Semantic roles labeling system for Slovak sentences

The proposed paper is focused on the automatic semantic roles labeling problem. First, semantic roles are discussed and a new two-layer set of semantic roles for Slovak language is described. Then the automatic semantic roles labeling system is introduced. It uses newly-designed training and recognition tool based on Viterbi algorithm, where HMM models are trained on a small “example-based” manually-annotated corpus of Slovak sentences. The classification system is able to classify “unseen” tokens using information about word suffices. The pilot evaluation was performed on the testing part of the corpus.