vnNLI - VLSP2021: An Empirical Study on Vietnamese-English Natural Language Inference Based on Pretrained Language Models with Data Augmentation

In this paper, we describe an empirical study of data augmentation techniques with various pre-trained language models on the bilingual dataset which was presented at VLSP 2021 - Vietnamese and English-Vietnamese Textual Entailment. We apply machine translation tool to generate new training set from original then investigate compare effectiveness a monolingual multilingual model set. Our experimental results show that fine-tuning XLM-R augmented gives best performance. system ranked third in shared-task F1-score about 0.88.