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

• DOI: 10.1609/aaai.v38i16.29804
• Publication Date: 2024-03-25T11:54:35Z
• Authors: Khoi M. Le; Trinh Pham; Tho Quan; Anh Tuan Luu
• Keywords: Rank (graph theory); Training set

Paraphrases are texts that convey the same meaning while using different words or sentence structures. It can be used as an automatic data augmentation tool for many Natural Language Processing tasks, especially when dealing with low-resource languages, where shortage is a significant problem. To generate paraphrase in multilingual settings, previous studies have leveraged knowledge from machine translation field, i.e., forming through zero-shot language. Despite good performance on human evaluation, those methods still require parallel datasets, thus making them inapplicable to languages do not corpora. mitigate problem, we proposed first unsupervised paraphrasing model, LAMPAT (Low-rank Adaptation Multilingual Paraphrasing Adversarial Training), by which monolingual dataset sufficient enough human-like and diverse sentence. Throughout experiments, found out our method only works well English but generalize unseen well. Data code available at https://github.com/phkhanhtrinh23/LAMPAT.