Extracting and filtering paraphrases by bridging natural language inference and paraphrasing

TL;DR

This paper introduces a novel method that leverages natural language inference to extract and refine paraphrasing datasets, improving quality and revealing noise in existing datasets using transformer models.

Contribution

It proposes a bidirectional entailment approach to extract paraphrases from NLI datasets and to clean existing paraphrasing datasets, demonstrating high-quality results.

Findings

High-quality paraphrasing datasets extracted

Significant noise detected in existing datasets

Transformer models effective in evaluation

Abstract

Paraphrasing is a useful natural language processing task that can contribute to more diverse generated or translated texts. Natural language inference (NLI) and paraphrasing share some similarities and can benefit from a joint approach. We propose a novel methodology for the extraction of paraphrasing datasets from NLI datasets and cleaning existing paraphrasing datasets. Our approach is based on bidirectional entailment; namely, if two sentences can be mutually entailed, they are paraphrases. We evaluate our approach using several large pretrained transformer language models in the monolingual and cross-lingual setting. The results show high quality of extracted paraphrasing datasets and surprisingly high noise levels in two existing paraphrasing datasets.