Can AI truly grasp the nuances of events described in text? A new study delves into the role of "triggers"—key phrases that signal event occurrences—in document-level event extraction. Researchers explored whether providing AI models with these triggers, or having the models identify them automatically, enhances their ability to extract comprehensive event information. They tested various AI models, including cutting-edge large language models (LLMs), across different datasets, comparing performance with and without triggers. Surprisingly, the research reveals that while triggers boost performance in some cases, they aren’t always essential. For instance, when documents contain numerous, closely related events, providing triggers, particularly high-quality human-annotated ones, proves beneficial. However, for simpler documents with fewer events, AI models can often infer event information without explicit trigger guidance. Interestingly, even low-quality or randomly generated triggers can sometimes improve the performance of LLMs, hinting that the mere concept of a trigger aids comprehension. This study’s insights could revolutionize how we design and train AI models for event extraction, potentially reducing the reliance on expensive human annotations and paving the way for more efficient and robust event understanding.