Data collection and labelling play a crucial role in many machine learning tasks, particularly in natural language processing. By asking the user to label the most instructive instances, active learning aims to reduce the quantity of labeled data needed to understand the target idea. This way, the concept is learned with fewer examples. It also involves techniques that enhance the efficiency of machine learning models by selectively querying the most informative data points for labelling. These strategies can be effectively integrated into different NLP tasks to enhance the model’s ability to generalize while reducing annotation costs. This paper provides a survey of natural language processing tasks using active learning. Natural language processing tasks are divided into two categories: classification and structured prediction. Six questions about natural language processing tasks with active learning were examined. The most common strategy was uncertainty, accounting for 21%. The tasks were diverse; however, most research focused on named entities. The language commonly used in the research was English, representing 45%. The most frequently used performance metric was the F1 score, representing 28%. Most of the research had been published in 2020.