Automatic speech translation based on the semantic structure

This paper describes a system for the semantic-based translation of spoken or written limited-domain utterances. The semantic structure as output of a semantic decoder serves as the interlingua-level. A word chain generator combined with a linguistic post-processor produces the according word chain in the target language. Both the semantic decoder and the word chain generator work with pure stochastic and trainable knowledge bases. The grammatical features of certain words can be easily extracted by the help of both the word chain and the semantic structure.