Semantic roles modeling using statistical language models

摘要

Automatic analysis of semantic roles can be seen not only as one of the natural language processing steps in the human-machine interfaces, but also as a tool to support linguistics analysis of written or spoken texts, which has many applications in education or in telecommunication services. There does not exist an automatic system for semantic roles labeling for Slovak texts, mainly because of the lack of labeled data. In our previous work, the small corpus SEMIENKO, which consists of sentences with semantic roles annotations, was prepared. Statistical modeling using n-gram models were applied to model relations between semanticaly-significant clause parts (chunks) and semantic roles labels. Using of four main types of chunks representations were researched and tested. The predicate-preposition-POStag-based representation has been identified as the well suitable representation of valence frames. Two different architectures of the automatic semantic roles labeling system for Slovak were designed and tested and obtained results were discussed inside the paper.