Toward Multimodal Support of Interpreted Telephone Dialogues

Abstract

Real-time interpretation of telephone dialogues presents a difficult array of long-term empirical and computational research problems. Certainly, we are a long way from fully understanding the unique discourse and performance characteristics that will require accommodation during such dialogues, much less are we prepared to automatically process and interpret them for foreign speakers actively engaged in real tasks. On the other hand, research has intensified towards the development of both spoken language systems and translation of text – two prerequisites for developing the more sophisticated systems needed to handle interpretation of telephone dialogues. Furthermore, the specific long-term goal at organizations like ATR’s Interpreting Telephony Research Laboratories in Japan has been real-time interpretation of spoken dialogues, and an interest in similar research pursuits is currently being generated by the Verbmobil research program in Germany. 

As a parallel development, there is emerging commercial interest among telecommunications companies in providing worldwide telephone interpretation services, in this case, with the aid of skilled human interpreters. For example, AT&T’s “Language Line” in the United States, has recently begun advertising rapid access 24 hours a day to professional telephone interpreters representing more than 140 languages – ranging from common languages like French to the more obscure Hausa. Similar services, although less ambitious with respect to the scope of languages represented, are offered by companies like KDD in Japan. Irrespective of whether our goal is the eventual automation of interpreted, telephone calls, or simply the support of human interpreters during such calls, our ultimate success would be enhanced by a clear understanding of how human telephone interpretation actually is conducted.