, including a statement about the required dataset characteristics for training and evaluation. - We have updated Fig. 2 and Fig. 5 to clarify the flow of information in our models. - We extended Sections 4.1 and 4.2 to describe our prompting strategies. - We extended Section 5.1.1 to compare our selected event ontologies to other event ontologies and to argue about our selected threshold for class and property selection. - We added a new experiment in Section "5.7 Consistency Analysis" where we demonstrate the robustness of our LLM-based approach when prompting the same LLMs multiple times with the same event classification/relation extraction prompts. - We expanded the Related Work section to incorporate recent studies published since the initial submission, ensuring our analysis reflects the latest advancements and contextualizes our findings within the current research landscape. Best regards The Authors