Abstract

OVR, a Dutch provider of information concerning national public transport, has recently started an investigation of the usability of Automated Speech Processing (ASP) in their company. In the future, part of their humanly operated information service would have to be automated by using ASP.

This report describes a coding scheme that was used to analyse a sample of nearly 500 dialogues out of a corpus of over 5000 dialogues held at OVR. The emphasis of this coding scheme was on analysis and not on design. The coding scheme is hierarchically made out of phases, moves (a.k.a. dialogue acts), and coding of information.

In the second half of the report we will describe the statistical analysis of the coding scheme applied to the corpus. We also give a classification of the dialogues with respect to the difficulty level of an ASP system that will have to process those dialogues.