Michael Paul
TDMT's German Generation Module
Abstract:TDMT is a multi-lingual approach for translating spoken language dialogues in order
to allow free communication between people who use different languages. We are using
a transfer-driven machine translation approach which integrates example-based and rule-based translation in a common framework. This report will give an overview of the TDMT
system. Its focus, however, is the German generation module.
The empirical linguistic knowledge provided by the application of the transfer rules to the
morphological analysis of the Japanese input sentence forms the input of the generation
process. Given the transfer result each linguistic constituent will be analyzed both on the
word level (inflection phenomena) and on the sentence level (selection of a topological
field).
On the word level we make use of a classification-based word analysis approach, which is
used to generate well-formed surface words in the generation output. In order to handle the
rather free word order of the German language in an accurate way, we use a topological field
approach, whereby a sentence model consists of several "fields", whose linear composition
specifies the chosen word order within the sentence.