An Efficient Annotation for Phrasal Verbs using Dependency Information

In this paper, we present an efficient semiautomatic method for annotating English phrasal verbs on the OntoNotes corpus. Our method first constructs a phrasal verb dictionary based on Wiktionary, then annotates each candidate example on the corpus as an either a phrasal verb usage or a literal one. For efficient annotation, we use the dependency structure of a sentence to filter out highly plausible positive and negative cases, resulting in a drastic reduction of annotation cost. We also show that a naive binary classification achieves better MWE identification performance than rule-based and sequence-labeling methods.