In this article we propose a novel method for automatically extracting and classifying argument components from raw texts. We introduce a multi-task deep learning framework exploiting weight parameters trained on auxiliary simple tasks, such as Part-Of-Speech tagging or chunking, in order to solve more complex tasks that require a fine-grained understanding of natural language. Interestingly, our results show that the use of advanced deep learning techniques framed in a multi-task setting enables competing with state-of-the-art systems that depend on handcrafted features.