Paraphrase Generation with Deep Reinforcement Learning

• DOI: 10.18653/v1/d18-1421
• Publication Date: 2019-06-29T20:01:17Z
• Authors: Zichao Li; Xin Jiang; Lifeng Shang; Hang Li
• Keywords: Paraphrase; Sequence (biology)

Automatic generation of paraphrases from a given sentence is an important yet challenging task in natural language processing (NLP). In this paper, we present deep reinforcement learning approach to paraphrase generation. Specifically, propose new framework for the task, which consists generator and evaluator, both are learned data. The generator, built as sequence-to-sequence model, can produce sentence. constructed matching judge whether two sentences each other. first trained by then further fine-tuned reward evaluator. For methods based on supervised inverse respectively, depending type available training Experimental results datasets demonstrate proposed models (the generators) more accurate outperform state-of-the-art automatic evaluation human evaluation.