Dial2Desc: End-to-end Dialogue Description Generation

  • 2018-11-01 02:10:50
  • Haojie Pan, Junpei Zhou, Zhou Zhao, Yan Liu, Deng Cai, Min Yang
  • 21


We first propose a new task named Dialogue Description (Dial2Desc). Unlikeother existing dialogue summarization tasks such as meeting summarization, wedo not maintain the natural flow of a conversation but describe an object or anaction of what people are talking about. The Dial2Desc system takes a dialoguetext as input, then outputs a concise description of the object or the actioninvolved in this conversation. After reading this short description, one canquickly extract the main topic of a conversation and build a clear picture inhis mind, without reading or listening to the whole conversation. Based on theexisting dialogue dataset, we build a new dataset, which has more than onehundred thousand dialogue-description pairs. As a step forward, we demonstratethat one can get more accurate and descriptive results using a new neuralattentive model that exploits the interaction between utterances from differentspeakers, compared with other baselines.


Introduction (beta)



Conclusion (beta)