Takahagi, K., & Shinnou, H. Data Augmentation by Shuffling Phrases in Recognizing Textual Entailment.
|
 |
|
Abstract
|
This study aimed to improve the performance of Recognizing Textual Entailment (RTE) using data augmentation, and demonstrated that augmenting the training data with this method improved the performance of the models. Data augmentation is a technique that aims to improve machine learning performance by increasing the number of training data. One method of data augmentation for Japanese sentences is to shuffle the order of phrases that compose a sentence while preserving the dependency relationships. This method has proven effective in improving the performance of text classification, especially when the training data is limited. In this study, we aimed to improve the performance of Recognizing Textual Entailment (RTE) using this method. RTE is recognized as a crucial technique for advancing natural language processing and is applied across various fields, including question-answering and machine translation. In the experiments, we addressed the RTE task using JSICK, a Japanese dataset, and pre-trained models, BERT and RoBERTa. The experimental results demonstrated that augmenting the training data with this method improved the performance of the models.
|
