Previous research [1, 2] showed how NooJ could automatically annotate English Phrasal Verbs (PV), both continuous and discontinuous, in large corpora. Due to certain restrictions, however, not all discontinuous PV listed in the PV Dictionary were successfully identified in texts. Further research [3] showed how a simplified PV grammar could identify more PV and improve recall, but it created an excessive amount of noise. Some of it could be automatically removed with disambiguation grammars, yet accuracy was still limited to 70–74%. In this article we show how incorporating additional dictionaries and disambiguation grammars – modifying them with unique NooJ functionalities such as +EXCLUDE and +UNAMB – can allow us to remove even more noise and achieve a better overall accuracy of 88%.
