Language Resources for human-centric NLP: the MYTHOTOPIA geo-tagged dataset

The quest for new explorations in the digital humanities domain brings about the notion of representation and organization of the data at hand that will allow not only for the efficient indexing and retrieval of the content but also for providing meaningful visualizations and new experiences for users. Lately, adding accurate geographical information that is relevant to the content of texts, images or videos has become of paramount importance not only to applications that seek to address the so-called "placing task" but also to tools aimed at visualizing the spatial features of texts. When performed manually, geo-tagging is a costly and time-consuming endeavour. To overcome this obstacle, domain-and language-specific tools for geotagging are required, that is, tools to automatically disambiguate names of localities, identify locations and geotag texts, images, audio, or video. In this regard, there is a need for hand-crafted quality datasets that are applicable for training and evaluating purposes. Our work seeks to address the above issues. We will, therefore, present an infrastructure aimed at providing a multi-faceted experience to visitors of Northern Greece using mythology as a starting point. This infrastructure comprises primarily a multi-lingual and multi-modal dataset (that is a corpus of textual data supplemented with images and videos) that belongs to the humanities domain; the corpus bears various annotations including geo-tagging. The backbone of the infrastructure is a content management system that facilitates advanced indexing, linking, and search functionalities. On top of this infrastructure, a tool creates sightseeing scenarios, i.e., itineraries for users based on their preferences. We will present the corpus itself focusing on the content and the steps taken towards rendering it accessible via the content management system in a way that also facilitates useful visualizations and human-centered applications.