ConvLab: Multi-Domain End-to-End Dialog System Platform

Lee, Sung‐Jin; Zhu, Qi; Takanobu, Ryuichi; Zhang, Zheng; Zhang, Yaoqin; Li, Xiang; Li, Jinchao; Peng, Baolin; Li, Xiujun; Huang, Minlie; Gao, Jianfeng

doi:10.18653/v1/p19-3011

Cited by 74 publications

(78 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The metrics used in the simulated evaluation includes task success rate, dialog length, average rewards, etc. Simulated evaluation has been widely applied in the recently proposed dialog system platforms, such as PyDial [60] and ConvLab [61,62]. The main advantage of simulated evaluation is that (1) the system can be evaluated in an end-to-end fashion;…”

Section: Simulated Evaluationmentioning

confidence: 99%

“…Recently, there are many methods trying to untie the domain scheme and model design. Convlab [61] provides additional user dialog act annotation in the MultiWOZ [69] dataset to enable developers to apply NLU models in multi-domain, multi-intent settings. While most DST makes assumption that a slot in a belief state can only be mapped to a single value within a single turn, COMER [31] extends the representation of dialog states with priority operator that considers the user's preference on slot values.…”

Section: Schema Integrationmentioning

confidence: 99%

See 1 more Smart Citation

Recent advances and challenges in task-oriented dialog systems

Zhang

Takanobu

Huang

et al. 2020

Sci. China Technol. Sci.

Self Cite

129

View full text Add to dashboard Cite

Due to the significance and value in human-computer interaction and natural language processing, task-oriented dialog systems are attracting more and more attention in both academic and industrial communities. In this paper, we survey recent advances and challenges in task-oriented dialog systems. We also discuss three critical topics for task-oriented dialog systems: (1) improving data efficiency to facilitate dialog modeling in low-resource settings, (2) modeling multi-turn dynamics for dialog policy learning to achieve better task-completion performance, and (3) integrating domain ontology knowledge into the dialog model. Besides, we review the recent progresses in dialog evaluation and some widely-used corpora. We believe that this survey, though incomplete, can shed a light on future research in task-oriented dialog systems. task-oriented dialog systems, natural language understanding, dialog policy, dialog state tracking, natural language generation

show abstract

Section: Simulated Evaluationmentioning

confidence: 99%

Section: Schema Integrationmentioning

confidence: 99%

Recent advances and challenges in task-oriented dialog systems

Zhang

Takanobu

Huang

et al. 2020

Sci. China Technol. Sci.

Self Cite

129

View full text Add to dashboard Cite

show abstract

“…They frame the dialogue as a context-to-response mapping and use an oracle belief tracker and a discrete database accessing component to augment a sequence-to-sequence model. In 2019, Lee et al [ 29 ] presented ConvLab, an open-source multi-domain dialogue system platform designed for allowing researchers to set up experiments in the field of dialogue management. This platform offers annotated datasets and pre-trained models that can been used in different setups in order to compare approaches.…”

Section: Related Workmentioning

confidence: 99%

Modelling Multimodal Dialogues for Social Robots Using Communicative Acts

Fernández-Rodicio

Castro‐González

Alonso-Martín

et al. 2020

Sensors

View full text Add to dashboard Cite

Social Robots need to communicate in a way that feels natural to humans if they are to effectively bond with the users and provide an engaging interaction. Inline with this natural, effective communication, robots need to perceive and manage multimodal information, both as input and output, and respond accordingly. Consequently, dialogue design is a key factor in creating an engaging multimodal interaction. These dialogues need to be flexible enough to adapt to unforeseen circumstances that arise during the conversation but should also be easy to create, so the development of new applications gets simpler. In this work, we present our approach to dialogue modelling based on basic atomic interaction units called Communicative Acts. They manage basic interactions considering who has the initiative (the robot or the user), and what is his/her intention. The two possible intentions are either ask for information or give information. In addition, because we focus on one-to-one interactions, the initiative can only be taken by the robot or the user. Communicative Acts can be parametrised and combined in a hierarchical manner to fulfil the needs of the robot’s applications, and they have been equipped with built-in functionalities that are in charge of low-level communication tasks. These tasks include communication error handling, turn-taking or user disengagement. This system has been integrated in Mini, a social robot that has been created to assist older adults with cognitive impairment. In a case of use, we demonstrate the operation of our system as well as its performance in real human–robot interactions.

show abstract

“…The absence of user simulators involved will largely reduce the required domain knowledge and supervised learning can lead to robust agent performance. Lee et al (2019) proposes that the action space covers both the atomic action space and the top-k most frequent atomic action combinations in the dataset and then the dialogue PL task can be regarded as a single label classification task. However, the expressive power of the dialogue agent is limited and it is beneficial if the agent can learn the action structure from the data and this could lead to more flexible and powerful system responses.…”

Section: Related Workmentioning

confidence: 99%

Rethinking Supervised Learning and Reinforcement Learning in Task-Oriented Dialogue Systems

Li¹,

Kiseleva²,

Rijke³

2020

Findings of the Association for Computational Linguistics: EMNLP 2020

View full text Add to dashboard Cite

Dialogue policy learning for Task-oriented Dialogue Systems (TDSs) has enjoyed great progress recently mostly through employing Reinforcement Learning (RL) methods. However, these approaches have become very sophisticated. It is time to re-evaluate it. Are we really making progress developing dialogue agents only based on RL? We demonstrate how (1) traditional supervised learning together with (2) a simulator-free adversarial learning method can be used to achieve performance comparable to state-of-the-art (SOTA) RL-based methods. First, we introduce a simple dialogue action decoder to predict the appropriate actions. Then, the traditional multilabel classification solution for dialogue policy learning is extended by adding dense layers to improve the dialogue agent performance. Finally, we employ the Gumbel-Softmax estimator to alternatively train the dialogue agent and the dialogue reward model without using RL. Based on our extensive experimentation, we can conclude the proposed methods can achieve more stable and higher performance with fewer efforts, such as the domain knowledge required to design a user simulator and the intractable parameter tuning in reinforcement learning. Our main goal is not to beat RL with supervised learning, but to demonstrate the value of rethinking the role of RL and supervised learning in optimizing TDSs.

show abstract

ConvLab: Multi-Domain End-to-End Dialog System Platform

Cited by 74 publications

References 25 publications

Recent advances and challenges in task-oriented dialog systems

Recent advances and challenges in task-oriented dialog systems

Modelling Multimodal Dialogues for Social Robots Using Communicative Acts

Rethinking Supervised Learning and Reinforcement Learning in Task-Oriented Dialogue Systems

Contact Info

Product

Resources

About