Mapping Named Entities from NKJP Corpus to Skladnica Treebank and Polish Wordnet

摘要

In this paper a method of mapping named entities from NKJP corpus, where their annotation is rather coarse, to Skladnica tree-bank, where their annotation is wordnet-based, is discussed. The method is based on the fact that Skladnica is a subcorpus of the one-million-word manually annotated balanced subcorpus of NKJP. The method to find a corresponding node in a parse tree is presented. Next, several heuristics to match the lemma of an NE in Polish Wordnet and to choose the most probable semantic interpretation of ambiguous ones are suggested. The results of the mapping are evaluated.