Comprehensive Study: How the Context Information of Different Granularity Affects Dialogue State Tracking?

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 2481–2491 August 1–6, 2021. ©2021 Association for Computational Linguistics 2481 Comprehensive Study: How the Context Information of Different Granularity Affects Dialogue State Tracking? Puhai Yang and Heyan Huang∗and Xian-Ling Mao School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China Beijing Engineering Research Center of High Volume Language Information Processing and Cloud Computing Applications, Beijing, China Southeast Academy of Information Technology, Beijing Institute of Technology, Fujian, China {phyang, hhy63, maoxl}@bit.edu.cn Abstract Dialogue state tracking (DST) plays a key role in task-oriented dialogue systems to monitor the user’s goal. In general, there are two strate- gies to track a dialogue state: predicting it from scratch and updating it from previous state. The scratch-based strategy obtains each slot value by inquiring all the dialogue his- tory, and the previous-based strategy relies on the current turn dialogue to update the previ- ous dialogue state. However, it is hard for the scratch-based strategy to correctly track short- dependency dialogue state because of noise; meanwhile, the previous-based strategy is not very useful for long-dependency dialogue state tracking. Obviously, it plays different roles for the context information of different granular- ity to track different kinds of dialogue states. Thus, in this paper, we will study and discuss how the context information of different granu- larity affects dialogue state tracking. First, we explore how greatly different granularities af- fect dialogue state tracking. Then, we further discuss how to combine multiple granularities for dialogue state tracking. Finally, we apply the findings about context granularity to few- shot learning scenario. Besides, we have pub- licly released all codes. 1 Introduction Currently, task-oriented dialogue systems have at- tracted great attention in academia and industry (Chen et al., 2017), which aim to assist the user to complete certain tasks, such as buying prod- ucts, booking a restaurant, etc. As a key compo- nent of task-oriented dialogue system, dialogue state tracking plays a important role in understand- ing the natural language given by the user and expressing it as a certain dialogue state (Rastogi et al., 2017, 2018; Goel et al., 2018). The dialogue ∗Corresponding author U: i am looking for a swimming pool in the south part of town . S: i am sorry , there are no swimming pools .. . would you be interested in visiting a different attraction ? U: how about a cinema instead ? S: cineworld cinema in the south side is located at cambridge leisure park . U: what s the postcode and enterance fee ? S: ... can i help you with anything else ? U: i am also looking for a restaurant that serve portuguese food in the south side of town . S: nandos meets your criteria . would you like to book a table ? U: yes . book for 7 people on saturday at 15:15 . S: it will be reserved for 15 minutes . …

Content truncated to 500 words