O.V.R. Dialogue management

title: O.V.R. Dialogue management
author: Alexandra Peters
published in: August 2000
appeared as: Master of Science thesis
Delft University of Technology
pages: 132
PDF (2.295 KB)


Many people are using the telephone as a medium to obtain information. They call a specific information service and ask the telephone operator specific questions. The telephone operator will try to answer these questions as good as possible with the knowledge she has regarding the specific knowledge domain. A dialogue emerges between the telephone operator and the client.
Most information providers are having problems handling the increasing number of clients. To still be able to help all these clients the dialogues could be made more efficient. A way to do this is to keep the dialogue as short as possible. This can be achieved by asking only the necessary items for providing the correct information. This way the telephone operator manages the dialogue in a directive manner that makes the dialogue seem very unfriendly. However, other clients want to take the initiative and require human friendly dialogues, though this implies longer and non-directive dialogues. The telephone operator has to find the right balance between efficiency and friendliness. An example of an information-providing organisation that has to deal with this situation is OVR (‘Openbaar Vervoer Reisinformatie’). The Dutch company OVR provides information concerning public transport services in the Netherlands by telephone.
To make the dialogue between the client and the telephone operator more efficient and effective while maintaining a high client appreciation the current dialogue management is analysed. One of the main disadvantages is that there is no underlying model or theory available as foundation of the dialogues. To execute the dialogue management task properly, knowledge of the dialogue model has been extracted.
A corpus-based approach is used to achieve this, and included the following steps:

  • The corpus of 200 transliterated OVR-dialogues is analysed.
  • Transliterated dialogues are not suitable for statistical processing; therefore, the dialogues are coded.
  • An OVR-dialogue model is constructed and validated.
  • The knowledge used by the operator during the dialogue is extracted and modelled.
  • With the acquired knowledge a prototype of a knowledge-based training environment is designed and implemented. With this training environment telephone operators are able to train parts of and the complete dialogue.

    blue line
    University logo