INFDok - Dokumentenvolltextserver des Fachbereichs Informatik

Kurzfassung auf Englisch:

Parsing learner data with high accuracy is important for all systems that want to analyze language learner input, such as computer-assisted language learning software. State-of-the-art parsers are typically trained on news text and not on language learner data since this kind of data is often not available in sufficient quantities. Our contribution is three-fold: We provide gold-standard syntactic annotations for sentences from language learners of German, evaluate the performance of state-of-the-art parser pipelines on this corpus and explore whether augmentation of a parser with weighted constraints to avoid common structural errors could lead to improvements.