Abstract: Aiming at the problems of low efficiency and insufficient accuracy existing in traditional manual parsing methods for railway passenger transport dispatching commands, this paper proposed a multi-task joint parsing model based on the RoBERTa-RGP-MRC architecture. The model integrated three core tasks: command classification, named entity recognition and relation extraction. With the support of pre-trained language models, semantic modeling mechanisms and multi-task collaborative learning strategies, it significantly improved the parsing accuracy and deep semantic understanding capability of dispatching commands. Experimental results show that the model achieves an accuracy of 94.2% on the command classification task, and F1-scores of 93.9% and 89.7% on named entity recognition and relation extraction tasks respectively, outperforming existing mainstream baseline methods on all indicators. This research provides a feasible technical solution for intelligent parsing and real-time processing of railway dispatching commands, and has high engineering application value and promotion potential.