ADVISER: A Dialog System Framework for Education & Research

Abstract

In this paper, we present ADVISER 1-an open source dialog system framework for education and research purposes. This system supports multi-domain task-oriented conversations in two languages. It additionally provides a flexible architecture in which modules can be arbitrarily combined or exchanged-allowing for easy switching between rules-based and neural network based implementations. Furthermore, ADVISER offers a transparent, user-friendly framework designed for interdisciplinary collaboration: from a flexible back end, allowing easy integration of new features, to an intuitive graphical user interface supporting nontechnical users.

ADVISER: A Toolkit for Developing Multi-modal, Multi-domain and Socially-engaged Conversational Agents

“…We extend and substantially modify our previous, text-based dialog system toolkit (Ortega et al, 2019) while following the same design choices. This means that our toolkit is meant to optimize the following four criteria: Modularity, Flexibility, Transparency and User-friendliness at different levels.…”

“…Policies To determine the correct system action, we provide three types of policy services: a handcrafted and a reinforcement learning policy for finding entities from a database (Ortega et al, 2019), as well as a handcrafted policy for looking up information through an API call. Both handcrafted policies use a series of rules to help the user find a single entity or, once an entity has been found (or directly provided by the user), find information about that entity.…”

“…However, open-source toolkits and frameworks for developing such systems are rare, especially for developing multi-modal systems comprised of speech, text, and vision. Most of the existing toolkits are designed for developing dialog systems focused only on core dialog components, with or without the option to access external speech processing services (Bohus and Rudnicky, 2009;Baumann and Schlangen, 2012;Lison and Kennington, 2016;Ultes et al, 2017;Ortega et al, 2019;Lee et al, 2019).…”

“…In this paper, we introduce a new version of ADVISER -previously a text-based, multi-domain dialog system toolkit (Ortega et al, 2019) -that supports multi-modal dialogs, including speech, text and vision information processing. This provides a new option for building dialog systems that is open-source and Python-based for easy use and fast prototyping.…”

“…We believe that these features will be key to developing humanlike dialog systems because it is well-known that social signals, such as emotional states and engagement levels, play an important role in human computer interaction (McTear et al, 2016). However in contrast to open-ended dialog systems (Weizenbaum, 1966), our toolkit focuses on task-oriented applications (Bobrow et al, 1977), such as searching for a lecturer at the university (Ortega et al, 2019). The purpose we envision for dialog systems developed using our toolkit is not the same as the objective of a social chatbot such as XiaoIce (Zhou et al, 2018).…”