Despite the existence of approximately 7000 languages, research in Natural Language Processing has predominantly focused on a select few high-resource languages, which does not serve the global linguistic diversity adequately. Multilingual Information Extraction aims to improve information access and communication across various languages, has therefore emerged as a vital research area. This field entails several key tasks, namely Event Trigger Detection, Event Argument Extraction, Entity Mention Recognition, and Relation Extraction, each contributing to the extraction of structured information from unstructured text. This work explores three primary research directions in Multilingual IE: (1) enhancing Multilingual IE upstream models, (2) developing language-agnostic downstream models, and (3) advancing cross-lingual transfer learning methods for situations with scarce training data. These directions are examined in detail, highlighting the recent advancements, enduring challenges, and future prospects, contributing to the overarching goal of democratizing communication and information access in the linguistic landscape of our world.