ViMRC - VLSP 2021: An empirical study of Vietnamese Machine Reading Comprehension with Unsupervised Context Selector and Adversarial Learning

Machine Reading Comprehension (MRC) is a great NLP task that requires concentration on making the machine read, scan documents, and extract meaning from text, just like human reader.One of MRC system challenges not only having to understand context answer but also being aware trust-worthy given question possible or not.Thought pre-trained language models (PTMs) have shown their performance many downstream tasks, it still has limitation in fixed-length input. We propose an unsupervised selector shortens contains answers within related contexts.In VLSP2021-MRC shared dataset, we empirical several training strategies consisting unanswerable sample selection different adversarial approaches, which slightly boost 2.5% EM score 1% F1 score.