Transformer Hybrid Neural Network Model for Natural Language Reasoning

Inference reasoning is a core task in natural language processing, and textual entailment inference is a particularly challenging task. In order to explore textual entailment inference, this paper proposes an enhanced hybrid neural network model based on the Transformer architecture. The Transformer encodes the premises and hypotheses by introducing positional information, obtaining contextual representations. By adding positional information to the premise and hypothesis texts, the model can better capture the sequential relationships in the sequences. Next, we calculate attention scores between the premises and hypotheses, allowing the model to focus more on important information between them, thereby enhancing the model’s inference ability for their meaning or inferential relationship. We evaluated the model on the Stanford Natural Language Inference (SNLI) dataset [1] and achieved an accuracy of 88.9%, surpassing previous research achievements. This indicates that our model has great potential in the task of textual entailment inference.