Abstract
Natural Language Inference (NLI) is a task within Natural Language Processing(NLP) that holds value for various AI applications. However, there have beenlimited studies on Natural Language Inference in Vietnamese that explore theconcept of joint models. Therefore, we conducted experiments using variouscombinations of contextualized language models (CLM) and neural networks. Weuse CLM to create contextualized work presentations and use Neural Networks forclassification. Furthermore, we have evaluated the strengths and weaknesses ofeach joint model and identified the model failure points in the Vietnamesecontext. The highest F1 score in this experiment, up to 82.78% in the benchmarkdataset (ViNLI). By conducting experiments with various models, the mostconsiderable size of the CLM is XLM-R (355M). That combination has consistentlydemonstrated superior performance compared to fine-tuning strong pre-trainedlanguage models like PhoBERT (+6.58%), mBERT (+19.08%), and XLM-R (+0.94%) interms of F1-score. This article aims to introduce a novel approach or modelthat attains improved performance for Vietnamese NLI. Overall, we find that thejoint approach of CLM and neural networks is simple yet capable of achievinghigh-quality performance, which makes it suitable for applications that requireefficient resource utilization.