Word-Based Dialog State Tracking with Recurrent Neural Networks

Abstract

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.

Multi-domain Dialog State Tracking using Recurrent Neural Networks

“…The recent Dialog State Tracking Challenge (DSTC) shared tasks Henderson et al, 2014a;Henderson et al, 2014b) saw a variety of novel approaches, including robust sets of hand-crafted rules (Wang and Lemon, 2013), conditional random fields (Lee and Eskenazi, 2013;Lee, 2013;Ren et al, 2013), maximum entropy models and web-style ranking (Williams, 2014). Henderson et al (2013;2014d;2014c) proposed a belief tracker based on recurrent neural networks. This approach maps directly from the ASR (automatic speech recognition) output to the belief state update, avoiding the use of complex semantic decoders while still attaining state-of-the-art performance.…”

“…Our starting point is the RNN framework for belief tracking that was introduced by Henderson et al (2014d;2014c). This is a single-hidden-layer recurrent neural network that outputs a distribution over all goal slot-value pairs for each user utterance in a dialog.…”

“…In recent shared tasks, approaches based on these models have shown competitive performance (Henderson et al, 2014d;Henderson et al, 2014c). This approach is particularly well suited to our goal of building open-domain dialog systems, as it does not require handcrafted domain-specific resources for semantic interpretation.…”

The Dialog State Tracking Challenge Series

“…Papers describing DSTC entries have broken new ground in dialog state tracking; the best-performing entries have been based on conditional random fields (Lee and Eskenazi 2013), recurrent neural networks (Henderson, Thomson, and Young 2014), and webstyle ranking (Williams 2014). At present, dialog state trackers are able to reliably exceed the performance of a carefully tuned hand-crafted tracker -for example, in DSTC2, the best trackers achieved approximately 78 percent accuracy versus the baseline's 72 percent.…”

Negotiation of Antibiotic Treatment in Medical Consultations: A Corpus Based Study

“…Previous studies (Lee and Eskenazi, 2013;Williams, 2014;Henderson et al, 2014) found that systems combining the classifier approach and the sequence model approach showed competitive results. In doing so, one can train several different models with different sets of parameters and join their results accordingly (Henderson et al, 2014).…”