In this master’s thesis, we analysed the productivity and accuracy of lexical markers introducing different semantic relations. In the first stage of our analysis, we carried out different search queries in a corpus specialized in karstology, in order to gather examples of use of the studied semantic relations. Then, we manually evaluated the obtained results and determined whether a certain marker truly indicates the right semantic relation. Based on our analysis, the markers were deemed efficient, with an average accuracy score of more than 50%. On the other hand, some markers turned out to be more ambiguous than others, with an ability to express different types of relations, which can hinder the development of automatic relation extraction methods. In order to test the applicability of lexical markers in more general texts, we performed a short analysis of their accuracy in a corpus of written standard Slovene, which provided similar results as the domain-specific one. This thesis aims to contribute to the development of different natural language processing (NLP) techniques, especially in terms of (semi)automatic relation extraction from Slovene texts.
|