Dialogue interpreting : data and analysis

Abstract

Up to the development of the VERBMOBIL-prototype mainly two types of speech data within the domain of appointment scheduling are recorded: 1. German-German dialogues serving as test and training data, 2. Wizard-of-Oz-dialogues and some authentic German-English and German-Japanese dialogues with human interpreter. Most of the dialogues of type 2 are recorded by the VERBMOBIL Subproject 13 in Hamburg and are stored in a special database (DiaBoLiC, Dialogue Interpreting: Analysis Base of Linguistic Corpora). In this paper, we describe characteristic features of the data of type 2, existing analysis and the structure of DiaBoLic, and we present new results of data analysis (extension of the list of dialogue acts for the domain TRAVEL PLANNING, analysis of translation strategies) . As dialogues of type 2 are regarded as relevant for the VERBMOBIL application, the data have been analyzed with reference to key-words in the realization of dialogue acts ([2]). A list of key-words for dialogue acts as well as a list of tested scenarios is attached to this paper.