運用集成式多通道類神經網路於科技英文寫作評估

Scientific Writing Evaluation Using Ensemble Multi-channel Neural Networks

現在有許多母語非英語人士撰寫的科學論文，幫助作者撰寫科學論文的自動化工具產生了巨大的需求。國際科技英文寫作評估評測任務藉由評估一個論文中的英文句子，是否需要語言編輯為任務目標，幫助開發自然語言處理工具，用以改善科技英文寫作的品質。本研究透過實驗設計比較通道數、模型架構和集成數，提出一個集成式多通道類神經網路架構，在該評測資料集下獲得F1分數63.28，比當時參與評測的系統有更好的效能。

A huge number of scientific papers have been authored by non-native English speakers. There is a large demand for effective computer-based writing tools to help writers composing scientific articles. The Automated Evaluation of Scientific Writing (AESW) shared task seeks to promote the use of NLP tools for improving the quality of scientific writing in English by predicting whether a given sentence needs language editing or not. In this study, we propose an ensemble multi-channel BiLSTM-CNN model based on a series of experiments in comparing the number of channels, network architectures, and ensemble size. Our model achieved an F1 score of 63.28 outperforms participating systems in the AESW 2016 task.