VOnDA is a framework for the implementation of reactive dialogue management functionality in dialogue systems for virtual agents. Although domain-independent, VOnDA is tailored towards dialogue systems with a focus on social communication, which implies the need of a long-term memory and high user adaptivity.
VOnDA heavily draws inspiration from the Information State-Update approach, which bears great similarity to the Belief-Desire-Intent approach to artificial agents. The Information State is realized using the special RDF data storage and reasoning component HFC that is designed to handle time-varying data.
Rather than being a complete dialogue framework, VOnDA is a hybrid rule-based/statistical implementation layer that can implement / simulate various approaches, from state-based to probabilistic reasoning.
The following installation instructions are tested on Ubuntu 22.04
Prerequisites for installing VOnDA are:
-
OpenJDK 11, maven build tool, and git
sudo apt install openjdk-11-jdk maven git -
get VOnDA and compile the framework
git clone https://github.com/bkiefer/vonda.git cd vonda mvn install
There is also a visual debugger/GUI for VOnDA.
The vonda-nlp github repository contains an example use of rasa NLP for a more sophisticated NLU. More NL componentes are planned to be added there.
Detailed user documentation can be found in the doc folder of the project. It contains a step-by-step guide how to build your first small conversational agent with VOnDA. the vonda-chatcat project contains the agent described in this guide, if you don't want to create it by hand
This is version 3.0 of VOnDA. If you have used an older version, especially 2.x, please consult the documentation about the breaking changes and the automatic converter, which is provided in this git repository on a special branch.