Textual Entailment Technique for the Bahasa Using BiLSTM

Recognizing Textual Entailment (RTE) is an important task in Natural Language Understanding (NLU). RTE aims to identify the entailment relationship between two text fragments (premise and hypothesis). Researchers on textual entailment have converged various languages. Research indicates that the use of deep learning approaches improved the accuracy performance of RTE in English. This study proposes an RTE model in Bahasa using a Bidirectional Long Short-Term Memory (BiLSTM). The method was trained on a large corpus in Bahasa. The corpus was generated by translating the SNLI corpus into Bahasa (SNLI Indo). The experimentation shows that the best training accuracy that the model produced, i.e., 79.37%, is when BiLSTM was combined with pre-trained IndoBERT as the word embedding method.