Multilingual complex word identification: convolutional neural networks with morphological and linguistic features

Abstract

Complex Word Identification (CWI) is an essential task in helping Lexical Simplification (LS) identify the difficult words that should be simplified. In this paper, we present an approach to CWI based on Convolutional Neural Networks (CNN) trained on pre-trained word embeddings with morphological and linguistic features. Generally, the majority of works on CWI are either feature-engineered or neural network with word embeddings. Both approaches have advantages and limitations, so here we combine both approaches in order to achieve higher performance and still support multilingualism. Our evaluation has shown that our system achieves quite similar performance as the state-ofthe-art system for English, and it outperforms the state-of-the-art systems for both Spanish and German.