Stomp, a POS-tagger with a different view

In this paper a data-driven method for Part-of-Speech tagging not using any n-grams of tags is presented. The method matches the text to be tagged to as long continuous strings from the training data as possible and assigns each match the same tags as the match-ing part of the training data. There is also a back-off method for short matches and the treatment of un-known words differs from the most common ways of handling them. The method is evaluated on Swedish text. The method is slightly less accurate than avail-able state of the art taggers. It is faster than some of these, and uses different information, which makes it a useful addition to an ensemble of taggers, especially for unknown words.