This paper explores active learning strategies for sequence labeling tasks, such as information extraction and document segmentation, where unlabeled data are abundant but annotation is costly. The authors survey existing query selection methods and propose several novel algorithms to address their limitations. They conduct a large-scale empirical comparison using multiple corpora, demonstrating that their proposed methods outperform state-of-the-art techniques. The paper introduces new query strategies, including information density, sequence vote entropy, and Fisher information, and evaluates their performance on various benchmark corpora. The results show that these methods, particularly information density, consistently improve accuracy and efficiency in active learning for sequence labeling tasks.This paper explores active learning strategies for sequence labeling tasks, such as information extraction and document segmentation, where unlabeled data are abundant but annotation is costly. The authors survey existing query selection methods and propose several novel algorithms to address their limitations. They conduct a large-scale empirical comparison using multiple corpora, demonstrating that their proposed methods outperform state-of-the-art techniques. The paper introduces new query strategies, including information density, sequence vote entropy, and Fisher information, and evaluates their performance on various benchmark corpora. The results show that these methods, particularly information density, consistently improve accuracy and efficiency in active learning for sequence labeling tasks.