Unsupervised Paraphrasing by Simulated Annealing

Abstract

Unsupervised paraphrase generation is a promising and important research topic in natural language processing. We propose UPSA, a novel approach that accomplishes Unsupervised Paraphrasing by Simulated Annealing. We model paraphrase generation as an optimization problem and propose a sophisticated objective function, involving semantic similarity, expression diversity, and language fluency of paraphrases. Then, UPSA searches the sentence space towards this objective by performing a sequence of local editing. Our method is unsupervised and does not require parallel corpora for training, so it could be easily applied to different domains. We evaluate our approach on a variety of benchmark datasets, namely, Quora, Wikianswers, MSCOCO, and Twitter. Extensive results show that UPSA achieves the state-of-the-art performance compared with previous unsupervised methods in terms of both automatic and human evaluations. Further, our approach outperforms most existing domain-adapted supervised models, showing the generalizability of UPSA.

References in corpus (7)

• Neural Discrete Representation Learning
• Unsupervised Learning of Sentence Embeddings using Compositional n-Gram Features
• On Using Monolingual Corpora in Neural Machine Translation
• Neural Paraphrase Generation with Stacked Residual LSTM Networks
• Building a Conversational Agent Overnight with Dialogue Self-Play
• A Deep Generative Framework for Paraphrase Generation
• Get To The Point: Summarization with Pointer-Generator Networks

Cited by in corpus (2)

• Unsupervised Paraphrase Generation using Pre-trained Language Models
• Paraphrase Generation as Unsupervised Machine Translation