Chinese Idiom Paraphrasing

Abstract Idioms are a kind of idiomatic expression in Chinese, most which consist four Chinese characters. Due to the properties non-compositionality and metaphorical meaning, idioms hard be understood by children non-native speakers. This study proposes novel task, denoted as Idiom Paraphrasing (CIP). CIP aims rephrase idiom-containing sentences non-idiomatic ones under premise preserving original sentence’s meaning. Since without more easily handled NLP systems, can used pre-process datasets, thereby facilitating improving performance tasks, e.g., machine translation idiom cloze, embeddings. In this study, we treat task special paraphrase generation task. To circumvent difficulties acquiring annotations, first establish large-scale dataset based on human collaboration, consists 115,529 sentence pairs. addition three sequence-to-sequence methods baselines, further propose infill-based approach text infilling. The results show that proposed method has better than baselines established dataset.