Improve Semantic Reasoning Ability by Combining Multi-task Learning and Deformation Model

At present, the research on literary text comprehension focuses on surface comprehension, ignoring deep reasoning and long-range dependencies, and it is difficult to deal with phenomena such as metaphors and symbols. This paper combines multi-task learning with the Transformer model to improve semantic reasoning ability. The study uses a multi-task framework to handle tasks such as plot comprehension and metaphor recognition, enhances reasoning ability by sharing knowledge across tasks, and ensures task collaboration with the help of dynamic task weight adjustment and independent output layers. Experiments show that the accuracy and recall rates of the model constructed in this paper for plot comprehension, character motivation analysis, and metaphor and symbol recognition are 89.2%, 88.6%, 85.7%, 84.3%, and 88.4%, 86.7%, respectively, which are better than other control models, and the performance in reasoning time is also better than other control models. It can be seen that this paper not only promotes the research on semantic reasoning in literary reading comprehension, but also provides new concepts and practical solutions for the integration of multi-task learning and Transformer models