LAMPAT: Low-Rank Adaption for Multilingual Paraphrasing Using Adversarial Training

Paraphrases are texts that convey the same meaning while using differentwords or sentence structures. It can be used as an automatic data augmentationtool for many Natural Language Processing tasks, especially when dealing withlow-resource languages, where data shortage is a significant problem. Togenerate a paraphrase in multilingual settings, previous studies have leveragedthe knowledge from the machine translation field, i.e., forming a paraphrasethrough zero-shot machine translation in the same language. Despite goodperformance on human evaluation, those methods still require paralleltranslation datasets, thus making them inapplicable to languages that do nothave parallel corpora. To mitigate that problem, we proposed the firstunsupervised multilingual paraphrasing model, LAMPAT ($\textbf{L}$ow-rank$\textbf{A}$daptation for $\textbf{M}$ultilingual $\textbf{P}$araphrasing using$\textbf{A}$dversarial $\textbf{T}$raining), by which monolingual dataset issufficient enough to generate a human-like and diverse sentence. Throughout theexperiments, we found out that our method not only works well for English butcan generalize on unseen languages as well. Data and code are available athttps://github.com/phkhanhtrinh23/LAMPAT.