Google 2013 European Doctoral Fellowship
FRAMPS is a new method for predicting syntactic structures that are hierarchical, but simpler and less deep than the hierarchies produced by standard parsers. In contrast to dependency and constituency parsing, FRAMPS is fast and robust because it relies on sequence prediction. In contrast to standard sequence prediction, FRAMPS produces hierarchical structure, which is needed for many NLP tasks. For predicting hierarchical structure in a purely sequential approach, FRAMPS makes extensive use of global (nonlocal) features, both sentence-internal and sentence-external global features, in a multi-pass framework.