The POS-tagger moot uses statistical methods for the disambiguation of lexical classes. In addition to the traditional bi-/trigram-based tagging methods the tagger takes into account sets of possible analyses (lexical classes) for each input word. It is thus possible to restrict the tagger’s output to a set of analyses provided by e.g. a rule-based morphological component. This approach has been shown to reduce the error rate by up to 21 % with respect to a traditional HMM.
Tool for creating own tools and resources
Berlin-Brandenburgische Akademie der Wissenschaften, Berlin