Machine Translation with Czech as a source language - recent experiments
Vladislav Kubon
The talk will concern recent MT experiments being carried at the Charles University in Prague. Although these experiments have at least one thing in common - the source language is Czech, they differ substantially.
The first system, Cesilko, concentrates on the translation from Czech into related languages (Slovak, Polish, Lithuanian, Sorbian) using very simple method exploiting the syntactic similarity of these languages. The other two experiments aim at the translation from Czech into English. Each of these systems uses different method - a Dependency Based Machine Translation project uses a stochastic approach. It tries to exploit the deep syntactic (tectogrammatical) level of description used in the Prague Dependency Treebank (and also in the Prague Czech-English Dependency Treebank). The second Czech-English MT experiment tries to find out whether it is possible to reuse parts of an existing MT system (the Czech-to-Russian MT system RUSLAN) in a system aiming at a different target language.