When word order and part-of-speech tags are not enough — Swedish dependency parsing with rich linguistic features

Even with high overall parsing accuracy, datadriven parsers often make errors in the assignment of core grammatical functions such as subject and object. Starting from a detailed error analysis of a state-of-the-art dependency parser for Swedish, we show that the addition of linguistically motivated features targeting specific error types may lead to substantial improvements, both for specific grammatical functions and in terms of overall parsing accuracy. In this way, we achieve the best reported results for dependency parsing of Swedish.