Differentiable Multi-Agent Actor-Critic for Multi-Step Radiology Report Summarization

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 115 - 126 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Dynamic Schema Graph Fusion Network for Multi-Domain Dialogue State Tracking Yue Feng† Aldo Lipani† Fanghua Ye† Qiang Zhang‡ ∗ Emine Yilmaz† †University College London, London, UK ‡Zhejiang University, Hangzhou, China † {yue.feng.20,aldo.lipani,fanghua.ye.19,emine.yilmaz}@ucl.ac.uk ‡qiang.zhang.cs@zju.edu.cn Abstract Dialogue State Tracking (DST) aims to keep track of users’ intentions during the course of a conversation. In DST, modelling the re- lations among domains and slots is still an under-studied problem. Existing approaches that have considered such relations generally fall short in: (1) fusing prior slot-domain mem- bership relations and dialogue-aware dynamic slot relations explicitly, and (2) generalizing to unseen domains. To address these issues, we propose a novel Dynamic Schema Graph Fusion Network (DSGFNet), which generates a dynamic schema graph to explicitly fuse the prior slot-domain membership relations and dialogue-aware dynamic slot relations. It also uses the schemata to facilitate knowledge trans- fer to new domains. DSGFNet consists of a dialogue utterance encoder, a schema graph encoder, a dialogue-aware schema graph evolv- ing network, and a schema graph enhanced dialogue state decoder. Empirical results on benchmark datasets (i.e., SGD, MultiWOZ2.1, and MultiWOZ2.2), show that DSGFNet out- performs existing methods. 1 Introduction Task-oriented dialogue systems can help users ac- complish different tasks (Huang et al., 2020), such as flight reservation, food ordering, and appoint- ment scheduling. Conventionally, task-oriented dia- logue systems consist of four modules (Zhang et al., 2020c): natural language understanding (NLU), dialogue state tracking (DST), dialogue manager (DM), and natural language generation (NLG). In this paper, we will focus on the DST module. The goal of DST is to extract users’ goals or intentions as dialogue states and keep these states updated over the whole dialogue. In order to track users’ goals, we need to have a predefined domain knowl- edge referred to as a schema, which consists of slot ∗Work in part done while at University College London. Movies: Location: Vacaville Slots: “Location”: City where the theatre is located. “Name”: Name of the movie. Could you look for films showing in Vacaville? I discovered 3 films. What do you think of Dumbo, Hellboy, or Shazam!? Dumbo is lovely. Could I assist you with something else? User System State Service: “Movies”: Search for movies by location, genre or other attributes. Schemata I'd also like to look for a diner there. I am searching for one that is intermediate priced. Japanese Restaurant is a lovely diner around there. That's prefect! Thanks! It's my pleasure. Slots: “City”: City in which the restaurant is located. “Name”: Name of the restaurant. “Price_Range”: Price range for the restaurant. Service: “Restaurants”: A leading provider for restaurant search and reservations. Movies: Location: Vacaville Name: Dumbo Movies: Location: Vacaville; Name: Dumbo Restaurants: City: Vacaville; Price_Range: intermediate Movies: Location: Vacaville; Name: Dumbo Restaurants: City: Vacaville; Price_Range: intermediate; Name: Japanese Restaurant co-reference co-update co-occurrence co-occurrence Figure 1: An example of DST. Given the schemata for …

Content truncated to 500 words