Syntax-guided controllable sentence simplification

Sentence simplification is to rephrase a sentence into a form that is easier to read and understand while preserving its essential meaning and information. Recently, monolingual neural machine translation methods have emerged as a popular approach for this task. However, these methods often overlook the syntactic tree information of sentences, which can be crucial for effective simplification. To address this issue, we propose a syntax-guided controllable sentence simplification model that leverages graph attention networks to incorporate the syntactic information of dependency trees. Specifically, besides the sentence encoder, we propose a graph encoder that encodes dependency trees to enrich the syntactic information. Within the decoder, we introduce a syntax-augmented cross-attention that aggregates both sentence and syntax information simultaneously to the target side for simplification. We evaluate our proposed model on two benchmark datasets, showcasing that it outperforms state-of-the-art methods by a significant margin. Our proposed model underscores the significance of incorporating syntactic knowledge in sentence simplification.