A Context-Driven Merge-Sort Model for Community-Oriented Lexical Simplification

Lexical simplification aims to convert complex words in a sentence into semantic equivalent but simple words. Most existing methods ignore sentence contextual information, which inevitably produces a large number of spurious substitute candidates. To that end, this paper proposes a new context-driven Merge-sort model which leverages contextual information in each step of lexical simplification, and a new merging method to combine ranking results produced by the proposed model. Based on standard datasets, our model outperforms a list of baselines including the state-of-the-art LSBert model, indicating its effectiveness in community-oriented lexical simplification.