Writing Support System that Enriches Expressions Using Encoder-Decoder Model

Abstract

Improving texts by using rhetorical expressions is important when we write literary works such as novels and essays. Although expression dictionaries can help authors broaden the range of their expressions, these dictionaries are often exhaustive, and selecting appropriate expressions from them requires a certain amount of writing experience. In this study, we propose a writing support system that help authors improve their wordings by generating and suggesting improved sentences with rhetorical expressions. This system is based on the Encoder-Decoder model that generates sentences in a context-sensitive manner. As a result, we observe that our system generates different modifiers for a same word depending on each context. Through a manual evaluation by a questionnaire, our system was shown to be comprehensively useful for writing support. Furthermore, we showed that attention scores calculated in our model can be used to understand the relationship between added expressions and words in input.