Transfer Learning on Natural YES/NO Questions
Heju Yin, Feng Zhou, Xiaoyong Li, Zheng Junqiang, Kai Liu · 2020
Although neural network approaches achieve remarkable success on QA, many of them struggle to answer questions that require retrieving relevant factual information from the given paragraph. Inferring from a given paragraph and answering the question to be true or false is an essential part of natural language understanding. To improve the model's inferring ability on Natural YES/NO Question, we provide a simple and effective method. First, we find the tasks related to main task Natural YES/NO Question to fine-tune the model by multi-task learning, then we fine-tune the model on the main task. Results on dataset BoolQ show this method is competitive with other recently published methods, which means transferring from the related datasets through multi-task learning in first stage can save more beneficial information about main task Natural YES/NO. Further analysis show that this method can not only have benefit in this task, it also can be used to other tasks.