Publication details
Abstracts and citations for selected publications. See the full publication list.
Efficient Methods for Natural Language Processing: A Survey
This work provides a comprehensive survey on methods that improve efficiency in NLP.
Annotation Curricula to Implicitly Train Non-Expert Annotators
In this work, we propose to re-order instances during annotation according to their difficulty. In experiments with simulated and real-world users, we show that such annotation curricula can significantly reduce the annotation time.
Empowering Active Learning to Jointly Optimize System and User Demands
In this work, we operationalize active learning for scenarios where the human oracle cannot be a mere labeler (such as language learning). We devise sampling strategies that aim to select instances that benefit both, model and user, and show that this leads to an improved model training over existing active learning strategies and at the same time, results in exercises that suit a learner well.
Manipulating the Difficulty of C-Tests
In this work, we propose two novel generation strategies for C-Tests, a type of gap filling exercises. In contrast to previous approaches, our strategies do not rely upon a static gap placement and size. This allows teachers to generate C-Tests of multiple difficulties from a single input text. In a user study, we show that both our proposed methods are capable successfully manipulating the C-Test difficulty.