Improving Phrase Structure Parsing for Icelandic

Abstract

Wepresent an improved state-of-the-art in phrase structure parsing for Icelandic, building on the Icelandic Parsed Historical Corpus (IcePaHC), a CLARIN resource, as well as previous milestones presented at the CLARIN Annual Conference in the past. The present parsing system utilizes the Stanford Stanza system as well as IceBERT, a freely available Icelandic BERT Model. Wedescribe previous work, the different configurations used for the present work, as well as the setup that yielded the best outcome, an F1 score of 90.38%.