To be a Knight-errant Novel Master: Knight-errant Style Transfer via Contrastive Learning

Anonymous

Abstract: Knight-errant style writing is a challenging task for novice writers due to the highly condensed terminology and highly literary language culture of the knight-errant works.To tackle this problem, in this paper, we propose a new large-scale parallel knight-errant dataset and model the knight-errant writing as a text style transfer (TST) task between modern style and knight-errant style. We establish the benchmark performance of six current SOTA models for knight-errant style transfer. Empirical results demonstrate that the existing SOTA TST models are unable to accurately identify and generate knight-errant style sentences. Therefore, we propose Knight, a TST framework based on contrastive learning. Knight uses multiple strategies to construct positive and negative samples, making it significantly better than existing SOTA models in terms of content fluency, style transfer accuracy, and factuality.The data and code are publicly available \footnote{https://anonymous.4open.science/r/knight-errant-style-transfer-C2E1/}.

Paper Type: long