Automatic Detection of Annotation Errors in Polish-Language Corpora

In this article we propose an extension to the variation ngram based method of detecting annotation errors. We also show an approach to nding anomalies in the morphosyntactic annotation layer by using association rule discovery. As no research has previously been done in the eld of morphosyntactic annotation error correction for Polish, we provide novel results based on experiments on the largest available Polish language corpus, the National Corpus of Polish (NCP). We also discuss the di erences in the approaches used earlier for English language data and the method proposed in this article, taking into account the characteristics of Polish language.