Academic Symposium: Short presentations & panel discussion

Author: Xu Huan

Title: Structure-aware Dialogue Paraphrasing for Task-oriented Conversational Data Augmentation

Enriching dialogue datasets by paraphrasing real conversations

Abstract: Paraphrasing is one of the NLP data augmentation techniques. However, existing paraphrasing strategies mainly focus on sentence-level short phrases with limited capacity to solve multiple-sentence tasks, e.g., dialogue. I will introduce a novel method to paraphrase conversations that keeps the exact meaning of each sentence and the same dialogue logic.