Word-Based Dialog State Tracking with Recurrent Neural Networks

By Matthew Henderson, Blaise Thomson, Susannah Young. 2014

336 citations recorded by the source corpus.

Recently discriminative methods for tracking the state of a spoken dialog have been shown to outperform traditional generative models. This paper presents a new wordbased tracking method which maps directly from the speech recognition results to the dialog state without using an explicit semantic decoder. The method is based on a recurrent neural network structure which is capable of generalising to unseen dialog state hypotheses, and which requires very little feature engineering. The method is evaluated on the second Dialog State Tracking Challenge (DSTC2) corpus and the results demonstrate consistently high performance across all of the metrics.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Efficient Natural Language Response Suggestion for Smart Reply (2017, background): “In recent work, the components of task-oriented dialog systems have been implemented as neural networks, enabling joint learning of robust models [7, 26, 27].”
Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models (2015, methodology): “…few examples using constraints on the POMDP (Gasic et al. 2013) as well as learning the observed features themselves with neural network models (Henderson, Thomson, and Young 2014), yet such approaches still require either hand-crafted features or large corpora of annotated task-speciﬁc…”
A Network-based End-to-End Trainable Task-oriented Dialogue System (2016, methodology): “to allow ambiguous inputs; and it uses delexicalisation2 and a weight tying strategy (Henderson et al., 2014) to reduce the data required to train the model, but still maintains a high degree of freedom should”
The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems (2015, methodology): “[11], which proposes an RNN structure, initialized with a denoising autoencoder, to tackle the DSTC 3 domain.”
Learning End-to-End Goal-Oriented Dialog (2016, methodology): “However, in spite of recent efforts to learn modules (Henderson et al., 2014b), they still require many hand-crafted features for the state and action space representations, which restrict their usage to narrow domains.”
A Survey of Available Corpora For Building Data-Driven Dialogue Systems: The Journal Version (2015, methodology): “DSTC2 introduces changing user goals in a restaurant booking system, while trying to provide a desired reservation(Henderson et al., 2014b).”
MultiWOZ - A Large-Scale Multi-Domain Wizard-of-Oz Dataset for Task-Oriented Dialogue Modelling (2018, methodology): “This is in contrast to previous approaches (Henderson et al., 2014a) and such WOZ set-up has been successfully validated by Wen et al. (2017) and Asri et al. (2017).”
Continuously Learning Neural Dialogue Management (2016, methodology): “As depicted in Figure 1, the input to the model is the belief state s which encodes the understood user in-tents along with the dialogue history (Henderson et al., 2014b; Mrkˇsi´c et al., 2015), and the output is the master dialogue action a that decides the semantic reply.”
Towards Scalable Multi-domain Conversational Agents: The Schema-Guided Dialogue Dataset (2019, background): “Popular approaches on small-scale datasets estimate the dialogue state as a distribution over all possible slot-values (Henderson, Thomson, and Young 2014; Wen et al. 2017; Ren et al. 2018) or individually score all slot-value combinations (Mrk ˇ si ´ c et al. 2017; Zhong, Xiong, and Socher 2018).”
Key-Value Retrieval Networks for Task-Oriented Dialogue (2017, methodology): “We note that other work on task-oriented dialogue by (Wen et al., 2016b; Henderson et al., 2014a) have reported the slot-tracking accuracy of their systems, which is a similar but perhaps more informative and ﬁne-grained notion of a system’s ability to capture user semantics.”
Dialogue Learning with Human Teaching and Feedback in End-to-End Trainable Task-Oriented Dialogue Systems (2018, methodology): “We evaluate the proposed method on DSTC2 (Henderson et al., 2014a) The movie booking dialogue corpus is generated (Shah et al., 2018) using a ﬁnite state machine based dialogue agent and an agenda based user simulator (Schatzmann et al., 2007) with nat-ural language utterances rewritten by real…”
End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning (2016, methodology): “Many DSTC entries learned from data how to update the state, using methods such as recurrent neural networks (Henderson et al., 2014), but the schema of the state be-”
Sample-efficient Actor-Critic Reinforcement Learning with Supervised Data for Dialogue Management (2017, background): “This includes a spoken language understanding module (Chen et al., 2016; Yang et al., 2017), a dialogue belief state tracker (Henderson et al., 2014; Perez and Liu, 2016; Mrkˇsi´c et al., 2017) to predict user intent and track the dialogue history, a dialogue policy (Young et al., 2013; Gaˇsi´c and…”
Iterative policy learning in end-to-end trainable task-oriented neural dialog models (2017, background): “We prepare the data in our study based the corpus from the second Dialog State Tracking Challenge (DSTC2) [41].”
Hybrid Code Networks: practical and efficient end-to-end dialog control with supervised and reinforcement learning (2017, methodology): “Task6 draws on human-computer dialog data from the second dialog state tracking challenge (DSTC2), where usability subjects (crowd-workers) inter-acted with several variants of a spoken dialog sys-tem (Henderson et al., 2014).”
Towards End-to-End Learning for Dialog State Tracking and Management using Deep Reinforcement Learning (2016, background): “Prior work in end-to-end state tracking (Henderson et al., 2014) learns a sequential classiﬁer that estimates the dialog state based on ASR output without the need of an NLU.”
Transferable Multi-Domain State Generator for Task-Oriented Dialogue Systems (2019, background): “Having a predefined ontology can simplify DST into a classification problem and improve performance (Henderson et al., 2014b; Mrkšić et al., 2017; Zhong et al., 2018).”