CityBrowser II: A Multimodal Restaurant Guide in Mandarin

In this paper we present a conversational dialogue system, CityBrowser II, which allows users to inquire about information about restaurants in Mandarin. Developed in the Galaxy infrastructure with a common, language-independent semantic representation, CityBrowser integrates portability and scalability. By inheriting the infrastructure and main language understanding/generation components from its English predecessor, CityBrowser can easily be transformed to a Mandarin language environment. This paper describes our system implementation, focusing on the language- specific modifications to the original English system. We show that our language-independent yet scalable system infrastructure makes multilingualism a promising task.

1. INTRODUCTION

For nearly two decades, our group has been conducting research leading to the development of multilingual spoken dialogue systems that combine multiple human language technologies (HLTs) to enable humans and machines to carry on a mixed-initiative conversation for interactive problem solving and information access [1], [2]. To ensure that these systems can easily be generalized to languages other than English, we have made two design choices. First, we assume that it is possible to extract a common, language-independent semantic representation from the languages of interest. Second, we require that each component in the system be as language transparent as possible to promote portability. As illustrated in Figure 1, the dialogue manager, discourse component, and the meaning representation are designed to be independent of the input or output language. Where language-dependent information is required, we have isolated it in the form of external models, tables, or rules.