Scalable Neural Dialogue State Tracking

Abstract

A Dialogue State Tracker (DST) is a key component in a dialogue system aiming at estimating the beliefs of possible user goals at each dialogue turn. Most of the current DST trackers make use of recurrent neural networks and are based on complex architectures that manage several aspects of a dialogue, including the user utterance, the system actions, and the slot-value pairs defined in a domain ontology. However, the complexity of such neural architectures incurs into a considerable latency in the dialogue state prediction, which limits the deployments of the models in real-world applications, particularly when task scalability (i.e. amount of slots) is a crucial factor. In this paper, we propose an innovative neural model for dialogue state tracking, named Global encoder and Slot-Attentive decoders (G-SAT), which can predict the dialogue state with a very low latency time, while maintaining high-level performance. We report experiments on three different languages (English, Italian, and German) of the WOZ2.0 dataset, and show that the proposed approach provides competitive advantages over state-of-art DST systems, both in terms of accuracy and in terms of time complexity for predictions, being over 15 times faster than the other systems.

Learn to Focus: Hierarchical Dynamic Copy Network for Dialogue State Tracking

“…DST -In recent years, neural network approaches have defined the state-of-the-art in DST research. Most previous works decomposed this task as a series of classification problems (Zhong et al, 2018;Mrksic et al, 2016;Balaraman and Magnini, 2019;Lee et al, 2019). They took each of the slot-value pairs as input for scoring, and output the value with the highest score as the predicted value for a slot.…”

“…To address these limitations, some recent works discarded the classification framework (Xu and Hu, 2018;Wu et al, 2019;Balaraman and Magnini, 2019;. (Xu and Hu, 2018) is the first model that applies pointer networks (Vinyals et al, 2015) to the single-domain DST problem, which generates both start and end pointers to perform index-based copying. )…”

“…With the development of deep learning, neural network-based methods have been widely applied to DST. Many previous studies (Mrksic et al, 2016;Zhong et al, 2018;Balaraman and Magnini, 2019) decompose DST into a series of classification problems. They first score all possible slot-values pairs and then choose the value with the highest probability as the predicted value for that slot.…”

Recent Neural Methods on Dialogue State Tracking for Task-Oriented Dialogue Systems: A Survey

“…Word-based DST (Henderson et al, 2014c) Closed Closed Fixed Function Multi-domain DST (Mrkšić et al, 2015) Closed Closed Fixed Function FS-NBT (Mrkšić and Vulić, 2018) Closed Closed Fixed Function Scalable Multi-domain DST (Rastogi et al, 2017) Closed Closed Fixed Rules CNN-Delex (Wen et al, 2017) Closed Closed Fixed Rules NBT (Mrkšić et al, 2017a) Closed Closed Fixed Rules StateNet (Ren et al, 2018) Closed Open* Fixed Rules Pointer (Xu and Hu, 2018) Open Closed Fixed Rules GLAD (Zhong et al, 2018) Closed Closed Fixed Rules GCE (Nouri and Hosseini-Asl, 2018) Closed Open Fixed Rules GSAT (Balaraman and Magnini, 2019) Closed Closed Fixed Rules BERT-DST (Chao and Lane, 2019) Open Closed Fixed Rules TRADE Open Open* Dynamic None DS-Picklist (Zhang et al, 2019) Closed Open Fixed None SUMBT Closed Open Fixed Function SST Closed Open* Fixed Function SGD-Baseline Open Open Dynamic Rules DA-DST (Balaraman and Magnini, 2021) Open Open Dynamic Rules SOM-DST (Kim et al, 2020) Open Open Dynamic Function TripPy (Heck et al, 2020) Open Open Dynamic Function MinTL Open Open Dynamic Function Nerual Reading (Gao et al, 2019) Open Open Dynamic Function NARDST (Le et al, 2020) Open Open Dynamic None…”

A Contextual Hierarchical Attention Network with Adaptive Objective for Dialogue State Tracking

“…Recent neural network models are proposed for further improvements (Mrkšić et al, 2015;Hori et al, 2016;Lei et al, 2018;Xu and Hu, 2018;Zhong et al, 2018;Nouri and Hosseini-Asl, 2018;Ren et al, 2019;Balaraman and Magnini, 2019). and use an RNN to encode the slot-related information of each turn, where slots can not attend to relevant information of past turns directly.…”