Complex Word Identification: Convolutional Neural Network vs. Feature Engineering

摘要: We describe the systems of NLP-CIC team that participated in Complex Word Identification (CWI) 2018 shared task. The task aimed to benchmark approaches for identifying complex words English and other languages from perspective non-native speakers. Our goal is compare two approaches: feature engineering a deep neural network. Both achieved comparable performance on test set. demonstrated flexibility deep-learning approach by using same network setup Spanish track. competitive results: all our were within 0.01 system with best macro-F1 score sets except Wikipedia set, which 0.04 below score.