The Multilingual Generation Game: Authoring Fluent Texts in Unfamiliar Languages

Abstract

There are obvious reasons for trying to automate the production of multilingual documents. Among them are the rapidly growing need for such documents, the high cost and low availability of good translators, and the fact that translators often need more time than is available to produce good multilingual versions. These problems are compounded when equivalent versions of a document are needed in not just two or three, but many, languages -- as is often the case in Europe, where there are now eleven official languages in the European Community. This talk presents some recent developments in Multilingual Natural Language Generation (MNLG). These allow the automatic production of high-quality multilingual documents, while avoiding many of the well-known pitfalls of the more familiar alternative of Machine Translation (MT) -- for example, the difficulty of information extraction from a source document and the danger of source-language bias.