Assigning part-of-speech to Dutch tweets

Mehdi Aminian,  Tetske Avontuur,  Zeynep Azar,  Iris Balemans,  Laura Elshof,  Rose Newell,  Nanne van Noord,  Alexandros Ntavelos,  Menno van Zaanen
Tilburg University


In this article we describe the development of a part-of-speech (POS) tagger for Dutch messages from the Twitter microblogging website. Initially we developed a POS tag set ourselves with the intention of building a corresponding tagger from scratch. However, it turned out that the output of Frog, an existing high-quality POS tagger for Dutch, is of such quality that we decided to develop a conversion tool that modifies the output of Frog. The conversion consists of retokenization and adding Twitter-specific tags. Frog annotates Dutch texts with the extensive D-Coi POS tag set, which is used in several corpus annotation projects in the Netherlands. We evaluated the resulting automatic annotation against a manually annotated sub-set of tweets. The annotation of tweets in this sub-set have a high inter-annotator agreement and our extension of Frog shows an accuracy of around 95%. The add-on conversion tool that adds Twitter-specific tags to the output of Frog will be made available to other users.