Modelling Specialized Knowledge With Conceptual Frames: The TermFrame Approach to a Structured Visual Domain Representation

We describe an emerging knowledge base for karstology developed in line with the frame-based approach with data for three languages, English, Slovene and Croatian. An annotation framework was developed to identify the definition elements, semantic categories, relations and relation definitors in definitions of karst concepts extracted from specialized corpora. A multi- layered annotation was performed for sets of validated English and Slovene definitions. We present the distribution of semantic categories and typical definition frames for the most prominent semantic categories: surface and underground landforms, hydrological forms and geomes, for English and Slovene. The definition frames specify the typical properties of concepts we expect to be described, and in our case they were initialized by domain experts and then verified through corpus data. The structured domain representation resulting from the annotated corpus allows us to compare knowledge structures between languages, generate ideal definitions and experiment with domain visualisations, graphs and maps of geolocations.