Virtual agents / Chatbots
in Netherlands

DEIRA - Dynamic Engaging Intelligent Reporter Agent is a virtual horse race reporter, specifically designed for reporting on horse races. The intention is to provide an alternative for real race reporters. Usage and adaptation The system may be adapted in many ways. The face can easily be replaced, as long as it can be used by Visage Interactive. The voice can be generated by any text-to-speech generator. What events are generated and the corresponding sentences that can be uttered about that event can also easily be adjusted, since they are expressed in rules and grammars respectively. This means adapting the reporter to report on a different type of race should be an easy thing to do. Also, the system is designed in a very modular way, which should make other structural changes fairly straight-forward to make.

DEIRA has a large vocabulary, which is easily extendible to include even more. Emotional display is done by intonation of speech as well as physically in the face. The design was highly influenced by the opinions of potential users (in the case of horse race reporting), who were polled multiple times during development. The system is also easily adjustable for use in other kinds of races.

The provided horse race simulation software is used to provide input on the state of the race. DEIRA analyses this information and eventually sends the resulting utterance including emotional expression to the visual representation software Visage Interactive which in turn shows the agent reporting the race.

Within DEIRA, horse race information is received by the Race Analysis Module, which scans the changing race state for events that are possibly interesting to report. These events are then passed on to the Mental Model which adjusts the global mental state and adds a local mental state to the incoming events. The events are then queued in the Event Queue, making them available for the other modules.

The Text Generation Module takes the highest priority event without text and generates suitable sentences based on the grammar to describe it. The Speech Generation Module is responsible for making this text suitable for the text-to-speech engine, also adding the emotional aspects to the speech.

The Face Animation Module provides the corresponding facial expression information. When an event has obtained all the necessary information from these modules, the Output Module sends it all through to Visage.