We present deep communicating agents in an encoder-decoder architecture to address the challenges of representing a long document for abstractive summarization. With deep communicating agents, the task of encoding a long text is divided across multiple collaborating agents, each in charge of a subsection of the input text. These encoders are connected to a single decoder, trained end-to-end using reinforcement learning to generate a focused and coherent summary. Empirical results demonstrate that multiple communicating encoders lead to a higher quality summary compared to several strong baselines, including those based on a single encoder or multiple non-communicating encoders.

翻译：我们在一个编码器-解码器结构中展示深度通信代理器,以应对代表一个长文件进行抽象总结的挑战。在深度通信代理器中,一个长文本编码的任务由多个合作代理商分担,每个代理商负责输入文本的一个小节。这些编码器与一个单一的编码器连接,这些编码器是经过培训的端对端,使用强化学习生成一个重点突出和连贯的概要。经验性结果显示,多个通信编码器导致一个质量更高的摘要,而多个强大的基线则不同,包括那些基于单一编码器或多个非连接编码器的基线。