# Unsupervised Paraphrasing without Translation

## TL;DR

This paper introduces a novel unsupervised method for paraphrasing using a residual vector-quantized variational auto-encoder, enabling effective monolingual paraphrasing without translation, outperforming translation-based methods in several tasks.

## Contribution

It presents a new residual vector-quantized variational auto-encoder model for monolingual paraphrasing, eliminating the need for bilingual translation data.

## Key findings

- Monolingual paraphrasing outperforms unsupervised translation in all tested tasks.
- Monolingual methods are effective for paraphrase identification and data augmentation.
- Supervised translation remains superior for paraphrase generation.

## Abstract

Paraphrasing exemplifies the ability to abstract semantic content from surface forms. Recent work on automatic paraphrasing is dominated by methods leveraging Machine Translation (MT) as an intermediate step. This contrasts with humans, who can paraphrase without being bilingual. This work proposes to learn paraphrasing models from an unlabeled monolingual corpus only. To that end, we propose a residual variant of vector-quantized variational auto-encoder. We compare with MT-based approaches on paraphrase identification, generation, and training augmentation. Monolingual paraphrasing outperforms unsupervised translation in all settings. Comparisons with supervised translation are more mixed: monolingual paraphrasing is interesting for identification and augmentation; supervised translation is superior for generation.