Robust and Statistical Parsing in GF

In this talk we review the state of turning Grammatical Framework to a hybride combining lingustic knowledge with statistical evidence. Our statistical parser already outperforms the existing state of the art parsers for related formalisms. For one and the same grammar we show a speed-up of up to two orders of magnitude without sacrificing the quality. In addition the grammarians can still use the high-level GF language for developing their grammars. In MOLTO we have promised a scale of hundreds of lemmas while in this experiment we have shown that we can scale to thousands. On top of the new hybride we have built an experimental robust translator which is based entirely on GF.